Add files using upload-large-folder tool

.launchpadlib/api.launchpad.net/cache/api.launchpad.net,devel,~deadsnakes,+archive,ubuntu,ppa,ws.op=getSigningKeyData-application,json,c76e9ed0b661c7fa5da42e8fb2da319e

@@ -0,0 +1,22 @@
+status: 200
+date: Mon, 05 Feb 2024 23:25:35 GMT
+server: gunicorn
+x-powered-by: Zope (www.zope.org), Python (www.python.org)
+content-security-policy: frame-ancestors 'self';
+content-type: application/json
+strict-transport-security: max-age=15552000
+vary: Accept,Accept-Encoding
+x-content-type-options: nosniff
+x-frame-options: SAMEORIGIN
+x-launchpad-revision: 9643586c585856148a18782148972ae9c1179d06
+x-lazr-notifications: []
+x-xss-protection: 1; mode=block
+x-vcs-revision: 9643586c585856148a18782148972ae9c1179d06
+x-request-id: 452e0c68-aa99-4bb4-abc3-237c7bb39fae
+content-length: 1641
+-content-encoding: gzip
+content-location: https://api.launchpad.net/devel/~deadsnakes/+archive/ubuntu/ppa?ws.op=getSigningKeyData
+-varied-accept: application/json
+-varied-accept-encoding: gzip, deflate
+
+"-----BEGIN PGP PUBLIC KEY BLOCK-----\n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\nyBDUU6BS7Su5nhQMStfdfr4FffLmnvFC9w==\n=7hFk\n-----END PGP PUBLIC KEY BLOCK-----\n"

.local/share/jupyter/nbextensions/snippets_menu/snippets_submenus_python/sympy.js

@@ -0,0 +1,750 @@
+define([
+    "require",
+    "./sympy_functions",
+    "./sympy_assumptions",
+], function (requirejs, sympy_functions, sympy_assumptions) {
+    return {
+        'name' : 'SymPy',
+        'sub-menu' : [
+            {
+                'name' : 'Setup',
+                'snippet'  : [
+                    'from __future__ import print_function, division',
+                    'from sympy import *',
+                    'a, s, t, u, v, w, x, y, z = symbols("a, s, t, u, v, w, x, y, z")',
+                    'k, m, n = symbols("k, m, n", integer=True)',
+                    'f, g, h = symbols("f, g, h", cls=Function)',
+                    'init_printing()',
+                ],
+            },
+            {
+                'name' : 'Documentation',
+                'external-link' : 'http://docs.sympy.org/latest/index.html',
+            },
+            '---',
+            {
+                'name' : 'Constants',
+                'sub-menu' : [
+                    {
+                        'name' : '1',
+                        'snippet' : ['S(1)',], //'S.One',],
+                    },
+                    // {
+                    //     'name' : '0',
+                    //     'snippet' : ['S.Zero',],
+                    // },
+                    // {
+                    //     'name' : '-1',
+                    //     'snippet' : ['S.NegativeOne',],
+                    // },
+                    {
+                        'name' : '1/2',
+                        'snippet' : ['S(1)/2',], //'S.Half',],
+                    },
+                    {
+                        'name' : 'Rational numbers',
+                        'snippet' : ['Rational(3, 7)',],
+                    },
+                    '---',
+                    {
+                        'name' : 'Base of natural logarithm, \\(e\\)',
+                        'snippet' : ['E',],
+                    },
+                    {
+                        'name' : 'Unit imaginary number, \\(i\\)',
+                        'snippet' : ['I',],
+                    },
+                    {
+                        'name' : 'Geometric constant, \\(\\pi\\)',
+                        'snippet' : ['pi',],
+                    },
+                    {
+                        'name' : 'Golden ratio, \\(\\phi\\)',
+                        'snippet' : ['GoldenRatio',],
+                    },
+                    {
+                        'name' : 'Euler-Mascheroni constant, \\(\\gamma\\)',
+                        'snippet' : ['EulerGamma',],
+                    },
+                    {
+                        'name' : 'Catalan\'s constant, \\(K\\)',
+                        'snippet' : ['Catalan',],
+                    },
+                    '---',
+                    {
+                        'name' : 'Infinity, \\(\\infty\\)',
+                        'snippet' : ['oo',], // 'S.Infinity'
+                    },
+                    // {
+                    //     'name' : 'Negative infinity, \\(-\\infty\\)',
+                    //     'snippet' : ['S.NegativeInfinity',],
+                    // },
+                    {
+                        'name' : 'Complex infinity, \\(\\tilde{\\infty}\\)',
+                        'snippet' : ['zoo'], //'S.ComplexInfinity',],
+                    },
+                    {
+                        'name' : 'NaN',
+                        'snippet' : ['nan',], // 'S.NaN'
+                    },
+                ],
+            },
+            sympy_functions,
+            {
+                'name' : 'Calculus',
+                'sub-menu' : [
+                    {
+                        'name' : 'Differentiate once',
+                        'snippet' : [
+                            'expr = exp(x**2)',
+                            'deriv = diff(expr, x)',
+                        ],
+                    },
+                    {
+                        'name' : 'Differentiate multiple times',
+                        'snippet' : [
+                            'expr = x**4',
+                            'deriv = diff(expr, x, 3)',
+                        ],
+                    },
+                    {
+                        'name' : 'Mixed partial derivatives',
+                        'snippet' : [
+                            'expr = exp(x*y*z)',
+                            'deriv = diff(expr, x, y, 2, z, 4)',
+                        ],
+                    },
+                    {
+                        'name' : 'Finite differences',
+                        'snippet' : [
+                            'dx0, dx1 = symbols("dx0, dx1")',
+                            'formula = as_finite_diff(f(x).diff(x), [x-dx0, x, x+dx1])',
+                        ],
+                    },
+                    '---',
+                    {
+                        'name' : 'Indefinite integral',
+                        'snippet' : [
+                            'integral = integrate(cos(x), x)',
+                        ],
+                    },
+                    {
+                        'name' : 'Definite integral',
+                        'snippet' : [
+                            'integral = integrate(exp(-x), (x, 0, oo))',
+                        ],
+                    },
+                    {
+                        'name' : 'Double integral',
+                        'snippet' : [
+                            'integral = integrate(exp(-x**2-y**2), (x, -oo, oo), (y, -oo, oo))',
+                        ],
+                    },
+                    '---',
+                    {
+                        'name' : 'Limits',
+                        'snippet' : [
+                            'lim = limit(sin(x)/x, x, 0, "+")',
+                        ],
+                    },
+                    {
+                        'name' : 'Series expansion',
+                        'snippet' : [
+                            'expr = exp(sin(x))',
+                            'ser = series(expr, x, 0, 6)',
+                        ],
+                    },
+                    {
+                        'name' : 'Series expansion, removing order term',
+                        'snippet' : [
+                            'expr = exp(sin(x))',
+                            'ser = series(expr, x, 0, 6).removeO()',
+                        ],
+                    },
+                    {
+                        'name' : 'Summations',
+                        'snippet' : [
+                            'ell_min,ell,ell_max = symbols("ell_min,ell,ell_max", integer=True)',
+                            'summ = summation((2*ell + 1), (ell, ell_min, ell_max))',
+                        ],
+                    },
+                ],
+            },
+
+            {
+                'name' : 'Solvers',
+                'sub-menu' : [
+                    {
+                        'name' : 'Solve for one variable',
+                        'snippet' : [
+                            'expr = x**4 - 4*x**3 + 2*x**2 - x',
+                            'eqn = Eq(expr, 0)',
+                            'soln = solve(eqn, x)',
+                        ],
+                    },
+                    {
+                        'name' : 'Solve for two variables',
+                        'snippet' : [
+                            'eqns = Eq(x + y, 4), Eq(x*y, 3)',
+                            'soln = solve(eqns, [x,y])',
+                        ],
+                    },
+                    {
+                        'name' : 'Solve differential equation',
+                        'snippet' : [
+                            'expr = f(x).diff(x, x) + 9*f(x)',
+                            "eqn = Eq(expr, 1)  # f''(x) + 9f(x) = 1",
+                            'soln = dsolve(eqn, f(x))',
+                        ],
+                    },
+                ],
+            },
+            {
+                'name' : 'Manipulating expressions',
+                'sub-menu' : [
+                    {
+                        'name' : 'Simplify',
+                        'snippet' : [
+                            'expr = (x**3 + x**2 - x - 1)/(x**2 + 2*x + 1)',
+                            'expr = simplify(expr)',
+                        ],
+                    },
+                    {
+                        'name' : 'Refine, using assumptions',
+                        // 'snippet' : [
+                        //     'expr = exp(pi*I*2*x)',
+                        //     'assumption = Q.integer(x) & Q.integer(y)',
+                        //     'expr = refine(expr, assumption)',
+                        // ],
+                        'sub-menu' : [
+                            {
+                                'name' : 'Refine',
+                                'snippet' : [
+                                    'expr = exp(pi*I*2*(x+y))',
+                                    'assumption = Q.integer(x) & Q.integer(y)',
+                                    'expr = refine(expr, assumption)',
+                                ],
+                            },
+                            {
+                                'name' : 'Refine in context manager',
+                                'snippet' : [
+                                    'expr = exp(pi*I*2*(x+y))',
+                                    'with assuming(Q.integer(x) & Q.integer(y)):',
+                                    '    expr = refine(expr)',
+                                ],
+                            },
+                            sympy_assumptions,
+                        ],
+                    },
+                    {
+                        'name' : 'Expansion',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Expand basic expressions',
+                                'snippet' : [
+                                    'expr = (x + 2)*(x - 3)',
+                                    'expr = expand(expr)',
+                                ],
+                            },
+                            '---',
+                            {
+                                'name' : 'Expand, including complex parts',
+                                'snippet' : [
+                                    'expr = cos(x)',
+                                    'expr = expand(expr, complex=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand, including functions',
+                                'snippet' : [
+                                    'expr = gamma(x+3)',
+                                    'expr = expand(expr, func=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand, including trig',
+                                'snippet' : [
+                                    'expr = sin(x+y)*(x+y)',
+                                    'expr = expand(expr, trig=True)',
+                                ],
+                            },
+                            '---',
+                            {
+                                'name' : 'Expand only real and imaginary parts',
+                                'snippet' : [
+                                    'expand_complex(x)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only functions',
+                                'snippet' : [
+                                    'expr = gamma(x + 2)',
+                                    'expr = expand_func(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only hypergeometric functions',
+                                'snippet' : [
+                                    'expr = hyper([1,1], [1,], z) + gamma(z)',
+                                    'expr = hyperexpand(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only logarithms',
+                                'snippet' : [
+                                    'a, b = symbols("a, b", positive=True)',
+                                    'expr = log(a**2*b)',
+                                    'expr = expand_log(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only multiplication over addition',
+                                'snippet' : [
+                                    'expr = y*(x + z)',
+                                    'expr = expand_mul(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only multinomials',
+                                'snippet' : [
+                                    'expr = (x + y + z)**3',
+                                    'expr = expand_multinomial(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only powers of multiplied bases',
+                                'snippet' : [
+                                    'a, b = symbols("a, b", positive=True)',
+                                    'expr = (a*b)**z',
+                                    'expr = expand_power_base(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only addition in exponents',
+                                'snippet' : [
+                                    'expr = x**(y + 2)',
+                                    'expr = expand_power_exp(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand only trig',
+                                'snippet' : [
+                                    'expr = sin(x+y)*(x+y)',
+                                    'expr = expand_trig(expr)',
+                                ],
+                            },
+                        ],
+                    },
+                    {
+                        'name' : 'Collect terms',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Collect as coefficients of one factor',
+                                'snippet' : [
+                                    'expr = y*x**2 + z*x**2 + t*x - 2*x + 3',
+                                    'expr = collect(expr, x)',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect as coefficients of multiple factors',
+                                'snippet' : [
+                                    'expr = x**2 + y*x**2 + x*y + y + z*y',
+                                    'expr = collect(expr, [x, y])',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect with respect to wild card',
+                                'snippet' : [
+                                    'w = Wild("w")',
+                                    'expr = z*x**y - t*z**y',
+                                    'expr = collect(expr, w**y)',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect and apply function to each coefficient',
+                                'snippet' : [
+                                    'expr = expand((x + y + 1)**3)',
+                                    'expr = collect(expr, x, factor)',
+                                ],
+                            },
+                            {
+                                'name' : 'Recursively collect',
+                                'snippet' : [
+                                    'expr = (x**2*y + x*y + x + y)/(x*y + z*y)',
+                                    'expr = rcollect(expr, y)',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect constants',
+                                'snippet' : [
+                                    'expr = sqrt(3)*x + sqrt(7)*x + sqrt(3) + sqrt(7)',
+                                    'expr = collect_const(expr)',
+                                ],
+                            },
+                        ],
+                    },
+                    {
+                        'name' : 'Substitutions and replacements',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Substitute one subexpression for another',
+                                'snippet' : [
+                                    'expr = 1 + x*y',
+                                    'expr = expr.subs(x, pi)',
+                                ],
+                            },
+                            {
+                                'name' : 'Substitute multiple subexpressions successively',
+                                'snippet' : [
+                                    'expr = (x+y)/y',
+                                    'substitutions = [(x+y, y), (y, x+y)]',
+                                    'expr = expr.subs(substitutions)',
+                                ],
+                            },
+                            {
+                                'name' : 'Substitute multiple subexpressions simultaneously',
+                                'snippet' : [
+                                    'expr = (x+y)/y',
+                                    'substitutions = [(x+y, y), (y, x+y)]',
+                                    'expr = expr.subs(substitutions, simultaneous=True)',
+                                ],
+                            },
+                            '---',
+                            {
+                                'name' : 'Replace one object with another',
+                                'snippet' : [
+                                    'expr = 1 + x*y',
+                                    'expr = expr.replace(x, pi)',
+                                ],
+                            },
+                            {
+                                'name' : 'Replace one object with some function of its arguments',
+                                'snippet' : [
+                                    'expr = log(sin(x)) + tan(sin(x**2))',
+                                    'expr = expr.replace(sin, lambda arg: sin(2*arg))',
+                                ],
+                            },
+                            {
+                                'name' : 'Replace a pattern with an object',
+                                'snippet' : [
+                                    '# Note: `exclude=` specifies that the Wild cannot match any item in the list',
+                                    'a, b = symbols("a, b", cls=Wild, exclude=[x,y])',
+                                    'expr = 2*x + y + z',
+                                    'wild = a*x + b',
+                                    'replacement = b - a',
+                                    '# Note: `exact=True` demands that all Wilds have nonzero matches',
+                                    'expr = expr.replace(wild, replacement, exact=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Replace a pattern with some function of that object',
+                                'snippet' : [
+                                    'a = symbols("a", cls=Wild, exclude=[])',
+                                    'expr = log(sin(x)) + tan(sin(x**2))',
+                                    'expr.replace(sin(a), lambda a: sin(2*a))',
+                                ],
+                            },
+                            {
+                                'name' : 'Replace anything with some function of that thing',
+                                'snippet' : [
+                                    'g = 2*sin(x**3)',
+                                    'g.replace(lambda expr: expr.is_Function, lambda expr: expr**2)',
+                                ],
+                            },
+                            '---',
+                            {
+                                'name' : 'Replace exact subexpressions',
+                                'snippet' : [
+                                    'expr = x**2 + x**4',
+                                    'replacements = {x**2: y}',
+                                    'expr = expr.xreplace(replacements)',
+                                ],
+                            },
+                            // {
+                            //     'name' : 'rewrite',
+                            //     'snippet' : [
+                            //         'expr = tan(x)',
+                            //         'expr = expr.rewrite(sin)',
+                            //     ],
+                            // },
+                        ],
+                    },
+                    {
+                        'name' : 'Evaluation',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Evaluate numerically to arbitrary precision',
+                                'snippet' : [
+                                    'expr = x * sqrt(8)',
+                                    'precision = 50',
+                                    'val = N(expr, precision, subs={x:2.4})',
+                                ],
+                            },
+                            {
+                                'name' : 'Evaluate numerically to python float',
+                                'snippet' : [
+                                    'expr = x * sqrt(8)',
+                                    'val = float(expr.subs([(x, 2.4)]))',
+                                ],
+                            },
+                            {
+                                'name' : 'Create numpy function for efficient evaluation',
+                                'snippet' : [
+                                    'import numpy',
+                                    'a = numpy.arange(10)',
+                                    'expr = sin(x)',
+                                    'f = lambdify(x, expr, "numpy")',
+                                    'vals = f(a)',
+                                ],
+                            },
+                        ],
+                    },
+                    '---',
+                    {
+                        'name' : 'Polynomials',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Factor polynomial over rationals',
+                                'snippet' : [
+                                    'expr = x**3 - x**2 + x - 1',
+                                    'expr = factor(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect common powers of a term',
+                                'snippet' : [
+                                    'expr = x*y + x - 3 + 2*x**2 - z*x**2 + x**3',
+                                    'expr = collect(expr, x)',
+                                ],
+                            },
+                            {
+                                'name' : 'Extract coefficient of a term',
+                                'snippet' : [
+                                    'expr = 3+2*x+4*x**2',
+                                    'expr = expr.coeff(x**2)',
+                                ],
+                            },
+                        ],
+                    },
+                    {
+                        'name' : 'Rational functions',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Cancel',
+                                'snippet' : [
+                                    'expr = (x**2 + 2*x + 1)/(x**2 + x)',
+                                    'expr = cancel(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Decompose into partial fractions',
+                                'snippet' : [
+                                    'expr = (4*x**3 + 21*x**2 + 10*x + 12)/(x**4 + 5*x**3 + 5*x**2 + 4*x)',
+                                    'expr = apart(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Join over common denominator',
+                                'snippet' : [
+                                    'expr = 1/x + 1/y',
+                                    'expr = ratsimp(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Remove square roots from denominator',
+                                'snippet' : [
+                                    'expr = 1/(1+I)',
+                                    'expr = radsimp(expr)',
+                                ],
+                            },
+                        ],
+                    },
+                    {
+                        'name' : 'Powers',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Important caveats',
+                                'external-link' : 'http://docs.sympy.org/dev/tutorial/simplification.html#powers'
+                            },
+                            '---',
+                            // {
+                            //     'name' : 'Setup for these snippets',
+                            //     'snippet' : [
+                            //         'x, y = symbols("x, y", positive=True)',
+                            //         'a, b = symbols("a, b", real=True)',
+                            //         'z, t, c = symbols("z, t, c")',
+                            //     ],
+                            // },
+                            {
+                                'name' : 'Simplify powers for general arguments',
+                                'snippet' : [
+                                    'powsimp(x**y * x**z)',
+                                ],
+                            },
+                            {
+                                'name' : 'Simplify powers, forcing assumptions',
+                                'snippet' : [
+                                    'powsimp(x**y * x**z, force=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand powers by exponent for general arguments',
+                                'snippet' : [
+                                    'expand_power_exp(x**(y + z))',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand powers of multiplied bases, forcing assumptions',
+                                'snippet' : [
+                                    'expand_power_base((x*y)**z, force=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect exponents on powers for general arguments',
+                                'snippet' : [
+                                    'powdenest((x**y)**z)',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect exponents on powers, forcing assumptions',
+                                'snippet' : [
+                                    'powdenest((x**y)**z, force=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Collect exponents on powers, forcing assumptions and polar simplifications',
+                                'snippet' : [
+                                    'powdenest((z**a)**b, force=True, polar=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Denest square-roots',
+                                'snippet' : [
+                                    'sqrtdenest(sqrt(5 + 2*sqrt(6)))',
+                                ],
+                            },
+                        ],
+                    },
+                    {
+                        'name' : 'Exponentials and Logarithms',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Important caveats',
+                                'external-link' : 'http://docs.sympy.org/dev/tutorial/simplification.html#exponentials-and-logarithms'
+                            },
+                            '---',
+                            // {
+                            //     'name' : 'Setup for these snippets',
+                            //     'snippet' : [
+                            //         'x, y = symbols("x, y", positive=True)',
+                            //         'n = symbols("n", real=True)',
+                            //     ],
+                            // },
+                            {
+                                'name' : 'Combine exponentials',
+                                'snippet' : [
+                                    'powsimp(exp(y) * exp(z))',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand logarithms for general arguments',
+                                'snippet' : [
+                                    'expand_log(log(x*y))',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand logarithms, forcing assumptions',
+                                'snippet' : [
+                                    'expand_log(log(z**2), force=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Combine logarithms for general arguments',
+                                'snippet' : [
+                                    'logcombine(log(x) + z*log(y))',
+                                ],
+                            },
+                            {
+                                'name' : 'Combine logarithms, forcing assumptions',
+                                'snippet' : [
+                                    'logcombine(log(x) + z*log(y))',
+                                ],
+                            },
+                            {
+                                'name' : 'Simplification, possibly to trig functions',
+                                'snippet' : [
+                                    'exptrigsimp(exp(z) + exp(-z))',
+                                ],
+                            },
+                        ],
+                    },
+                    {
+                        'name' : 'Trigonometric functions',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Expansion',
+                                'snippet' : [
+                                    'expr = sin(x + y)',
+                                    'expr = expand(expr, trig=True)',
+                                ],
+                            },
+                            {
+                                'name' : 'Simplification',
+                                'snippet' : [
+                                    'expr = sin(x)**4 - 2*cos(x)**2*sin(x)**2 + cos(x)**4',
+                                    'expr = trigsimp(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Simplification, possibly to exponentials',
+                                'snippet' : [
+                                    'expr = cosh(z) - sinh(z)',
+                                    'expr = exptrigsimp(expr)',
+                                ],
+                            },
+                        ],
+                    },
+                    {
+                        'name' : 'Miscellaneous',
+                        'sub-menu' : [
+                            {
+                                'name' : 'Simplify factorials',
+                                'snippet' : [
+                                    'expr = factorial(n)/factorial(n - 3)',
+                                    'expr = combsimp(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Simplify binomials',
+                                'snippet' : [
+                                    'expr = binomial(n+1, k+1)/binomial(n, k)',
+                                    'expr = combsimp(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Simplify numerical expressions to exact values',
+                                'snippet' : [
+                                    'nsimplify(4.0/(1+sqrt(5.0)), constants=[GoldenRatio,])',
+                                ],
+                            },
+                            {
+                                'name' : 'Expand gamma functions',
+                                'snippet' : [
+                                    'expr = gamma(z+3)',
+                                    'expr = expand_func(expr)',
+                                ],
+                            },
+                            {
+                                'name' : 'Simplify Bessel functions',
+                                'snippet' : [
+                                    'expr = besselj(x, z*polar_lift(-1))',
+                                    'expr = besselsimp(expr)',
+                                ],
+                            },
+                        ],
+                    },
+                ],
+            },
+        ],
+    };
+});

.local/share/jupyter/nbextensions/toc2/toc2.js

@@ -0,0 +1,826 @@
+(requirejs.specified('base/js/namespace') ? define : function(deps, callback) {
+    "use strict";
+    // if here, the Jupyter namespace hasn't been specified to be loaded.
+    // This means that we're probably embedded in a page, so we need to make
+    // our definition with a specific module name
+    return define('nbextensions/toc2/toc2', deps, callback);
+})(['jquery', 'require'], function($, requirejs) {
+    "use strict";
+
+    var IPython;
+    var events;
+    var liveNotebook = false;
+    var all_headers = $("#notebook").find(":header");
+
+    // default values for system-wide configurable parameters
+    var default_cfg = {
+        colors: {
+            hover_highlight: '#DAA520',
+            selected_highlight: '#FFD700',
+            running_highlight: '#FF0000',
+            wrapper_background: '#FFFFFF',
+            sidebar_border: '#EEEEEE',
+            navigate_text: '#333333',
+            navigate_num: '#000000',
+            on_scroll: '#2447f0',
+        },
+        collapse_to_match_collapsible_headings: false,
+        markTocItemOnScroll: true,
+        moveMenuLeft: true,
+        navigate_menu: true,
+        threshold: 4,
+        widenNotebook: false,
+    };
+    // default values for per-notebook configurable parameters
+    var metadata_settings = {
+        nav_menu: {},
+        number_sections: true,
+        sideBar: true,
+        skip_h1_title: false,
+        base_numbering: 1,
+        title_cell: 'Table of Contents',
+        title_sidebar: 'Contents',
+        toc_cell: false,
+        toc_position: {},
+        toc_section_display: true,
+        toc_window_display: false,
+    };
+    $.extend(true, default_cfg, metadata_settings);
+
+    /**
+     *  Read our config from server config & notebook metadata
+     *  This function should only be called when both:
+     *   1. the notebook (and its metadata) has fully loaded
+     *  AND
+     *   2. Jupyter.notebook.config.loaded has resolved
+     */
+    var read_config = function () {
+        var cfg = default_cfg;
+
+        if (!liveNotebook) {
+            return cfg;
+        }
+
+        // config may be specified at system level or at document level.
+        // first, update defaults with config loaded from server
+        $.extend(true, cfg, IPython.notebook.config.data.toc2);
+        // ensure notebook metadata has toc object, cache old values
+        var md = IPython.notebook.metadata.toc || {};
+        // reset notebook metadata to remove old values
+        IPython.notebook.metadata.toc = {};
+        // then update cfg with any found in current notebook metadata
+        // and save in nb metadata (then can be modified per document)
+        Object.keys(metadata_settings).forEach(function (key) {
+            cfg[key] = IPython.notebook.metadata.toc[key] = (md.hasOwnProperty(key) ? md : cfg)[key];
+        });
+        return cfg;
+    };
+
+    // globally-used status variables:
+    var rendering_toc_cell = false;
+    // toc_position default also serves as the defaults for a non-live notebook
+    var toc_position = {height: 'calc(100% - 180px)', width: '20%', left: '10px', top: '150px'};
+
+    try {
+        // this will work in a live notebook because nbextensions & custom.js
+        // are loaded by/after notebook.js, which requires base/js/namespace
+        IPython = requirejs('base/js/namespace');
+        events = requirejs('base/js/events');
+        liveNotebook = true;
+    } catch (err) {
+        // We *are* theoretically in a non-live notebook
+        console.log('[toc2] working in non-live notebook'); //, err);
+        // in non-live notebook, there's no event structure, so we make our own
+        if (window.events === undefined) {
+            var Events = function() {};
+            window.events = $([new Events()]);
+        }
+        events = window.events;
+    }
+    var Jupyter = IPython;
+
+    var setMd = function(key, value) {
+        if (liveNotebook) {
+            var md = IPython.notebook.metadata.toc;
+            if (md === undefined) {
+                md = IPython.notebook.metadata.toc = {};
+            }
+            var old_val = md[key];
+            md[key] = value;
+            if (typeof _ !== undefined ? !_.isEqual(value, old_val) : old_val != value) {
+                IPython.notebook.set_dirty();
+            }
+        }
+        return value;
+    };
+
+    function incr_lbl(ary, h_idx) { //increment heading label  w/ h_idx (zero based)
+        ary[h_idx]++;
+        for (var j = h_idx + 1; j < ary.length; j++) {
+            ary[j] = 0;
+        }
+        return ary.slice(0, h_idx + 1);
+    }
+
+    function removeMathJaxPreview(elt) {
+        elt.children('.anchor-link, .toc-mod-link').remove();
+        elt.find("script[type='math/tex']").each(
+            function(i, e) {
+                $(e).replaceWith('$' + $(e).text() + '$')
+            })
+        elt.find("span.MathJax_Preview").remove()
+        elt.find("span.MathJax").remove()
+        return elt
+    }
+
+    var callback_toc_link_click = function(evt) {
+        // workaround for https://github.com/jupyter/notebook/issues/699
+        setTimeout(function() {
+            $.ajax()
+        }, 100);
+        evt.preventDefault();
+        // Each time a link is clicked in the toc, save the current position and target in the history
+        var currentSection = $('#toc  .highlight_on_scroll a').data('tocModifiedId')
+        if (window.history.state != null){
+            if (window.history.state.back != currentSection) {
+                window.history.pushState({'back':currentSection},"",'')
+            }
+        }
+        var trg_id = $(evt.currentTarget).attr('data-toc-modified-id');
+        window.history.pushState({'back':trg_id},"",'');
+        window.history.lastjump = trg_id;
+
+        // use native scrollIntoView method with semi-unique id
+        // ! browser native click does't follow links on all browsers
+        document.getElementById(trg_id).scrollIntoView(true)
+        if (liveNotebook) {
+            // use native document method as jquery won't cope with characters
+            // like . in an id
+            var cell = $(document.getElementById(trg_id)).closest('.cell').data('cell');
+            Jupyter.notebook.select(Jupyter.notebook.find_cell_index(cell));
+            highlight_toc_item("toc_link_click", {
+                cell: cell
+            });
+        }
+    };
+
+    //  
+    window.addEventListener('popstate', 
+        function(e) { 
+            if (e.state != null && e.state.back != null) {
+                var back_id = e.state.back;
+                document.getElementById(back_id).scrollIntoView(true)
+                if (liveNotebook) {
+                    var cell = $(document.getElementById(back_id)).closest('.cell').data('cell');
+                    Jupyter.notebook.select(Jupyter.notebook.find_cell_index(cell));
+                    highlight_toc_item("toc_link_click", {
+                        cell: cell
+                    });
+                }
+            }
+    });
+
+    var make_link = function(h, toc_mod_id) {
+        var a = $('<a>')
+            .attr({
+                'href': h.find('.anchor-link').attr('href'),
+                'data-toc-modified-id': toc_mod_id,
+            });
+        // get the text *excluding* the link text, whatever it may be
+        var hclone = h.clone();
+        hclone = removeMathJaxPreview(hclone);
+        a.html(hclone.html());
+        a.on('click', callback_toc_link_click);
+        return a;
+    };
+
+    function highlight_toc_item(evt, data) {
+        var c = $(data.cell.element);
+        if (c.length < 1) {
+            return;
+        }
+        var trg_id = c.find('.toc-mod-link').attr('id') ||
+            c.prevAll().find('.toc-mod-link').eq(-1).attr('id');
+        var highlighted_item = $();
+        if (trg_id !== undefined) {
+            highlighted_item = $('.toc a').filter(function(idx, elt) {
+                return $(elt).attr('data-toc-modified-id') === trg_id;
+            });
+        }
+        if (evt.type === 'execute') {
+            // remove the selected class and add execute class
+            // if the cell is selected again, it will be highligted as selected+running
+            highlighted_item.removeClass('toc-item-highlight-select').addClass('toc-item-highlight-execute');
+        } else {
+            $('.toc .toc-item-highlight-select').removeClass('toc-item-highlight-select');
+            highlighted_item.addClass('toc-item-highlight-select');
+        }
+    }
+
+    var create_navigate_menu = function(cfg, callback) {
+        $('#kernel_menu').parent().after('<li id="Navigate"/>')
+        $('#Navigate').addClass('dropdown').append($('<a/>').attr('href', '#').attr('id', 'Navigate_sub'))
+        $('#Navigate_sub').text('Navigate').addClass('dropdown-toggle').attr('data-toggle', 'dropdown')
+        $('#Navigate').append($('<ul/>').attr('id', 'Navigate_menu').addClass('dropdown-menu')
+            .append($("<div/>").attr("id", "navigate_menu").addClass('toc')))
+
+        if (cfg['nav_menu']) {
+            $('#Navigate_menu').css(cfg['nav_menu'])
+            $('#navigate_menu').css('width', $('#Navigate_menu').css('width'))
+            $('#navigate_menu').css('height', $('#Navigate_menu').height())
+        } else {
+            cfg.nav_menu = {};
+            events.on("before_save.Notebook",
+                function() {
+                    try {
+                        cfg.nav_menu['width'] = $('#Navigate_menu').css('width')
+                        cfg.nav_menu['height'] = $('#Navigate_menu').css('height')
+                    } catch (e) {
+                        console.log("[toc2] Error in metadata (navigation menu) - Proceeding", e)
+                    }
+                })
+        }
+
+        $('#Navigate_menu').resizable({
+            resize: function(event, ui) {
+                $('#navigate_menu').css('width', $('#Navigate_menu').css('width'))
+                $('#navigate_menu').css('height', $('#Navigate_menu').height())
+            },
+            stop: function(event, ui) {
+                cfg.nav_menu['width'] = $('#Navigate_menu').css('width')
+                cfg.nav_menu['height'] = $('#Navigate_menu').css('height')
+            }
+        })
+
+        callback && callback();
+    }
+
+    function setNotebookWidth(cfg, st) {
+        var margin = 20;
+        var nb_inner = $('#notebook-container');
+        var nb_wrap_w = $('#notebook').width();
+        var sidebar = $('#toc-wrapper');
+        var visible_sidebar = cfg.sideBar && sidebar.is(':visible');
+        var sidebar_w = visible_sidebar ? sidebar.outerWidth() : 0;
+        var available_space = nb_wrap_w - 2 * margin - sidebar_w;
+        var inner_css = {marginLeft: '', width: ''};
+        if (cfg.widenNotebook) {
+            inner_css.width = available_space;
+        }
+        if (visible_sidebar) {
+            var nb_inner_w = nb_inner.outerWidth();
+            inner_css.marginLeft = sidebar_w + margin; // shift notebook rightward to fit the sidebar in
+            if (available_space <= nb_inner_w) {
+                inner_css.width = available_space; // also slim notebook to fit sidebar
+            }
+        }
+        nb_inner.css(inner_css);
+    }
+
+    var saveTocPosition = function () {
+        var toc_wrapper = $('#toc-wrapper');
+        var new_values = toc_wrapper.hasClass('sidebar-wrapper') ? ['width'] : ['left', 'top', 'height', 'width'];
+        $.extend(toc_position, toc_wrapper.css(new_values));
+        setMd('toc_position', toc_position);
+    };
+
+    var makeUnmakeMinimized = function (cfg, animate) {
+        var open = cfg.sideBar || cfg.toc_section_display;
+        var new_css, wrap = $('#toc-wrapper');
+        var anim_opts = {duration: animate ? 'fast' : 0};
+        if (open) {
+            $('#toc').show();
+            new_css = cfg.sideBar ? {} : {height: toc_position.height, width: toc_position.width};
+        }
+        else {
+            new_css = {
+                height: wrap.outerHeight() - wrap.find('#toc').outerHeight(),
+            };
+            anim_opts.complete = function () {
+                $('#toc').hide();
+                $('#toc-wrapper').css('width', '');
+            };
+        }
+        wrap.toggleClass('closed', !open)
+            .animate(new_css, anim_opts)
+            .find('.hide-btn').attr('title', open ? 'Hide ToC' : 'Show ToC');
+        return open;
+    };
+
+    var makeUnmakeSidebar = function (cfg) {
+        var make_sidebar = cfg.sideBar;
+        var wrap = $('#toc-wrapper')
+            .toggleClass('sidebar-wrapper', make_sidebar)
+            .toggleClass('float-wrapper', !make_sidebar)
+            .resizable('option', 'handles', make_sidebar ? 'e' : 'all');
+        wrap.children('.ui-resizable-se').toggleClass('ui-icon', !make_sidebar);
+        wrap.children('.ui-resizable-e').toggleClass('ui-icon ui-icon-grip-dotted-vertical', make_sidebar);
+        if (make_sidebar) {
+            var sidebar_top = liveNotebook ? document.getElementById('site').top : 0
+            wrap.css({top: sidebar_top,height: "",left: 0});
+        }
+        else {
+            wrap.css({height: toc_position.height});
+        }
+        setNotebookWidth(cfg);
+    };
+
+    var create_toc_div = function(cfg, st) {
+
+        var callbackPageResize = function (evt) {
+            setNotebookWidth(cfg);
+        };
+
+        var toc_wrapper = $('<div id="toc-wrapper"/>')
+            .css('display', 'none')
+            .append(
+                $('<div id="toc-header"/>')
+                .append('<span class="header"/>')
+                .append(
+                    $('<i class="fa fa-fw hide-btn" title="Hide ToC">')
+                    .on('click', function (evt) {
+                        cfg.toc_section_display = setMd('toc_section_display', !cfg.toc_section_display);
+                        makeUnmakeMinimized(cfg, true);
+                    })
+                ).append(
+                    $('<i class="fa fa-fw fa-refresh" title="Reload ToC">')
+                    .on('click', function(evt) {
+                        var icon = $(evt.currentTarget).addClass('fa-spin');
+                        table_of_contents(cfg, st);
+                        icon.removeClass('fa-spin');
+                    })
+                ).append(
+                    $('<i class="fa fa-fw fa-cog" title="ToC settings"/>')
+                    .on('click', function(evt) {
+                        show_settings_dialog(cfg, st);
+                    })
+                )
+            ).append(
+                $("<div/>").attr("id", "toc").addClass('toc')
+            )
+            .prependTo(liveNotebook ? '#site' : document.body);
+
+        // enable dragging and save position on stop moving
+        toc_wrapper.draggable({
+            drag: function(event, ui) {
+                var make_sidebar = ui.position.left < 20; // 20 is snapTolerance
+                if (make_sidebar) {
+                    ui.position.top = liveNotebook ? document.getElementById('site').top : 0
+                    ui.position.left = 0;
+                }
+                if (make_sidebar !== cfg.sideBar) {
+                    cfg.toc_section_display = setMd('toc_section_display', true);
+                    cfg.sideBar = setMd('sideBar', make_sidebar);
+                    makeUnmakeMinimized(cfg);
+                    makeUnmakeSidebar(cfg);
+                }
+            }, //end of drag function
+            stop: saveTocPosition,
+            containment: 'parent',
+            snap: 'body, #site',
+            snapTolerance: 20,
+        });
+
+        toc_wrapper.resizable({
+            handles: 'all',
+            resize: function(event, ui) {
+                if (cfg.sideBar) {
+                    // unset the height set by jquery resizable
+                    $('#toc-wrapper').css('height', '');
+                    setNotebookWidth(cfg, st)
+                }
+            },
+            start: function(event, ui) {
+                if (!cfg.sideBar) {
+                    cfg.toc_section_display = setMd('toc_section_display', true);
+                    makeUnmakeMinimized(cfg);
+                }
+            },
+            stop: saveTocPosition,
+            containment: 'parent',
+            minHeight: 100,
+            minWidth: 165,
+        });
+
+        // On header/menu/toolbar resize, resize the toc itself
+        $(window).on('resize', callbackPageResize);
+        if (liveNotebook) {
+            events.on("resize-header.Page toggle-all-headers", callbackPageResize);
+            $.extend(toc_position, IPython.notebook.metadata.toc.toc_position);
+        }
+        else {
+            // default to true for non-live notebook
+            cfg.toc_window_display = true;
+        }
+        // restore toc position at load
+        toc_wrapper.css(cfg.sideBar ? {width: toc_position.width} : toc_position);
+        // older toc2 versions stored string representations, so update those
+        if (cfg.toc_window_display === 'none') {
+            cfg.toc_window_display = setMd('toc_window_display', false);
+        }
+        if (cfg.toc_section_display === 'none') {
+            cfg.toc_section_display = setMd('toc_section_display', false);
+        }
+        toc_wrapper.toggle(cfg.toc_window_display);
+        makeUnmakeSidebar(cfg);
+        $("#toc_button").toggleClass('active', cfg.toc_window_display);
+        if (!cfg.toc_section_display) {
+            makeUnmakeMinimized(cfg);
+        }
+    };
+
+    //----------------------------------------------------------------------------
+    // on scroll - mark the toc item corresponding to the first header visible in
+    // the viewport with 'highlight_on_scroll' class
+    // some elements from https://stackoverflow.com/questions/20791374/jquery-check-if-element-is-visible-in-viewport
+    function highlightTocItemOnScroll(cfg, st) {
+        if (cfg.markTocItemOnScroll) {
+            var scrolling_elt = liveNotebook ? '#site' : window
+            $(scrolling_elt).scroll(function() {
+                var headerVisibleHeight = $('#header').is(':visible') ? $('#header').height() : 0
+                var headerHeight = liveNotebook ? headerVisibleHeight : 0
+                var bottom_of_screen = $(window).scrollTop() + $(scrolling_elt).height() + headerHeight;
+                var top_of_screen = $(window).scrollTop() + headerHeight;
+                //loop over all headers
+                all_headers.each(function(i, h) {
+                    var top_of_element = $(h).offset().top;
+
+                    if ((bottom_of_screen > top_of_element) && (top_of_screen < top_of_element)) {
+                        // The element is visible
+                        var trg_id = $(h).attr('data-toc-modified-id')
+                        if (trg_id !== undefined) {
+                            var highlighted_item = $('#toc a').filter(function(idx, elt) {
+                                return $(elt).attr('data-toc-modified-id') === trg_id;
+                            });
+                            $('#toc .highlight_on_scroll').removeClass('highlight_on_scroll')
+                            highlighted_item.parent().addClass('highlight_on_scroll')
+                        }
+                        return false;
+                    } else {
+                        // The element is not visible
+                        // If the current header is already below the viewport then break
+                        if (bottom_of_screen < top_of_element) return false
+                        else return
+                    }
+                })
+            });
+        }
+    }
+    //----------------------------------------------------------------------------
+    // TOC CELL -- if cfg.toc_cell=true, add and update a toc cell in the notebook.
+    //             This cell, initially at the very beginning, can be moved.
+    //             Its contents are automatically updated.
+    //             Optionnaly, the sections in the toc can be numbered.
+
+    function process_cell_toc(cfg, st) {
+        var new_html = '<h1>' +
+            $('<div>').text(cfg.title_cell).html() + '<span class="tocSkip"></span></h1>\n' +
+            '<div class="toc">' +
+            $('#toc').html() +
+            '</div>';
+        if (!liveNotebook) {
+            if (cfg.toc_cell) {
+                $('.cell > .toc').parent(':has(.tocSkip)')
+                    .html(new_html)
+                    .find('.toc-item li a')
+                        .on('click', callback_toc_link_click);
+            }
+            return;
+        }
+        var cell_toc;
+        // look for a possible toc cell
+        var cells = IPython.notebook.get_cells();
+        var lcells = cells.length;
+        for (var i = 0; i < lcells; i++) {
+            if (cells[i].metadata.toc) {
+                // delete if we don't want it
+                if (!cfg.toc_cell) {
+                    return IPython.notebook.delete_cell(i);
+                }
+                cell_toc = cells[i];
+                break;
+            }
+        }
+        //if toc_cell=true, we want a cell_toc.
+        //  If it does not exist, create it at the beginning of the notebook
+        if (cfg.toc_cell) {
+            if (cell_toc === undefined) {
+                // set rendering_toc_cell flag to avoid loop on insert_cell_above
+                rendering_toc_cell = true;
+                cell_toc = IPython.notebook.insert_cell_above('markdown', 0);
+                cell_toc.metadata.toc = true;
+                rendering_toc_cell = false;
+            }
+            // set rendering_toc_cell flag to avoid loop on render
+            rendering_toc_cell = true;
+            cell_toc.set_text(new_html);
+            cell_toc.render();
+            rendering_toc_cell = false;
+            cell_toc.element.find('.toc-item li a').on('click', callback_toc_link_click);
+        }
+    } //end function process_cell_toc --------------------------
+
+    var collapse_by_id = function(trg_id, show, trigger_event) {
+        var anchors = $('.toc .toc-item > li > span > a').filter(function(idx, elt) {
+            return $(elt).attr('data-toc-modified-id') === trg_id;
+        });
+        anchors.siblings('i')
+            .toggleClass('fa-caret-right', !show)
+            .toggleClass('fa-caret-down', show);
+        anchors.parent().siblings('ul')[show ? 'slideDown' : 'slideUp']('fast');
+        if (trigger_event !== false) {
+            // fire event for collapsible_heading to catch
+            var cell = $(document.getElementById(trg_id)).closest('.cell').data('cell');
+            events.trigger((show ? 'un' : '') + 'collapse.Toc', {
+                cell: cell
+            });
+        }
+    };
+
+    var callback_toc2_collapsible_headings = function(evt, data) {
+        var trg_id = data.cell.element.find(':header').filter(function(idx, elt) {
+            return Boolean($(elt).attr('data-toc-modified-id'));
+        }).attr('data-toc-modified-id');
+        var show = evt.type.indexOf('un') >= 0;
+        // use trigger_event false to avoid re-triggering collapsible_headings
+        collapse_by_id(trg_id, show, false);
+    };
+
+    var callback_collapser = function(evt) {
+        var clicked_i = $(evt.currentTarget);
+        var trg_id = clicked_i.siblings('a').attr('data-toc-modified-id');
+        var show = clicked_i.hasClass('fa-caret-right');
+        collapse_by_id(trg_id, show);
+    };
+
+    // Table of Contents =================================================================
+    var table_of_contents = function(cfg, st) {
+
+        // if this call is a result of toc_cell rendering, do nothing to avoid
+        // looping, as we're already in a table_of_contents call
+        if (rendering_toc_cell) {
+            return
+        }
+
+        // In a live notebook, read_config will have been called already, but
+        // in non-live notebooks, ensure that all config values are defined.
+        if (!liveNotebook) {
+            cfg = $.extend(true, {}, default_cfg, cfg);
+        }
+
+        var toc_wrapper = $("#toc-wrapper");
+        if (toc_wrapper.length === 0) { // toc window doesn't exist at all
+            create_toc_div(cfg, st); // create it
+            highlightTocItemOnScroll(cfg, st); // initialize highlighting on scroll
+        }
+        var ul = $('<ul/>').addClass('toc-item');
+
+        // update sidebar/window title
+        $('#toc-header > .header').text(cfg.title_sidebar + ' ');
+
+        // update toc element
+        $("#toc").empty().append(ul);
+
+        var depth = 1;
+        // update all headers with id that are in rendered text cell outputs,
+        // excepting any header which contains an html tag with class 'tocSkip'
+        // eg in ## title <a class='tocSkip'>,
+        // or the ToC cell.
+        all_headers = $('.text_cell_render').find('[id]:header:not(:has(.tocSkip))');
+        var min_lvl = 1 + Number(Boolean(cfg.skip_h1_title)),
+            lbl_ary = [];
+        for (; min_lvl <= 6; min_lvl++) {
+            if (all_headers.is('h' + min_lvl)) {
+                break;
+            }
+        }
+        lbl_ary[0] = cfg.base_numbering-1 // begin numbering at base_numbering
+        for (var i = min_lvl+1; i <= 6; i++) {
+            lbl_ary[i - min_lvl] = 0;
+        }
+
+        //loop over all headers
+        all_headers.each(function(i, h) {
+            // remove pre-existing number
+            $(h).children('.toc-item-num').remove();
+
+            var level = parseInt(h.tagName.slice(1), 10) - min_lvl + 1;
+            // skip below threshold, or h1 ruled out by cfg.skip_h1_title
+            if (level < 1 || level > cfg.threshold) {
+                return;
+            }
+            h = $(h);
+            // numbered heading labels
+            var num_str = incr_lbl(lbl_ary, level - 1).join('.');
+            if (cfg.number_sections) {
+                $('<span>')
+                    .text(num_str + '\u00a0\u00a0')
+                    .addClass('toc-item-num')
+                    .prependTo(h);
+            }
+
+            // walk down levels
+            for (; depth < level; depth++) {
+                var li = ul.children('li:last-child');
+                if (li.length < 1) {
+                    li = $('<li>').appendTo(ul);
+                }
+                ul = $('<ul class="toc-item">').appendTo(li);
+            }
+            // walk up levels
+            for (; depth > level; depth--) {
+                ul = ul.parent().closest('.toc-item');
+            }
+
+            var toc_mod_id = h.attr('id') + '-' + num_str;
+            h.attr('data-toc-modified-id', toc_mod_id);
+            // add an anchor with modified id (if it doesn't already exist)
+            h.children('.toc-mod-link').remove();
+            $('<a>').addClass('toc-mod-link').attr('id', toc_mod_id).prependTo(h);
+
+            // Create toc entry, append <li> tag to the current <ol>.
+            ul.append(
+                $('<li>').append(
+                    $('<span>').append(
+                        make_link(h, toc_mod_id))));
+        });
+
+        // update navigation menu
+        if (cfg.navigate_menu) {
+            var pop_nav = function() { //callback for create_nav_menu
+                $('#navigate_menu').empty().append($('#toc > .toc-item').clone());
+            }
+            if ($('#Navigate_menu').length == 0) {
+                create_navigate_menu((liveNotebook ? IPython.notebook.metadata.toc : cfg), pop_nav);
+            } else {
+                pop_nav()
+            }
+        } else { // If navigate_menu is false but the menu already exists, then remove it
+            if ($('#Navigate_menu').length > 0) $('#Navigate_sub').remove()
+        }
+
+        // if cfg.toc_cell=true, find/add and update a toc cell in the notebook.
+        process_cell_toc(cfg, st);
+
+        // add collapse controls
+        $('<i>')
+            .addClass('fa fa-fw fa-caret-down')
+            .on('click', callback_collapser) // callback
+            .prependTo('.toc li:has(ul) > span'); // only if li has descendants
+        $('<i>').addClass('fa fa-fw ').prependTo('.toc li:not(:has(ul)) > span'); // otherwise still add <i> to keep things aligned
+
+        events[cfg.collapse_to_match_collapsible_headings ? 'on' : 'off'](
+            'collapse.CollapsibleHeading uncollapse.CollapsibleHeading', callback_toc2_collapsible_headings);
+    };
+
+    var toggle_toc = function(cfg, st) {
+        // toggle draw (first because of first-click behavior)
+        var wrap = $("#toc-wrapper");
+        var show = wrap.is(':hidden');
+        wrap.toggle(show);
+        cfg['toc_window_display'] = setMd('toc_window_display', show);
+        setNotebookWidth(cfg);
+        table_of_contents(cfg);
+        $("#toc_button").toggleClass('active', show);
+    };
+
+    var show_settings_dialog = function (cfg, st) {
+
+        var callback_setting_change = function (evt) {
+            var input = $(evt.currentTarget);
+            var md_key = input.attr('tocMdKey');
+            cfg[md_key] = setMd(md_key, input.attr('type') == 'checkbox' ? Boolean(input.prop('checked')) : input.val());
+            table_of_contents(cfg, st);
+        };
+        var build_setting_input = function (md_key, md_label, input_type) {
+            var opts = liveNotebook ? IPython.notebook.metadata.toc : cfg;
+            var id = 'toc-settings-' + md_key;
+            var fg = $('<div>').append(
+                $('<label>').text(md_label).attr('for', id));
+            var input = $('<input/>').attr({
+                type: input_type || 'text', id: id, tocMdKey: md_key,
+            }).on('change', callback_setting_change);
+            if (input_type == 'checkbox') {
+                fg.addClass('checkbox');
+                input
+                    .prop('checked', opts[md_key])
+                    .prependTo(fg.children('label'));
+            }
+            else {
+                fg.addClass('form-group');
+                input
+                    .addClass('form-control')
+                    .val(opts[md_key])
+                    .appendTo(fg);
+            }
+            return fg;
+        };
+
+        var modal = $('<div class="modal fade" role="dialog"/>');
+        var dialog_content = $("<div/>")
+            .addClass("modal-content")
+            .appendTo($('<div class="modal-dialog">').appendTo(modal));
+        $('<div class="modal-header">')
+            .append('<button type="button" class="close" data-dismiss="modal" aria-hidden="true">&times;</button>')
+            .append('<h4 class="modal-title">ToC2 settings</h4>')
+            .on('mousedown', function() { $('.modal').draggable({handle: '.modal-header'});})
+            .appendTo(dialog_content);
+        $('<div>')
+            .addClass('modal-body')
+            .append([
+                $('<div>').text(
+                    'These settings apply to this notebook only, and are stored in its metadata. ' +
+                    liveNotebook ? 'The defaults for new notebooks can be edited from the nbextensions configurator.' :
+                    'The settings won\'t persist in non-live notebooks though.'),
+                build_setting_input('number_sections', 'Automatically number headings', 'checkbox'),
+                build_setting_input('skip_h1_title', 'Leave h1 items out of ToC', 'checkbox'),
+                build_setting_input('base_numbering', 'Begin numbering at'),
+                build_setting_input('toc_cell', 'Add notebook ToC cell', 'checkbox'),
+                build_setting_input('title_cell', 'ToC cell title'),
+                build_setting_input('title_sidebar', 'Sidebar/window title'),
+                build_setting_input('sideBar', 'Display as a sidebar (otherwise as a floating window)', 'checkbox'),
+                build_setting_input('toc_window_display', 'Display ToC window/sidebar at startup', 'checkbox'),
+                build_setting_input('toc_section_display', 'Expand window/sidebar at startup', 'checkbox'),
+            ])
+            .appendTo(dialog_content);
+        $('<div class="modal-footer">')
+            .append('<button class="btn btn-default btn-sm btn-primary" data-dismiss="modal">Ok</button>')
+            .appendTo(dialog_content);
+        // focus button on open
+        modal.on('shown.bs.modal', function () {
+            setTimeout(function () {
+                dialog_content.find('.modal-footer button').last().focus();
+            }, 0);
+        });
+
+        if (liveNotebook) {
+            Jupyter.notebook.keyboard_manager.disable();
+            modal.on('hidden.bs.modal', function () {
+                modal.remove(); // destroy modal on hide
+                Jupyter.notebook.keyboard_manager.enable();
+                Jupyter.notebook.keyboard_manager.command_mode();
+                var cell = Jupyter.notebook.get_selected_cell();
+                if (cell) cell.select();
+            });
+        }
+
+        // Try to use bootstrap modal, but bootstrap's js may not be available
+        // (e.g. as in non-live notebook), so we provide a poor-man's version
+        try {
+            return modal.modal({backdrop: 'static'});
+        }
+        catch (err) {
+            // show the backdrop
+            $(document.body).addClass('modal-open');
+            var $backdrop = $('<div class="modal-backdrop fade">').appendTo($(document.body));
+            $backdrop[0].offsetWidth; // force reflow
+            $backdrop.addClass('in');
+            // hook up removals
+            modal.on('click', '[data-dismiss="modal"]', function modal_close() {
+                // hide the modal foreground
+                modal.removeClass('in');
+                setTimeout(function on_foreground_hidden() {
+                    modal.remove();
+                    // now hide the backdrop
+                    $backdrop.removeClass('in');
+                    // wait for transition
+                    setTimeout(function on_backdrop_hidden() {
+                        $(document.body).removeClass('modal-open');
+                        $backdrop.remove();
+                    }, 150);
+                }, 300);
+            });
+            // wait for transition
+            setTimeout(function () {
+                // now show the modal foreground
+                modal.appendTo(document.body).show().scrollTop(0);
+                modal[0].offsetWidth; // force reflow
+                modal.addClass('in');
+                // wait for transition, then trigger callbacks
+                setTimeout(function on_foreground_shown() {
+                    modal.trigger('shown.bs.modal');
+                }, 300);
+            }, 150);
+            return modal;
+        }
+    };
+
+    return {
+        highlight_toc_item: highlight_toc_item,
+        table_of_contents: table_of_contents,
+        toggle_toc: toggle_toc,
+        read_config: read_config,
+    };
+});
+// export table_of_contents to global namespace for backwards compatibility
+// Do export synchronously, so that it's defined as soon as this file is loaded
+if (!requirejs.specified('base/js/namespace')) {
+    window.table_of_contents = function(cfg, st) {
+        "use strict";
+        // use require to ensure the module is correctly loaded before the
+        // actual call is made
+        requirejs(['nbextensions/toc2/toc2'], function(toc2) {
+            toc2.table_of_contents(cfg, st);
+        });
+    };
+}

.local/share/jupyter/nbextensions/toc2/toc2.yaml

@@ -0,0 +1,104 @@
+Type: Jupyter Notebook Extension
+Name: Table of Contents (2)
+Description: The toc2 extension enables to collect all running headers and display them in a floating window, as a sidebar or with a navigation menu. The extension is also draggable, resizable, collapsable, dockable and features automatic numerotation with unique links ids, and an optional toc cell.
+Link: README.md
+Icon: icon.png
+Main: main.js
+Compatibility: 4.x, 5.x
+Parameters:
+- name: toc2.number_sections
+  description: Automatically number notebook's sections
+  input_type: checkbox
+  default: true
+- name: toc2.threshold
+  description: Maximum level of nested sections to display on the tables of contents
+  input_type: number
+  min: -1
+  step: 1
+  default: 4
+
+- name: toc2.skip_h1_title
+  description: |
+    Skip h1 headings from numbering, so that they can serve as a notebook title.
+    See the README for details, caveats and alternatives
+  input_type: checkbox
+  default: false
+
+- name: toc2.toc_cell
+  description: Add a Table of Contents cell at the top of the notebook
+  input_type: checkbox
+  default: false
+
+- name: toc2.title_cell
+  description: Default heading used for ToC cell (can also be set per-notebook)
+  default: 'Table of Contents'
+
+- name: toc2.title_sidebar
+  description: Default title used for ToC sidebar/window (can also be set per-notebook)
+  default: 'Contents'
+
+- name: toc2.toc_window_display
+  description: Display toc window/sidebar at startup
+  input_type: checkbox
+  default: false
+- name: toc2.sideBar
+  description: Display Table of Contents as a sidebar (otherwise as a floating window)
+  input_type: checkbox
+  default: true
+- name: toc2.markTocItemOnScroll
+  description: Mark toc item of header in viewport when scrolling
+  input_type: checkbox
+  default: true
+- name: toc2.widenNotebook
+  description: Widen the display area to fit the browser window (may be useful with sidebar option)
+  input_type: checkbox
+  default: true
+- name: toc2.navigate_menu
+  description: Display Table of Contents as a navigation menu
+  input_type: checkbox
+  default: true
+- name: toc2.moveMenuLeft
+  description: Move notebook's title and menu on the left instead of being centered -- This provides a better look when the toc/sidebar is present
+  input_type: checkbox
+  default: true
+
+- name: toc2.collapse_to_match_collapsible_headings
+  input_type: checkbox
+  default: false
+  description: |
+    Collapse/uncollapse ToC sections when the collapsible_headings nbextension
+    is used to collapse/uncollapse sections in the notebook. For the inverse
+    behaviour, see collapsible_headings' configuration
+
+- name: toc2.colors.hover_highlight
+  input_type: color
+  description: Hover color in toc
+  default: "#DAA520"
+- name: toc2.colors.on_scroll
+  input_type: color
+  description: Color of highlight mark on scrolling
+  default: '#2447f0'
+- name: toc2.colors.selected_highlight
+  input_type: color
+  description: Color of sections with selected elements
+  default: "#FFD700"
+- name: toc2.colors.running_highlight
+  input_type: color
+  description: Color of sections with running cells
+  default: "#FF0000"
+- name: toc2.colors.wrapper_background
+  input_type: color
+  description: Color of wrapper window background
+  default: "#FFFFFF"
+- name: toc2.colors.sidebar_border
+  input_type: color
+  description: Color of sidebar border
+  default: "#EEEEEE"
+- name: toc2.colors.navigate_text
+  input_type: color
+  description: Color of navigate text
+  default: "#333333"
+- name: toc2.colors.navigate_num
+  input_type: color
+  description: Color of navigate number
+  default: "#000000"

.local/share/jupyter/nbextensions/toggle_all_line_numbers/main.js

@@ -0,0 +1,82 @@
+// toggle on/off linenumber display in all codecells
+
+define([
+    'jquery',
+    'base/js/namespace'
+], function(
+    $,
+    Jupyter
+) {
+    "use strict";
+
+    // define default values for config parameters
+    var params = {
+        toggle_all_linenumbers_hotkey : 'Alt-N',
+        toggle_all_linenumbers_enable_hotkey : true
+    };
+
+    // to be called once config is loaded, this updates default config vals
+    // with the ones specified by the server's config file
+    var update_params = function() {
+        var config = Jupyter.notebook.config;
+        for (var key in params) {
+            if (config.data.hasOwnProperty(key) ){
+                params[key] = config.data[key];
+            }
+        }
+    };
+
+    var toggle_all = function() {
+        var toolbar_button = $('#toggle_all_linenumbers');
+        toolbar_button.toggleClass('active', !toolbar_button.hasClass('active'));
+        var cells = Jupyter.notebook.get_cells();
+        for(var i in cells ){
+            cells[i].toggle_line_numbers();
+        }
+    };
+
+    // define action, register with ActionHandler instance
+    var prefix = 'auto';
+    var action_name = 'toggle-all-line-numbers';
+    var action = {
+        icon: 'fa-list-ol',
+        help: 'Toggle linenumbers in all codecells',
+        help_index : 'zz',
+        id: 'toggle_all_linenumbers',
+        handler: toggle_all
+    };
+    var action_full_name; // will be set on registration
+
+    var initialize = function () {
+        // update default config vals with the newly loaded ones
+        update_params();
+
+        // register actions with ActionHandler instance
+        action_full_name = Jupyter.keyboard_manager.actions.register(action, action_name, prefix);
+
+        // create toolbar button
+        Jupyter.toolbar.add_buttons_group([action_full_name]);
+
+        // (maybe) define hotkey
+        if (params.toggle_all_linenumbers_enable_hotkey &&
+            params.toggle_all_linenumbers_hotkey) {
+
+            console.log('toggle_all_linenumbers enabling hotkey:',
+                        params.toggle_all_linenumbers_hotkey);
+
+            Jupyter.keyboard_manager.edit_shortcuts.add_shortcut(
+                params.toggle_all_linenumbers_hotkey, action_full_name);
+            Jupyter.keyboard_manager.command_shortcuts.add_shortcut(
+                params.toggle_all_linenumbers_hotkey, action_full_name);
+        }
+    };
+
+    var load_ipython_extension = function() {
+        return Jupyter.notebook.config.loaded.then(initialize);
+    };
+
+    var extension = {
+        load_ipython_extension : load_ipython_extension
+    };
+    return extension;
+});

.local/share/jupyter/nbextensions/toggle_all_line_numbers/main.yaml

@@ -0,0 +1,16 @@
+Type: IPython Notebook Extension
+Name: Toggle all line numbers
+Description: "Add a toolbar button and hotkey to toggle all cells' line numbers on or off"
+Icon: icon.png
+Link: readme.md
+Main: main.js
+Compatibility: 3.x, 4.x, 5.x
+Parameters:
+- name: toggle_all_linenumbers_hotkey
+  description: Hotkey combination to which to bind linenumber-toggling
+  input_type: hotkey
+  default: Alt-N
+- name: toggle_all_linenumbers_enable_hotkey
+  description: Enable the hotkey
+  input_type: checkbox
+  default: true

.local/share/jupyter/nbextensions/toggle_all_line_numbers/readme.md

@@ -0,0 +1,5 @@
+Toggle all line numbers
+=======================
+This extension adds a toolbar button, along with an optional hotkey,
+to toggle all cells' line numbers on or off in one action.
+

.local/share/jupyter/nbextensions/varInspector/jquery.tablesorter.min.js

@@ -0,0 +1,2 @@
+!function(a){"function"==typeof define&&define.amd?define(["jquery"],a):"object"==typeof module&&"object"==typeof module.exports?module.exports=a(require("jquery")):a(jQuery)}(function(a){return function(a){"use strict";var b=a.tablesorter={version:"2.25.7",parsers:[],widgets:[],defaults:{theme:"default",widthFixed:!1,showProcessing:!1,headerTemplate:"{content}",onRenderTemplate:null,onRenderHeader:null,cancelSelection:!0,tabIndex:!0,dateFormat:"mmddyyyy",sortMultiSortKey:"shiftKey",sortResetKey:"ctrlKey",usNumberFormat:!0,delayInit:!1,serverSideSorting:!1,resort:!0,headers:{},ignoreCase:!0,sortForce:null,sortList:[],sortAppend:null,sortStable:!1,sortInitialOrder:"asc",sortLocaleCompare:!1,sortReset:!1,sortRestart:!1,emptyTo:"bottom",stringTo:"max",duplicateSpan:!0,textExtraction:"basic",textAttribute:"data-text",textSorter:null,numberSorter:null,widgets:[],widgetOptions:{zebra:["even","odd"]},initWidgets:!0,widgetClass:"widget-{name}",initialized:null,tableClass:"",cssAsc:"",cssDesc:"",cssNone:"",cssHeader:"",cssHeaderRow:"",cssProcessing:"",cssChildRow:"tablesorter-childRow",cssInfoBlock:"tablesorter-infoOnly",cssNoSort:"tablesorter-noSort",cssIgnoreRow:"tablesorter-ignoreRow",cssIcon:"tablesorter-icon",cssIconNone:"",cssIconAsc:"",cssIconDesc:"",pointerClick:"click",pointerDown:"mousedown",pointerUp:"mouseup",selectorHeaders:"> thead th, > thead td",selectorSort:"th, td",selectorRemove:".remove-me",debug:!1,headerList:[],empties:{},strings:{},parsers:[]},css:{table:"tablesorter",cssHasChild:"tablesorter-hasChildRow",childRow:"tablesorter-childRow",colgroup:"tablesorter-colgroup",header:"tablesorter-header",headerRow:"tablesorter-headerRow",headerIn:"tablesorter-header-inner",icon:"tablesorter-icon",processing:"tablesorter-processing",sortAsc:"tablesorter-headerAsc",sortDesc:"tablesorter-headerDesc",sortNone:"tablesorter-headerUnSorted"},language:{sortAsc:"Ascending sort applied, ",sortDesc:"Descending sort applied, ",sortNone:"No sort applied, ",sortDisabled:"sorting is disabled",nextAsc:"activate to apply an ascending sort",nextDesc:"activate to apply a descending sort",nextNone:"activate to remove the sort"},regex:{templateContent:/\{content\}/g,templateIcon:/\{icon\}/g,templateName:/\{name\}/i,spaces:/\s+/g,nonWord:/\W/g,formElements:/(input|select|button|textarea)/i,chunk:/(^([+\-]?(?:\d*)(?:\.\d*)?(?:[eE][+\-]?\d+)?)?$|^0x[0-9a-f]+$|\d+)/gi,chunks:/(^\\0|\\0$)/,hex:/^0x[0-9a-f]+$/i,comma:/,/g,digitNonUS:/[\s|\.]/g,digitNegativeTest:/^\s*\([.\d]+\)/,digitNegativeReplace:/^\s*\(([.\d]+)\)/,digitTest:/^[\-+(]?\d+[)]?$/,digitReplace:/[,.'"\s]/g},string:{max:1,min:-1,emptymin:1,emptymax:-1,zero:0,none:0,"null":0,top:!0,bottom:!1},keyCodes:{enter:13},dates:{},instanceMethods:{},setup:function(c,d){if(!c||!c.tHead||0===c.tBodies.length||c.hasInitialized===!0)return void(d.debug&&(c.hasInitialized?console.warn("Stopping initialization. Tablesorter has already been initialized"):console.error("Stopping initialization! No table, thead or tbody",c)));var e="",f=a(c),g=a.metadata;c.hasInitialized=!1,c.isProcessing=!0,c.config=d,a.data(c,"tablesorter",d),d.debug&&(console[console.group?"group":"log"]("Initializing tablesorter"),a.data(c,"startoveralltimer",new Date)),d.supportsDataObject=function(a){return a[0]=parseInt(a[0],10),a[0]>1||1===a[0]&&parseInt(a[1],10)>=4}(a.fn.jquery.split(".")),d.emptyTo=d.emptyTo.toLowerCase(),d.stringTo=d.stringTo.toLowerCase(),d.last={sortList:[],clickedIndex:-1},/tablesorter\-/.test(f.attr("class"))||(e=""!==d.theme?" tablesorter-"+d.theme:""),d.table=c,d.$table=f.addClass(b.css.table+" "+d.tableClass+e).attr("role","grid"),d.$headers=f.find(d.selectorHeaders),d.namespace?d.namespace="."+d.namespace.replace(b.regex.nonWord,""):d.namespace=".tablesorter"+Math.random().toString(16).slice(2),d.$table.children().children("tr").attr("role","row"),d.$tbodies=f.children("tbody:not(."+d.cssInfoBlock+")").attr({"aria-live":"polite","aria-relevant":"all"}),d.$table.children("caption").length&&(e=d.$table.children("caption")[0],e.id||(e.id=d.namespace.slice(1)+"caption"),d.$table.attr("aria-labelledby",e.id)),d.widgetInit={},d.textExtraction=d.$table.attr("data-text-extraction")||d.textExtraction||"basic",b.buildHeaders(d),b.fixColumnWidth(c),b.addWidgetFromClass(c),b.applyWidgetOptions(c),b.setupParsers(d),d.totalRows=0,d.delayInit||b.buildCache(d),b.bindEvents(c,d.$headers,!0),b.bindMethods(d),d.supportsDataObject&&"undefined"!=typeof f.data().sortlist?d.sortList=f.data().sortlist:g&&f.metadata()&&f.metadata().sortlist&&(d.sortList=f.metadata().sortlist),b.applyWidget(c,!0),d.sortList.length>0?b.sortOn(d,d.sortList,{},!d.initWidgets):(b.setHeadersCss(d),d.initWidgets&&b.applyWidget(c,!1)),d.showProcessing&&f.unbind("sortBegin"+d.namespace+" sortEnd"+d.namespace).bind("sortBegin"+d.namespace+" sortEnd"+d.namespace,function(a){clearTimeout(d.timerProcessing),b.isProcessing(c),"sortBegin"===a.type&&(d.timerProcessing=setTimeout(function(){b.isProcessing(c,!0)},500))}),c.hasInitialized=!0,c.isProcessing=!1,d.debug&&(console.log("Overall initialization time: "+b.benchmark(a.data(c,"startoveralltimer"))),d.debug&&console.groupEnd&&console.groupEnd()),f.triggerHandler("tablesorter-initialized",c),"function"==typeof d.initialized&&d.initialized(c)},bindMethods:function(c){var d=c.$table,e=c.namespace,f="sortReset update updateRows updateAll updateHeaders addRows updateCell updateComplete sorton appendCache updateCache applyWidgetId applyWidgets refreshWidgets destroy mouseup mouseleave ".split(" ").join(e+" ");d.unbind(f.replace(b.regex.spaces," ")).bind("sortReset"+e,function(a,c){a.stopPropagation(),b.sortReset(this.config,c)}).bind("updateAll"+e,function(a,c,d){a.stopPropagation(),b.updateAll(this.config,c,d)}).bind("update"+e+" updateRows"+e,function(a,c,d){a.stopPropagation(),b.update(this.config,c,d)}).bind("updateHeaders"+e,function(a,c){a.stopPropagation(),b.updateHeaders(this.config,c)}).bind("updateCell"+e,function(a,c,d,e){a.stopPropagation(),b.updateCell(this.config,c,d,e)}).bind("addRows"+e,function(a,c,d,e){a.stopPropagation(),b.addRows(this.config,c,d,e)}).bind("updateComplete"+e,function(){this.isUpdating=!1}).bind("sorton"+e,function(a,c,d,e){a.stopPropagation(),b.sortOn(this.config,c,d,e)}).bind("appendCache"+e,function(c,d,e){c.stopPropagation(),b.appendCache(this.config,e),a.isFunction(d)&&d(this)}).bind("updateCache"+e,function(a,c,d){a.stopPropagation(),b.updateCache(this.config,c,d)}).bind("applyWidgetId"+e,function(a,c){a.stopPropagation(),b.applyWidgetId(this,c)}).bind("applyWidgets"+e,function(a,c){a.stopPropagation(),b.applyWidget(this,c)}).bind("refreshWidgets"+e,function(a,c,d){a.stopPropagation(),b.refreshWidgets(this,c,d)}).bind("removeWidget"+e,function(a,c,d){a.stopPropagation(),b.removeWidget(this,c,d)}).bind("destroy"+e,function(a,c,d){a.stopPropagation(),b.destroy(this,c,d)}).bind("resetToLoadState"+e,function(d){d.stopPropagation(),b.removeWidget(this,!0,!1),c=a.extend(!0,b.defaults,c.originalSettings),this.hasInitialized=!1,b.setup(this,c)})},bindEvents:function(c,d,e){c=a(c)[0];var f,g=c.config,h=g.namespace,i=null;e!==!0&&(d.addClass(h.slice(1)+"_extra_headers"),f=a.fn.closest?d.closest("table")[0]:d.parents("table")[0],f&&"TABLE"===f.nodeName&&f!==c&&a(f).addClass(h.slice(1)+"_extra_table")),f=(g.pointerDown+" "+g.pointerUp+" "+g.pointerClick+" sort keyup ").replace(b.regex.spaces," ").split(" ").join(h+" "),d.find(g.selectorSort).add(d.filter(g.selectorSort)).unbind(f).bind(f,function(c,e){var f,h,j,k=a(c.target),l=" "+c.type+" ";if(!(1!==(c.which||c.button)&&!l.match(" "+g.pointerClick+" | sort | keyup ")||" keyup "===l&&c.which!==b.keyCodes.enter||l.match(" "+g.pointerClick+" ")&&"undefined"!=typeof c.which||l.match(" "+g.pointerUp+" ")&&i!==c.target&&e!==!0)){if(l.match(" "+g.pointerDown+" "))return i=c.target,j=k.jquery.split("."),void("1"===j[0]&&j[1]<4&&c.preventDefault());if(i=null,b.regex.formElements.test(c.target.nodeName)||k.hasClass(g.cssNoSort)||k.parents("."+g.cssNoSort).length>0||k.parents("button").length>0)return!g.cancelSelection;g.delayInit&&b.isEmptyObject(g.cache)&&b.buildCache(g),f=a.fn.closest?a(this).closest("th, td"):/TH|TD/.test(this.nodeName)?a(this):a(this).parents("th, td"),j=d.index(f),g.last.clickedIndex=0>j?f.attr("data-column"):j,h=g.$headers[g.last.clickedIndex],h&&!h.sortDisabled&&b.initSort(g,h,c)}}),g.cancelSelection&&d.attr("unselectable","on").bind("selectstart",!1).css({"user-select":"none",MozUserSelect:"none"})},buildHeaders:function(c){var d,e,f,g;for(c.headerList=[],c.headerContent=[],c.sortVars=[],c.debug&&(f=new Date),c.columns=b.computeColumnIndex(c.$table.children("thead, tfoot").children("tr")),e=c.cssIcon?'<i class="'+(c.cssIcon===b.css.icon?b.css.icon:c.cssIcon+" "+b.css.icon)+'"></i>':"",c.$headers=a(a.map(c.$table.find(c.selectorHeaders),function(d,f){var g,h,i,j,k,l=a(d);if(!l.parent().hasClass(c.cssIgnoreRow))return g=b.getColumnData(c.table,c.headers,f,!0),c.headerContent[f]=l.html(),""===c.headerTemplate||l.find("."+b.css.headerIn).length||(j=c.headerTemplate.replace(b.regex.templateContent,l.html()).replace(b.regex.templateIcon,l.find("."+b.css.icon).length?"":e),c.onRenderTemplate&&(h=c.onRenderTemplate.apply(l,[f,j]),h&&"string"==typeof h&&(j=h)),l.html('<div class="'+b.css.headerIn+'">'+j+"</div>")),c.onRenderHeader&&c.onRenderHeader.apply(l,[f,c,c.$table]),i=parseInt(l.attr("data-column"),10),d.column=i,k=b.getData(l,g,"sortInitialOrder")||c.sortInitialOrder,c.sortVars[i]={count:-1,order:b.getOrder(k)?[1,0,2]:[0,1,2],lockedOrder:!1},k=b.getData(l,g,"lockedOrder")||!1,"undefined"!=typeof k&&k!==!1&&(c.sortVars[i].lockedOrder=!0,c.sortVars[i].order=b.getOrder(k)?[1,1,1]:[0,0,0]),c.headerList[f]=d,l.addClass(b.css.header+" "+c.cssHeader).parent().addClass(b.css.headerRow+" "+c.cssHeaderRow).attr("role","row"),c.tabIndex&&l.attr("tabindex",0),d})),c.$headerIndexed=[],g=0;g<c.columns;g++)b.isEmptyObject(c.sortVars[g])&&(c.sortVars[g]={}),d=c.$headers.filter('[data-column="'+g+'"]'),c.$headerIndexed[g]=d.length?d.not(".sorter-false").length?d.not(".sorter-false").filter(":last"):d.filter(":last"):a();c.$table.find(c.selectorHeaders).attr({scope:"col",role:"columnheader"}),b.updateHeader(c),c.debug&&(console.log("Built headers:"+b.benchmark(f)),console.log(c.$headers))},addInstanceMethods:function(c){a.extend(b.instanceMethods,c)},setupParsers:function(a,c){var d,e,f,g,h,i,j,k,l,m,n,o,p,q,r=a.table,s=0,t={};if(a.$tbodies=a.$table.children("tbody:not(."+a.cssInfoBlock+")"),p="undefined"==typeof c?a.$tbodies:c,q=p.length,0===q)return a.debug?console.warn("Warning: *Empty table!* Not building a parser cache"):"";for(a.debug&&(o=new Date,console[console.group?"group":"log"]("Detecting parsers for each column")),e={extractors:[],parsers:[]};q>s;){if(d=p[s].rows,d.length)for(h=0,g=a.columns,i=0;g>i;i++){if(j=a.$headerIndexed[h],j&&j.length&&(k=b.getColumnData(r,a.headers,h),n=b.getParserById(b.getData(j,k,"extractor")),m=b.getParserById(b.getData(j,k,"sorter")),l="false"===b.getData(j,k,"parser"),a.empties[h]=(b.getData(j,k,"empty")||a.emptyTo||(a.emptyToBottom?"bottom":"top")).toLowerCase(),a.strings[h]=(b.getData(j,k,"string")||a.stringTo||"max").toLowerCase(),l&&(m=b.getParserById("no-parser")),n||(n=!1),m||(m=b.detectParserForColumn(a,d,-1,h)),a.debug&&(t["("+h+") "+j.text()]={parser:m.id,extractor:n?n.id:"none",string:a.strings[h],empty:a.empties[h]}),e.parsers[h]=m,e.extractors[h]=n,f=j[0].colSpan-1,f>0))for(h+=f,g+=f;f+1>0;)e.parsers[h-f]=m,e.extractors[h-f]=n,f--;h++}s+=e.parsers.length?q:1}a.debug&&(b.isEmptyObject(t)?console.warn("  No parsers detected!"):console[console.table?"table":"log"](t),console.log("Completed detecting parsers"+b.benchmark(o)),console.groupEnd&&console.groupEnd()),a.parsers=e.parsers,a.extractors=e.extractors},addParser:function(a){var c,d=b.parsers.length,e=!0;for(c=0;d>c;c++)b.parsers[c].id.toLowerCase()===a.id.toLowerCase()&&(e=!1);e&&(b.parsers[b.parsers.length]=a)},getParserById:function(a){if("false"==a)return!1;var c,d=b.parsers.length;for(c=0;d>c;c++)if(b.parsers[c].id.toLowerCase()===a.toString().toLowerCase())return b.parsers[c];return!1},detectParserForColumn:function(c,d,e,f){for(var g,h,i,j=b.parsers.length,k=!1,l="",m=!0;""===l&&m;)e++,i=d[e],i&&50>e?i.className.indexOf(b.cssIgnoreRow)<0&&(k=d[e].cells[f],l=b.getElementText(c,k,f),h=a(k),c.debug&&console.log("Checking if value was empty on row "+e+", column: "+f+': "'+l+'"')):m=!1;for(;--j>=0;)if(g=b.parsers[j],g&&"text"!==g.id&&g.is&&g.is(l,c.table,k,h))return g;return b.getParserById("text")},getElementText:function(c,d,e){if(!d)return"";var f,g=c.textExtraction||"",h=d.jquery?d:a(d);return"string"==typeof g?"basic"===g&&"undefined"!=typeof(f=h.attr(c.textAttribute))?a.trim(f):a.trim(d.textContent||h.text()):"function"==typeof g?a.trim(g(h[0],c.table,e)):"function"==typeof(f=b.getColumnData(c.table,g,e))?a.trim(f(h[0],c.table,e)):a.trim(h[0].textContent||h.text())},getParsedText:function(a,c,d,e){"undefined"==typeof e&&(e=b.getElementText(a,c,d));var f=""+e,g=a.parsers[d],h=a.extractors[d];return g&&(h&&"function"==typeof h.format&&(e=h.format(e,a.table,c,d)),f="no-parser"===g.id?"":g.format(""+e,a.table,c,d),a.ignoreCase&&"string"==typeof f&&(f=f.toLowerCase())),f},buildCache:function(c,d,e){var f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,A,B=c.table,C=c.parsers;if(c.$tbodies=c.$table.children("tbody:not(."+c.cssInfoBlock+")"),l="undefined"==typeof e?c.$tbodies:e,c.cache={},c.totalRows=0,!C)return c.debug?console.warn("Warning: *Empty table!* Not building a cache"):"";for(c.debug&&(q=new Date),c.showProcessing&&b.isProcessing(B,!0),k=0;k<l.length;k++){for(u=[],f=c.cache[k]={normalized:[]},r=l[k]&&l[k].rows.length||0,i=0;r>i;++i)if(s={child:[],raw:[]},m=a(l[k].rows[i]),n=[],m.hasClass(c.cssChildRow)&&0!==i)for(z=f.normalized.length-1,t=f.normalized[z][c.columns],t.$row=t.$row.add(m),m.prev().hasClass(c.cssChildRow)||m.prev().addClass(b.css.cssHasChild),o=m.children("th, td"),z=t.child.length,t.child[z]=[],w=0,y=c.columns,j=0;y>j;j++)p=o[j],p&&(t.child[z][j]=b.getParsedText(c,p,j),v=o[j].colSpan-1,v>0&&(w+=v,y+=v)),w++;else{for(s.$row=m,s.order=i,w=0,y=c.columns,j=0;y>j;++j){if(p=m[0].cells[j],p&&w<c.columns&&(x="undefined"!=typeof C[w],!x&&c.debug&&console.warn("No parser found for row: "+i+", column: "+j+'; cell containing: "'+a(p).text()+'"; does it have a header?'),g=b.getElementText(c,p,w),s.raw[w]=g,h=b.getParsedText(c,p,w,g),n[w]=h,x&&"numeric"===(C[w].type||"").toLowerCase()&&(u[w]=Math.max(Math.abs(h)||0,u[w]||0)),v=p.colSpan-1,v>0)){for(A=0;v>=A;)s.raw[w+A]=c.duplicateSpan||0===A?g:"",n[w+A]=c.duplicateSpan||0===A?g:"",A++;w+=v,y+=v}w++}n[c.columns]=s,f.normalized[f.normalized.length]=n}f.colMax=u,c.totalRows+=f.normalized.length}if(c.showProcessing&&b.isProcessing(B),c.debug){for(z=Math.min(5,c.cache[0].normalized.length),console[console.group?"group":"log"]("Building cache for "+c.totalRows+" rows (showing "+z+" rows in log)"+b.benchmark(q)),g={},j=0;j<c.columns;j++)for(w=0;z>w;w++)g["row: "+w]||(g["row: "+w]={}),g["row: "+w][c.$headerIndexed[j].text()]=c.cache[0].normalized[w][j];console[console.table?"table":"log"](g),console.groupEnd&&console.groupEnd()}a.isFunction(d)&&d(B)},getColumnText:function(c,d,e,f){c=a(c)[0];var g,h,i,j,k,l,m,n,o,p,q="function"==typeof e,r="all"===d,s={raw:[],parsed:[],$cell:[]},t=c.config;if(!b.isEmptyObject(t)){for(k=t.$tbodies.length,g=0;k>g;g++)for(i=t.cache[g].normalized,l=i.length,h=0;l>h;h++)j=i[h],f&&!j[t.columns].$row.is(f)||(p=!0,n=r?j.slice(0,t.columns):j[d],j=j[t.columns],m=r?j.raw:j.raw[d],o=r?j.$row.children():j.$row.children().eq(d),q&&(p=e({tbodyIndex:g,rowIndex:h,parsed:n,raw:m,$row:j.$row,$cell:o})),p!==!1&&(s.parsed[s.parsed.length]=n,s.raw[s.raw.length]=m,s.$cell[s.$cell.length]=o));return s}t.debug&&console.warn("No cache found - aborting getColumnText function!")},setHeadersCss:function(c){var d,e,f,g=c.sortList,h=g.length,i=b.css.sortNone+" "+c.cssNone,j=[b.css.sortAsc+" "+c.cssAsc,b.css.sortDesc+" "+c.cssDesc],k=[c.cssIconAsc,c.cssIconDesc,c.cssIconNone],l=["ascending","descending"],m=c.$table.find("tfoot tr").children("td, th").add(a(c.namespace+"_extra_headers")).removeClass(j.join(" "));for(c.$headers.removeClass(j.join(" ")).addClass(i).attr("aria-sort","none").find("."+b.css.icon).removeClass(k.join(" ")).addClass(k[2]),e=0;h>e;e++)if(2!==g[e][1]&&(d=c.$headers.filter(function(a){for(var d=!0,e=c.$headers.eq(a),f=parseInt(e.attr("data-column"),10),g=f+c.$headers[a].colSpan;g>f;f++)d=d?d||b.isValueInArray(f,c.sortList)>-1:!1;return d}),d=d.not(".sorter-false").filter('[data-column="'+g[e][0]+'"]'+(1===h?":last":"")),d.length)){for(f=0;f<d.length;f++)d[f].sortDisabled||d.eq(f).removeClass(i).addClass(j[g[e][1]]).attr("aria-sort",l[g[e][1]]).find("."+b.css.icon).removeClass(k[2]).addClass(k[g[e][1]]);m.length&&m.filter('[data-column="'+g[e][0]+'"]').removeClass(i).addClass(j[g[e][1]])}for(h=c.$headers.length,e=0;h>e;e++)b.setColumnAriaLabel(c,c.$headers.eq(e))},setColumnAriaLabel:function(c,d,e){if(d.length){var f=parseInt(d.attr("data-column"),10),g=d.hasClass(b.css.sortAsc)?"sortAsc":d.hasClass(b.css.sortDesc)?"sortDesc":"sortNone",h=a.trim(d.text())+": "+b.language[g];d.hasClass("sorter-false")||e===!1?h+=b.language.sortDisabled:(e=c.sortVars[f].order[(c.sortVars[f].count+1)%(c.sortReset?3:2)],h+=b.language[0===e?"nextAsc":1===e?"nextDesc":"nextNone"]),d.attr("aria-label",h)}},updateHeader:function(a){var c,d,e,f,g=a.table,h=a.$headers.length;for(c=0;h>c;c++)e=a.$headers.eq(c),f=b.getColumnData(g,a.headers,c,!0),d="false"===b.getData(e,f,"sorter")||"false"===b.getData(e,f,"parser"),b.setColumnSort(a,e,d)},setColumnSort:function(a,b,c){var d=a.table.id;b[0].sortDisabled=c,b[c?"addClass":"removeClass"]("sorter-false").attr("aria-disabled",""+c),a.tabIndex&&(c?b.removeAttr("tabindex"):b.attr("tabindex","0")),d&&(c?b.removeAttr("aria-controls"):b.attr("aria-controls",d))},updateHeaderSortCount:function(c,d){var e,f,g,h,i,j,k,l,m=d||c.sortList,n=m.length;for(c.sortList=[],h=0;n>h;h++)if(k=m[h],e=parseInt(k[0],10),e<c.columns){switch(c.sortVars[e].order||(l=c.sortVars[e].order=b.getOrder(c.sortInitialOrder)?[1,0,2]:[0,1,2],c.sortVars[e].count=0),l=c.sortVars[e].order,f=(""+k[1]).match(/^(1|d|s|o|n)/),f=f?f[0]:""){case"1":case"d":f=1;break;case"s":f=i||0;break;case"o":j=l[(i||0)%(c.sortReset?3:2)],f=0===j?1:1===j?0:2;break;case"n":f=l[++c.sortVars[e].count%(c.sortReset?3:2)];break;default:f=0}i=0===h?f:i,g=[e,parseInt(f,10)||0],c.sortList[c.sortList.length]=g,f=a.inArray(g[1],l),c.sortVars[e].count=f>=0?f:g[1]%(c.sortReset?3:2)}},updateAll:function(a,c,d){var e=a.table;e.isUpdating=!0,b.refreshWidgets(e,!0,!0),b.buildHeaders(a),b.bindEvents(e,a.$headers,!0),b.bindMethods(a),b.commonUpdate(a,c,d)},update:function(a,c,d){var e=a.table;e.isUpdating=!0,b.updateHeader(a),b.commonUpdate(a,c,d)},updateHeaders:function(a,c){a.table.isUpdating=!0,b.buildHeaders(a),b.bindEvents(a.table,a.$headers,!0),b.resortComplete(a,c)},updateCell:function(c,d,e,f){if(b.isEmptyObject(c.cache))return b.updateHeader(c),void b.commonUpdate(c,e,f);c.table.isUpdating=!0,c.$table.find(c.selectorRemove).remove();var g,h,i,j,k,l,m=c.$tbodies,n=a(d),o=m.index(a.fn.closest?n.closest("tbody"):n.parents("tbody").filter(":first")),p=c.cache[o],q=a.fn.closest?n.closest("tr"):n.parents("tr").filter(":first");if(d=n[0],m.length&&o>=0){if(i=m.eq(o).find("tr").index(q),k=p.normalized[i],l=q[0].cells.length,l!==c.columns)for(j=0,g=!1,h=0;l>h;h++)g||q[0].cells[h]===d?g=!0:j+=q[0].cells[h].colSpan;else j=n.index();g=b.getElementText(c,d,j),k[c.columns].raw[j]=g,g=b.getParsedText(c,d,j,g),k[j]=g,k[c.columns].$row=q,"numeric"===(c.parsers[j].type||"").toLowerCase()&&(p.colMax[j]=Math.max(Math.abs(g)||0,p.colMax[j]||0)),g="undefined"!==e?e:c.resort,g!==!1?b.checkResort(c,g,f):b.resortComplete(c,f)}else c.debug&&console.error("updateCell aborted, tbody missing or not within the indicated table"),c.table.isUpdating=!1},addRows:function(c,d,e,f){var g,h,i,j,k,l,m,n,o,p,q,r,s,t="string"==typeof d&&1===c.$tbodies.length&&/<tr/.test(d||""),u=c.table;if(t)d=a(d),c.$tbodies.append(d);else if(!(d&&d instanceof jQuery&&(a.fn.closest?d.closest("table")[0]:d.parents("table")[0])===c.table))return c.debug&&console.error("addRows method requires (1) a jQuery selector reference to rows that have already been added to the table, or (2) row HTML string to be added to a table with only one tbody"),!1;if(u.isUpdating=!0,b.isEmptyObject(c.cache))b.updateHeader(c),b.commonUpdate(c,e,f);else{for(k=d.filter("tr").attr("role","row").length,i=c.$tbodies.index(d.parents("tbody").filter(":first")),c.parsers&&c.parsers.length||b.setupParsers(c),j=0;k>j;j++){for(o=0,m=d[j].cells.length,n=c.cache[i].normalized.length,q=[],p={child:[],raw:[],$row:d.eq(j),order:n},l=0;m>l;l++)r=d[j].cells[l],g=b.getElementText(c,r,o),p.raw[o]=g,h=b.getParsedText(c,r,o,g),q[o]=h,"numeric"===(c.parsers[o].type||"").toLowerCase()&&(c.cache[i].colMax[o]=Math.max(Math.abs(h)||0,c.cache[i].colMax[o]||0)),s=r.colSpan-1,s>0&&(o+=s),o++;q[c.columns]=p,c.cache[i].normalized[n]=q}b.checkResort(c,e,f)}},updateCache:function(a,c,d){a.parsers&&a.parsers.length||b.setupParsers(a,d),b.buildCache(a,c,d)},appendCache:function(a,c){var d,e,f,g,h,i,j,k=a.table,l=a.widgetOptions,m=a.$tbodies,n=[],o=a.cache;if(b.isEmptyObject(o))return a.appender?a.appender(k,n):k.isUpdating?a.$table.triggerHandler("updateComplete",k):"";for(a.debug&&(j=new Date),i=0;i<m.length;i++)if(f=m.eq(i),f.length){for(g=b.processTbody(k,f,!0),d=o[i].normalized,e=d.length,h=0;e>h;h++)n[n.length]=d[h][a.columns].$row,a.appender&&(!a.pager||a.pager.removeRows&&l.pager_removeRows||a.pager.ajax)||g.append(d[h][a.columns].$row);b.processTbody(k,g,!1)}a.appender&&a.appender(k,n),a.debug&&console.log("Rebuilt table"+b.benchmark(j)),c||a.appender||b.applyWidget(k),k.isUpdating&&a.$table.triggerHandler("updateComplete",k)},commonUpdate:function(a,c,d){a.$table.find(a.selectorRemove).remove(),b.setupParsers(a),b.buildCache(a),b.checkResort(a,c,d)},initSort:function(c,d,e){if(c.table.isUpdating)return setTimeout(function(){b.initSort(c,d,e)},50);var f,g,h,i,j,k,l,m=!e[c.sortMultiSortKey],n=c.table,o=c.$headers.length,p=parseInt(a(d).attr("data-column"),10),q=c.sortVars[p].order;if(c.$table.triggerHandler("sortStart",n),c.sortVars[p].count=e[c.sortResetKey]?2:(c.sortVars[p].count+1)%(c.sortReset?3:2),c.sortRestart)for(h=0;o>h;h++)l=c.$headers.eq(h),k=parseInt(l.attr("data-column"),10),p!==k&&(m||l.hasClass(b.css.sortNone))&&(c.sortVars[k].count=-1);if(m){if(c.sortList=[],c.last.sortList=[],null!==c.sortForce)for(f=c.sortForce,g=0;g<f.length;g++)f[g][0]!==p&&(c.sortList[c.sortList.length]=f[g]);if(i=q[c.sortVars[p].count],2>i&&(c.sortList[c.sortList.length]=[p,i],d.colSpan>1))for(g=1;g<d.colSpan;g++)c.sortList[c.sortList.length]=[p+g,i],c.sortVars[p+g].count=a.inArray(i,q)}else if(c.sortList=a.extend([],c.last.sortList),b.isValueInArray(p,c.sortList)>=0)for(g=0;g<c.sortList.length;g++)k=c.sortList[g],k[0]===p&&(k[1]=q[c.sortVars[p].count],2===k[1]&&(c.sortList.splice(g,1),c.sortVars[p].count=-1));else if(i=q[c.sortVars[p].count],2>i&&(c.sortList[c.sortList.length]=[p,i],d.colSpan>1))for(g=1;g<d.colSpan;g++)c.sortList[c.sortList.length]=[p+g,i],c.sortVars[p+g].count=a.inArray(i,q);if(c.last.sortList=a.extend([],c.sortList),c.sortList.length&&c.sortAppend&&(f=a.isArray(c.sortAppend)?c.sortAppend:c.sortAppend[c.sortList[0][0]],!b.isEmptyObject(f)))for(g=0;g<f.length;g++)if(f[g][0]!==p&&b.isValueInArray(f[g][0],c.sortList)<0){if(i=f[g][1],j=(""+i).match(/^(a|d|s|o|n)/))switch(k=c.sortList[0][1],j[0]){case"d":i=1;break;case"s":i=k;break;case"o":i=0===k?1:0;break;case"n":i=(k+1)%(c.sortReset?3:2);break;default:i=0}c.sortList[c.sortList.length]=[f[g][0],i]}c.$table.triggerHandler("sortBegin",n),setTimeout(function(){b.setHeadersCss(c),b.multisort(c),b.appendCache(c),c.$table.triggerHandler("sortBeforeEnd",n),c.$table.triggerHandler("sortEnd",n)},1)},multisort:function(a){var c,d,e,f,g=a.table,h=0,i=a.textSorter||"",j=a.sortList,k=j.length,l=a.$tbodies.length;if(!a.serverSideSorting&&!b.isEmptyObject(a.cache)){for(a.debug&&(d=new Date),c=0;l>c;c++)e=a.cache[c].colMax,f=a.cache[c].normalized,f.sort(function(c,d){var f,l,m,n,o,p,q;for(f=0;k>f;f++){if(m=j[f][0],n=j[f][1],h=0===n,a.sortStable&&c[m]===d[m]&&1===k)return c[a.columns].order-d[a.columns].order;if(l=/n/i.test(b.getSortType(a.parsers,m)),l&&a.strings[m]?(l="boolean"==typeof b.string[a.strings[m]]?(h?1:-1)*(b.string[a.strings[m]]?-1:1):a.strings[m]?b.string[a.strings[m]]||0:0,o=a.numberSorter?a.numberSorter(c[m],d[m],h,e[m],g):b["sortNumeric"+(h?"Asc":"Desc")](c[m],d[m],l,e[m],m,a)):(p=h?c:d,q=h?d:c,o="function"==typeof i?i(p[m],q[m],h,m,g):"object"==typeof i&&i.hasOwnProperty(m)?i[m](p[m],q[m],h,m,g):b["sortNatural"+(h?"Asc":"Desc")](c[m],d[m],m,a)),o)return o}return c[a.columns].order-d[a.columns].order});a.debug&&console.log("Applying sort "+j.toString()+b.benchmark(d))}},resortComplete:function(b,c){b.table.isUpdating&&b.$table.triggerHandler("updateComplete",b.table),a.isFunction(c)&&c(b.table)},checkResort:function(c,d,e){var f=a.isArray(d)?d:c.sortList,g="undefined"==typeof d?c.resort:d;g===!1||c.serverSideSorting||c.table.isProcessing?(b.resortComplete(c,e),b.applyWidget(c.table,!1)):f.length?b.sortOn(c,f,function(){b.resortComplete(c,e)},!0):b.sortReset(c,function(){b.resortComplete(c,e),b.applyWidget(c.table,!1)})},sortOn:function(c,d,e,f){var g=c.table;c.$table.triggerHandler("sortStart",g),b.updateHeaderSortCount(c,d),b.setHeadersCss(c),c.delayInit&&b.isEmptyObject(c.cache)&&b.buildCache(c),c.$table.triggerHandler("sortBegin",g),b.multisort(c),b.appendCache(c,f),c.$table.triggerHandler("sortBeforeEnd",g),c.$table.triggerHandler("sortEnd",g),b.applyWidget(g),a.isFunction(e)&&e(g)},sortReset:function(c,d){c.sortList=[],b.setHeadersCss(c),b.multisort(c),b.appendCache(c),a.isFunction(d)&&d(c.table)},getSortType:function(a,b){return a&&a[b]?a[b].type||"":""},getOrder:function(a){return/^d/i.test(a)||1===a},sortNatural:function(a,c){if(a===c)return 0;var d,e,f,g,h,i,j=b.regex;if(j.hex.test(c)){if(d=parseInt(a.match(j.hex),16),e=parseInt(c.match(j.hex),16),e>d)return-1;if(d>e)return 1}for(d=a.replace(j.chunk,"\\0$1\\0").replace(j.chunks,"").split("\\0"),e=c.replace(j.chunk,"\\0$1\\0").replace(j.chunks,"").split("\\0"),i=Math.max(d.length,e.length),h=0;i>h;h++){if(f=isNaN(d[h])?d[h]||0:parseFloat(d[h])||0,g=isNaN(e[h])?e[h]||0:parseFloat(e[h])||0,isNaN(f)!==isNaN(g))return isNaN(f)?1:-1;if(typeof f!=typeof g&&(f+="",g+=""),g>f)return-1;if(f>g)return 1}return 0},sortNaturalAsc:function(a,c,d,e){if(a===c)return 0;var f=b.string[e.empties[d]||e.emptyTo];return""===a&&0!==f?"boolean"==typeof f?f?-1:1:-f||-1:""===c&&0!==f?"boolean"==typeof f?f?1:-1:f||1:b.sortNatural(a,c)},sortNaturalDesc:function(a,c,d,e){if(a===c)return 0;var f=b.string[e.empties[d]||e.emptyTo];return""===a&&0!==f?"boolean"==typeof f?f?-1:1:f||1:""===c&&0!==f?"boolean"==typeof f?f?1:-1:-f||-1:b.sortNatural(c,a)},sortText:function(a,b){return a>b?1:b>a?-1:0},getTextValue:function(a,b,c){if(c){var d,e=a?a.length:0,f=c+b;for(d=0;e>d;d++)f+=a.charCodeAt(d);return b*f}return 0},sortNumericAsc:function(a,c,d,e,f,g){if(a===c)return 0;var h=b.string[g.empties[f]||g.emptyTo];return""===a&&0!==h?"boolean"==typeof h?h?-1:1:-h||-1:""===c&&0!==h?"boolean"==typeof h?h?1:-1:h||1:(isNaN(a)&&(a=b.getTextValue(a,d,e)),isNaN(c)&&(c=b.getTextValue(c,d,e)),a-c)},sortNumericDesc:function(a,c,d,e,f,g){if(a===c)return 0;var h=b.string[g.empties[f]||g.emptyTo];return""===a&&0!==h?"boolean"==typeof h?h?-1:1:h||1:""===c&&0!==h?"boolean"==typeof h?h?1:-1:-h||-1:(isNaN(a)&&(a=b.getTextValue(a,d,e)),isNaN(c)&&(c=b.getTextValue(c,d,e)),c-a)},sortNumeric:function(a,b){return a-b},addWidget:function(a){a.id&&!b.isEmptyObject(b.getWidgetById(a.id))&&console.warn('"'+a.id+'" widget was loaded more than once!'),b.widgets[b.widgets.length]=a},hasWidget:function(b,c){return b=a(b),b.length&&b[0].config&&b[0].config.widgetInit[c]||!1},getWidgetById:function(a){var c,d,e=b.widgets.length;for(c=0;e>c;c++)if(d=b.widgets[c],d&&d.id&&d.id.toLowerCase()===a.toLowerCase())return d},applyWidgetOptions:function(c){var d,e,f=c.config,g=f.widgets.length;if(g)for(d=0;g>d;d++)e=b.getWidgetById(f.widgets[d]),e&&e.options&&(f.widgetOptions=a.extend(!0,{},e.options,f.widgetOptions))},addWidgetFromClass:function(a){var c,d,e=a.config,f="^"+e.widgetClass.replace(b.regex.templateName,"(\\S+)+")+"$",g=new RegExp(f,"g"),h=(a.className||"").split(b.regex.spaces);if(h.length)for(c=h.length,d=0;c>d;d++)h[d].match(g)&&(e.widgets[e.widgets.length]=h[d].replace(g,"$1"))},applyWidgetId:function(c,d,e){c=a(c)[0];var f,g,h,i=c.config,j=i.widgetOptions,k=b.getWidgetById(d);k&&(h=k.id,f=!1,a.inArray(h,i.widgets)<0&&(i.widgets[i.widgets.length]=h),i.debug&&(g=new Date),!e&&i.widgetInit[h]||(i.widgetInit[h]=!0,c.hasInitialized&&b.applyWidgetOptions(c),"function"==typeof k.init&&(f=!0,i.debug&&console[console.group?"group":"log"]("Initializing "+h+" widget"),k.init(c,k,i,j))),e||"function"!=typeof k.format||(f=!0,i.debug&&console[console.group?"group":"log"]("Updating "+h+" widget"),k.format(c,i,j,!1)),i.debug&&f&&(console.log("Completed "+(e?"initializing ":"applying ")+h+" widget"+b.benchmark(g)),console.groupEnd&&console.groupEnd()))},applyWidget:function(c,d,e){c=a(c)[0];var f,g,h,i,j,k=c.config,l=[];if(d===!1||!c.hasInitialized||!c.isApplyingWidgets&&!c.isUpdating){if(k.debug&&(j=new Date),b.addWidgetFromClass(c),clearTimeout(k.timerReady),k.widgets.length){for(c.isApplyingWidgets=!0,k.widgets=a.grep(k.widgets,function(b,c){return a.inArray(b,k.widgets)===c}),h=k.widgets||[],g=h.length,f=0;g>f;f++)i=b.getWidgetById(h[f]),i&&i.id?(i.priority||(i.priority=10),l[f]=i):k.debug&&console.warn('"'+h[f]+'" widget code does not exist!');for(l.sort(function(a,b){return a.priority<b.priority?-1:a.priority===b.priority?0:1}),g=l.length,k.debug&&console[console.group?"group":"log"]("Start "+(d?"initializing":"applying")+" widgets"),f=0;g>f;f++)i=l[f],i&&i.id&&b.applyWidgetId(c,i.id,d);k.debug&&console.groupEnd&&console.groupEnd(),d||"function"!=typeof e||e(c)}k.timerReady=setTimeout(function(){c.isApplyingWidgets=!1,a.data(c,"lastWidgetApplication",new Date),k.$table.triggerHandler("tablesorter-ready")},10),k.debug&&(i=k.widgets.length,console.log("Completed "+(d===!0?"initializing ":"applying ")+i+" widget"+(1!==i?"s":"")+b.benchmark(j)))}},removeWidget:function(c,d,e){c=a(c)[0];var f,g,h,i,j=c.config;if(d===!0)for(d=[],i=b.widgets.length,h=0;i>h;h++)g=b.widgets[h],g&&g.id&&(d[d.length]=g.id);else d=(a.isArray(d)?d.join(","):d||"").toLowerCase().split(/[\s,]+/);for(i=d.length,f=0;i>f;f++)g=b.getWidgetById(d[f]),h=a.inArray(d[f],j.widgets),h>=0&&e!==!0&&j.widgets.splice(h,1),g&&g.remove&&(j.debug&&console.log((e?"Refreshing":"Removing")+' "'+d[f]+'" widget'),g.remove(c,j,j.widgetOptions,e),j.widgetInit[d[f]]=!1)},refreshWidgets:function(c,d,e){c=a(c)[0];var f,g,h=c.config,i=h.widgets,j=b.widgets,k=j.length,l=[],m=function(b){a(b).triggerHandler("refreshComplete")};for(f=0;k>f;f++)g=j[f],g&&g.id&&(d||a.inArray(g.id,i)<0)&&(l[l.length]=g.id);b.removeWidget(c,l.join(","),!0),e!==!0?(b.applyWidget(c,d||!1,m),d&&b.applyWidget(c,!1,m)):m(c)},benchmark:function(a){return" ( "+((new Date).getTime()-a.getTime())+"ms )"},log:function(){console.log(arguments)},isEmptyObject:function(a){for(var b in a)return!1;return!0},isValueInArray:function(a,b){var c,d=b&&b.length||0;for(c=0;d>c;c++)if(b[c][0]===a)return c;return-1},formatFloat:function(c,d){if("string"!=typeof c||""===c)return c;var e,f=d&&d.config?d.config.usNumberFormat!==!1:"undefined"!=typeof d?d:!0;return c=f?c.replace(b.regex.comma,""):c.replace(b.regex.digitNonUS,"").replace(b.regex.comma,"."),b.regex.digitNegativeTest.test(c)&&(c=c.replace(b.regex.digitNegativeReplace,"-$1")),e=parseFloat(c),isNaN(e)?a.trim(c):e},isDigit:function(a){return isNaN(a)?b.regex.digitTest.test(a.toString().replace(b.regex.digitReplace,"")):""!==a},computeColumnIndex:function(b,c){var d,e,f,g,h,i,j,k,l,m,n=c&&c.columns||0,o=[],p=new Array(n);
+for(d=0;d<b.length;d++)for(i=b[d].cells,e=0;e<i.length;e++){for(h=i[e],j=h.parentNode.rowIndex,k=h.rowSpan||1,l=h.colSpan||1,"undefined"==typeof o[j]&&(o[j]=[]),f=0;f<o[j].length+1;f++)if("undefined"==typeof o[j][f]){m=f;break}for(n&&h.cellIndex===m||(h.setAttribute?h.setAttribute("data-column",m):a(h).attr("data-column",m)),f=j;j+k>f;f++)for("undefined"==typeof o[f]&&(o[f]=[]),p=o[f],g=m;m+l>g;g++)p[g]="x"}return p.length},fixColumnWidth:function(c){c=a(c)[0];var d,e,f,g,h,i=c.config,j=i.$table.children("colgroup");if(j.length&&j.hasClass(b.css.colgroup)&&j.remove(),i.widthFixed&&0===i.$table.children("colgroup").length){for(j=a('<colgroup class="'+b.css.colgroup+'">'),d=i.$table.width(),f=i.$tbodies.find("tr:first").children(":visible"),g=f.length,h=0;g>h;h++)e=parseInt(f.eq(h).width()/d*1e3,10)/10+"%",j.append(a("<col>").css("width",e));i.$table.prepend(j)}},getData:function(b,c,d){var e,f,g="",h=a(b);return h.length?(e=a.metadata?h.metadata():!1,f=" "+(h.attr("class")||""),"undefined"!=typeof h.data(d)||"undefined"!=typeof h.data(d.toLowerCase())?g+=h.data(d)||h.data(d.toLowerCase()):e&&"undefined"!=typeof e[d]?g+=e[d]:c&&"undefined"!=typeof c[d]?g+=c[d]:" "!==f&&f.match(" "+d+"-")&&(g=f.match(new RegExp("\\s"+d+"-([\\w-]+)"))[1]||""),a.trim(g)):""},getColumnData:function(b,c,d,e,f){if("undefined"!=typeof c&&null!==c){b=a(b)[0];var g,h,i=b.config,j=f||i.$headers,k=i.$headerIndexed&&i.$headerIndexed[d]||j.filter('[data-column="'+d+'"]:last');if(c[d])return e?c[d]:c[j.index(k)];for(h in c)if("string"==typeof h&&(g=k.filter(h).add(k.find(h)),g.length))return c[h]}},isProcessing:function(c,d,e){c=a(c);var f=c[0].config,g=e||c.find("."+b.css.header);d?("undefined"!=typeof e&&f.sortList.length>0&&(g=g.filter(function(){return this.sortDisabled?!1:b.isValueInArray(parseFloat(a(this).attr("data-column")),f.sortList)>=0})),c.add(g).addClass(b.css.processing+" "+f.cssProcessing)):c.add(g).removeClass(b.css.processing+" "+f.cssProcessing)},processTbody:function(b,c,d){if(b=a(b)[0],d)return b.isProcessing=!0,c.before('<colgroup class="tablesorter-savemyplace"/>'),a.fn.detach?c.detach():c.remove();var e=a(b).find("colgroup.tablesorter-savemyplace");c.insertAfter(e),e.remove(),b.isProcessing=!1},clearTableBody:function(b){a(b)[0].config.$tbodies.children().detach()},characterEquivalents:{a:"áàâãäąå",A:"ÁÀÂÃÄĄÅ",c:"çćč",C:"ÇĆČ",e:"éèêëěę",E:"ÉÈÊËĚĘ",i:"íìİîïı",I:"ÍÌİÎÏ",o:"óòôõöō",O:"ÓÒÔÕÖŌ",ss:"ß",SS:"ẞ",u:"úùûüů",U:"ÚÙÛÜŮ"},replaceAccents:function(a){var c,d="[",e=b.characterEquivalents;if(!b.characterRegex){b.characterRegexArray={};for(c in e)"string"==typeof c&&(d+=e[c],b.characterRegexArray[c]=new RegExp("["+e[c]+"]","g"));b.characterRegex=new RegExp(d+"]")}if(b.characterRegex.test(a))for(c in e)"string"==typeof c&&(a=a.replace(b.characterRegexArray[c],c));return a},restoreHeaders:function(c){var d,e,f=a(c)[0].config,g=f.$table.find(f.selectorHeaders),h=g.length;for(d=0;h>d;d++)e=g.eq(d),e.find("."+b.css.headerIn).length&&e.html(f.headerContent[d])},destroy:function(c,d,e){if(c=a(c)[0],c.hasInitialized){b.removeWidget(c,!0,!1);var f,g=a(c),h=c.config,i=h.debug,j=g.find("thead:first"),k=j.find("tr."+b.css.headerRow).removeClass(b.css.headerRow+" "+h.cssHeaderRow),l=g.find("tfoot:first > tr").children("th, td");d===!1&&a.inArray("uitheme",h.widgets)>=0&&(g.triggerHandler("applyWidgetId",["uitheme"]),g.triggerHandler("applyWidgetId",["zebra"])),j.find("tr").not(k).remove(),f="sortReset update updateRows updateAll updateHeaders updateCell addRows updateComplete sorton appendCache updateCache applyWidgetId applyWidgets refreshWidgets removeWidget destroy mouseup mouseleave "+"keypress sortBegin sortEnd resetToLoadState ".split(" ").join(h.namespace+" "),g.removeData("tablesorter").unbind(f.replace(b.regex.spaces," ")),h.$headers.add(l).removeClass([b.css.header,h.cssHeader,h.cssAsc,h.cssDesc,b.css.sortAsc,b.css.sortDesc,b.css.sortNone].join(" ")).removeAttr("data-column").removeAttr("aria-label").attr("aria-disabled","true"),k.find(h.selectorSort).unbind("mousedown mouseup keypress ".split(" ").join(h.namespace+" ").replace(b.regex.spaces," ")),b.restoreHeaders(c),g.toggleClass(b.css.table+" "+h.tableClass+" tablesorter-"+h.theme,d===!1),c.hasInitialized=!1,delete c.config.cache,"function"==typeof e&&e(c),i&&console.log("tablesorter has been removed")}}};a.fn.tablesorter=function(c){return this.each(function(){var d=this,e=a.extend(!0,{},b.defaults,c,b.instanceMethods);e.originalSettings=c,!d.hasInitialized&&b.buildTable&&"TABLE"!==this.nodeName?b.buildTable(d,e):b.setup(d,e)})},window.console&&window.console.log||(b.logs=[],console={},console.log=console.warn=console.error=console.table=function(){var a=arguments.length>1?arguments:arguments[0];b.logs[b.logs.length]={date:Date.now(),log:a}}),b.addParser({id:"no-parser",is:function(){return!1},format:function(){return""},type:"text"}),b.addParser({id:"text",is:function(){return!0},format:function(c,d){var e=d.config;return c&&(c=a.trim(e.ignoreCase?c.toLocaleLowerCase():c),c=e.sortLocaleCompare?b.replaceAccents(c):c),c},type:"text"}),b.regex.nondigit=/[^\w,. \-()]/g,b.addParser({id:"digit",is:function(a){return b.isDigit(a)},format:function(c,d){var e=b.formatFloat((c||"").replace(b.regex.nondigit,""),d);return c&&"number"==typeof e?e:c?a.trim(c&&d.config.ignoreCase?c.toLocaleLowerCase():c):c},type:"numeric"}),b.regex.currencyReplace=/[+\-,. ]/g,b.regex.currencyTest=/^\(?\d+[\u00a3$\u20ac\u00a4\u00a5\u00a2?.]|[\u00a3$\u20ac\u00a4\u00a5\u00a2?.]\d+\)?$/,b.addParser({id:"currency",is:function(a){return a=(a||"").replace(b.regex.currencyReplace,""),b.regex.currencyTest.test(a)},format:function(c,d){var e=b.formatFloat((c||"").replace(b.regex.nondigit,""),d);return c&&"number"==typeof e?e:c?a.trim(c&&d.config.ignoreCase?c.toLocaleLowerCase():c):c},type:"numeric"}),b.regex.urlProtocolTest=/^(https?|ftp|file):\/\//,b.regex.urlProtocolReplace=/(https?|ftp|file):\/\//,b.addParser({id:"url",is:function(a){return b.regex.urlProtocolTest.test(a)},format:function(c){return c?a.trim(c.replace(b.regex.urlProtocolReplace,"")):c},parsed:!0,type:"text"}),b.regex.dash=/-/g,b.regex.isoDate=/^\d{4}[\/\-]\d{1,2}[\/\-]\d{1,2}/,b.addParser({id:"isoDate",is:function(a){return b.regex.isoDate.test(a)},format:function(a,c){var d=a?new Date(a.replace(b.regex.dash,"/")):a;return d instanceof Date&&isFinite(d)?d.getTime():a},type:"numeric"}),b.regex.percent=/%/g,b.regex.percentTest=/(\d\s*?%|%\s*?\d)/,b.addParser({id:"percent",is:function(a){return b.regex.percentTest.test(a)&&a.length<15},format:function(a,c){return a?b.formatFloat(a.replace(b.regex.percent,""),c):a},type:"numeric"}),b.addParser({id:"image",is:function(a,b,c,d){return d.find("img").length>0},format:function(b,c,d){return a(d).find("img").attr(c.config.imgAttr||"alt")||b},parsed:!0,type:"text"}),b.regex.dateReplace=/(\S)([AP]M)$/i,b.regex.usLongDateTest1=/^[A-Z]{3,10}\.?\s+\d{1,2},?\s+(\d{4})(\s+\d{1,2}:\d{2}(:\d{2})?(\s+[AP]M)?)?$/i,b.regex.usLongDateTest2=/^\d{1,2}\s+[A-Z]{3,10}\s+\d{4}/i,b.addParser({id:"usLongDate",is:function(a){return b.regex.usLongDateTest1.test(a)||b.regex.usLongDateTest2.test(a)},format:function(a,c){var d=a?new Date(a.replace(b.regex.dateReplace,"$1 $2")):a;return d instanceof Date&&isFinite(d)?d.getTime():a},type:"numeric"}),b.regex.shortDateTest=/(^\d{1,2}[\/\s]\d{1,2}[\/\s]\d{4})|(^\d{4}[\/\s]\d{1,2}[\/\s]\d{1,2})/,b.regex.shortDateReplace=/[\-.,]/g,b.regex.shortDateXXY=/(\d{1,2})[\/\s](\d{1,2})[\/\s](\d{4})/,b.regex.shortDateYMD=/(\d{4})[\/\s](\d{1,2})[\/\s](\d{1,2})/,b.convertFormat=function(a,c){a=(a||"").replace(b.regex.spaces," ").replace(b.regex.shortDateReplace,"/"),"mmddyyyy"===c?a=a.replace(b.regex.shortDateXXY,"$3/$1/$2"):"ddmmyyyy"===c?a=a.replace(b.regex.shortDateXXY,"$3/$2/$1"):"yyyymmdd"===c&&(a=a.replace(b.regex.shortDateYMD,"$1/$2/$3"));var d=new Date(a);return d instanceof Date&&isFinite(d)?d.getTime():""},b.addParser({id:"shortDate",is:function(a){return a=(a||"").replace(b.regex.spaces," ").replace(b.regex.shortDateReplace,"/"),b.regex.shortDateTest.test(a)},format:function(a,c,d,e){if(a){var f=c.config,g=f.$headerIndexed[e],h=g.length&&g.data("dateFormat")||b.getData(g,b.getColumnData(c,f.headers,e),"dateFormat")||f.dateFormat;return g.length&&g.data("dateFormat",h),b.convertFormat(a,h)||a}return a},type:"numeric"}),b.regex.timeTest=/^([1-9]|1[0-2]):([0-5]\d)(\s[AP]M)$|^((?:[01]\d|[2][0-4]):[0-5]\d)$/i,b.regex.timeMatch=/([1-9]|1[0-2]):([0-5]\d)(\s[AP]M)|((?:[01]\d|[2][0-4]):[0-5]\d)/i,b.addParser({id:"time",is:function(a){return b.regex.timeTest.test(a)},format:function(a,c){var d,e=(a||"").match(b.regex.timeMatch),f=new Date(a),g=a&&(null!==e?e[0]:"00:00 AM"),h=g?new Date("2000/01/01 "+g.replace(b.regex.dateReplace,"$1 $2")):g;return h instanceof Date&&isFinite(h)?(d=f instanceof Date&&isFinite(f)?f.getTime():0,d?parseFloat(h.getTime()+"."+f.getTime()):h.getTime()):a},type:"numeric"}),b.addParser({id:"metadata",is:function(){return!1},format:function(b,c,d){var e=c.config,f=e.parserMetadataName?e.parserMetadataName:"sortValue";return a(d).metadata()[f]},type:"numeric"}),b.addWidget({id:"zebra",priority:90,format:function(b,c,d){var e,f,g,h,i,j,k,l=new RegExp(c.cssChildRow,"i"),m=c.$tbodies.add(a(c.namespace+"_extra_table").children("tbody:not(."+c.cssInfoBlock+")"));for(i=0;i<m.length;i++)for(g=0,e=m.eq(i).children("tr:visible").not(c.selectorRemove),k=e.length,j=0;k>j;j++)f=e.eq(j),l.test(f[0].className)||g++,h=g%2===0,f.removeClass(d.zebra[h?1:0]).addClass(d.zebra[h?0:1])},remove:function(a,c,d,e){if(!e){var f,g,h=c.$tbodies,i=(d.zebra||["even","odd"]).join(" ");for(f=0;f<h.length;f++)g=b.processTbody(a,h.eq(f),!0),g.children().removeClass(i),b.processTbody(a,g,!1)}}})}(jQuery),a.tablesorter});

.local/share/jupyter/nbextensions/varInspector/main.css

@@ -0,0 +1,119 @@
+
+.varInspector {
+  max-height: 500px;
+  min-height: 100px;
+  font-size: 80%;
+  padding: 0px;
+  overflow-y: auto;
+  font-weight: normal;
+  color: #333333;
+  white-space: nowrap;
+  overflow-x: auto;
+} 
+
+.varInspector-float-wrapper {
+  position: fixed !important;
+  top: 120px;
+  width:350px;
+  max-width:800px;
+  right: 20px;
+  border: thin solid rgba(0, 0, 0, 0.38);
+  border-radius: 5px;
+  padding:10px;
+  background-color: #fff;
+  opacity: .95;
+  z-index: 100;
+  overflow: hidden;
+}
+
+.hide-btn{
+  float: right;
+}
+
+.reload-btn{
+  float: right;
+}
+
+.kill-btn{
+  float: right;
+}
+
+.col-md-9 {
+  overflow:hidden;
+  margin-left: 14%;
+  width: 80%}
+
+#varInspector-wrapper.closed {
+  min-width: 250px;
+  width: auto;
+  transition: width;
+}
+#varInspector-wrapper:hover{
+  opacity: 1;
+}
+#varInspector-wrapper .header {
+  font-size: 16px;
+  font-weight: bold;
+}
+#varInspector-wrapper .hide-btn {
+  font-size: 14px;
+  font-family: monospace;
+}
+
+#varInspector-wrapper .reload-btn {
+  font-size: 14px;
+  font-family: monospace;
+}
+
+#varInspector-wrapper .kill-btn {
+  font-size: 14px;
+  font-family: monospace;
+}
+
+
+
+/* don't waste so much screen space... */
+#varInspector-wrapper .toc-item{
+  padding-left: 20px;
+}
+
+#varInspector-wrapper .toc-item .toc-item{
+  padding-left: 10px;
+}
+
+
+
+table.table, table.table tr, table.table td, table.table th {
+    border: 0;
+}
+table.table-nonfluid {
+    width: auto !important;
+}
+table.table {
+    margin-left: 0;
+    margin-right: 0;
+}
+/* stuff for tablesorter plugin */
+.tablesorter-default .header,
+.tablesorter-default .tablesorter-header {
+  background-image: url(data:image/gif;base64,R0lGODlhFQAJAIAAACMtMP///yH5BAEAAAEALAAAAAAVAAkAAAIXjI+AywnaYnhUMoqt3gZXPmVg94yJVQAAOw==);
+  background-position: right center;
+  background-repeat: no-repeat;
+  cursor: pointer;
+  padding-right: 20px;
+}
+.tablesorter-default thead .headerSortUp,
+.tablesorter-default thead .tablesorter-headerSortUp,
+.tablesorter-default thead .tablesorter-headerAsc {
+  background-image: url(data:image/gif;base64,R0lGODlhFQAEAIAAACMtMP///yH5BAEAAAEALAAAAAAVAAQAAAINjI8Bya2wnINUMopZAQA7);
+}
+.tablesorter-default thead .headerSortDown,
+.tablesorter-default thead .tablesorter-headerSortDown,
+.tablesorter-default thead .tablesorter-headerDesc {
+  background-image: url(data:image/gif;base64,R0lGODlhFQAEAIAAACMtMP///yH5BAEAAAEALAAAAAAVAAQAAAINjB+gC+jP2ptn0WskLQA7);
+}
+.tablesorter-default thead .sorter-false {
+  background-image: none;
+  cursor: default;
+  padding-right: 5px;
+}

.local/share/jupyter/nbextensions/varInspector/main.js

@@ -0,0 +1,462 @@
+define([
+    'require',
+    'jquery',
+    'base/js/namespace',
+    'base/js/events',
+    'notebook/js/codecell'
+], function(
+    requirejs,
+    $,
+    Jupyter,
+    events,
+    codecell
+) {
+    "use strict";
+
+    var mod_name = "varInspector";
+    var log_prefix = '[' + mod_name + '] ';
+
+
+    // ...........Parameters configuration......................
+    // define default values for config parameters if they were not present in general settings (notebook.json)
+    var cfg = {
+        'window_display': false,
+        'cols': {
+            'lenName': 16,
+            'lenType': 16,
+            'lenVar': 40
+        },
+        'kernels_config' : {
+            'python': {
+                library: 'var_list.py',
+                delete_cmd_prefix: 'del ',
+                delete_cmd_postfix: '',
+                varRefreshCmd: 'print(var_dic_list())'
+            },
+            'r': {
+                library: 'var_list.r',
+                delete_cmd_prefix: 'rm(',
+                delete_cmd_postfix: ') ',
+                varRefreshCmd: 'cat(var_dic_list()) '
+            }
+        },
+        'types_to_exclude': ['module', 'function', 'builtin_function_or_method', 'instance', '_Feature']
+    }
+
+
+
+    //.....................global variables....
+
+
+    var st = {}
+    st.config_loaded = false;
+    st.extension_initialized = false;
+    st.code_init = "";
+
+    function read_config(cfg, callback) { // read after nb is loaded
+        var config = Jupyter.notebook.config;
+        config.loaded.then(function() {
+            // config may be specified at system level or at document level.
+            // first, update defaults with config loaded from server
+            cfg = $.extend(true, cfg, config.data.varInspector);
+            // then update cfg with some vars found in current notebook metadata
+            // and save in nb metadata (then can be modified per document)
+
+            // window_display is taken from notebook metadata
+            if (Jupyter.notebook.metadata.varInspector) {
+                if (Jupyter.notebook.metadata.varInspector.window_display)
+                    cfg.window_display = Jupyter.notebook.metadata.varInspector.window_display;
+            }
+
+            cfg = Jupyter.notebook.metadata.varInspector = $.extend(true,
+            cfg, Jupyter.notebook.metadata.varInspector);       
+
+            // but cols and kernels_config are taken from system (if defined)
+            if (config.data.varInspector) {
+                if (config.data.varInspector.cols) {
+                    cfg.cols = $.extend(true, cfg.cols, config.data.varInspector.cols);  
+                }
+                if (config.data.varInspector.kernels_config) {
+                    cfg.kernels_config = $.extend(true, cfg.kernels_config, config.data.varInspector.kernels_config);  
+                }
+            }
+
+            // call callbacks
+            callback && callback();
+            st.config_loaded = true;
+        })
+        return cfg;
+    }
+
+    var sortable;
+
+    function toggleVarInspector() {
+        toggle_varInspector(cfg, st)
+    }
+
+    var varInspector_button = function() {
+        if (!Jupyter.toolbar) {
+            events.on("app_initialized.NotebookApp", varInspector_button);
+            return;
+        }
+        if ($("#varInspector_button").length === 0) {
+            $(Jupyter.toolbar.add_buttons_group([
+                Jupyter.keyboard_manager.actions.register ({
+                    'help'   : 'Variable Inspector',
+                    'icon'   : 'fa-crosshairs',
+                    'handler': toggleVarInspector,
+                }, 'toggle-variable-inspector', 'varInspector')
+            ])).find('.btn').attr('id', 'varInspector_button');
+        }
+    };
+
+    var load_css = function() {
+        var link = document.createElement("link");
+        link.type = "text/css";
+        link.rel = "stylesheet";
+        link.href = requirejs.toUrl("./main.css");
+        document.getElementsByTagName("head")[0].appendChild(link);
+    };
+
+
+function html_table(jsonVars) {
+    function _trunc(x, L) {
+        x = String(x)
+        if (x.length < L) return x
+        else return x.substring(0, L - 3) + '...'
+    }
+    var kernelLanguage = Jupyter.notebook.metadata.kernelspec.language.toLowerCase()
+    var kernel_config = cfg.kernels_config[kernelLanguage];
+    var varList = JSON.parse(String(jsonVars))
+
+    var shape_str = '';
+    var has_shape = false;
+    if (varList.some(listVar => "varShape" in listVar && listVar.varShape !== '')) { //if any of them have a shape
+        shape_str = '<th >Shape</th>';
+        has_shape = true;
+    }
+    var beg_table = '<div class=\"inspector\"><table class=\"table fixed table-condensed table-nonfluid \"><col /> \
+ <col  /><col /><thead><tr><th >X</th><th >Name</th><th >Type</th><th >Size</th>' + shape_str + '<th >Value</th></tr></thead><tr><td> \
+ </td></tr>';
+    varList.forEach(listVar => {
+        var shape_col_str = '</td><td>';
+        if (has_shape) {
+            shape_col_str = '</td><td>' + listVar.varShape + '</td><td>';
+        }
+        beg_table +=
+            '<tr><td><a href=\"#\" onClick=\"Jupyter.notebook.kernel.execute(\'' +
+            kernel_config.delete_cmd_prefix + listVar.varName + kernel_config.delete_cmd_postfix + '\'' + '); ' +
+            'Jupyter.notebook.events.trigger(\'varRefresh\'); \">x</a></td>' +
+            '<td>' + _trunc(listVar.varName, cfg.cols.lenName) + '</td><td>' + _trunc(listVar.varType, cfg.cols.lenType) +
+            '</td><td>' + listVar.varSize + shape_col_str + _trunc(listVar.varContent, cfg.cols.lenVar) +
+            '</td></tr>';
+    });
+    var full_table = beg_table + '</table></div>';
+    return full_table;
+    }
+
+
+
+    function code_exec_callback(msg) {
+        var jsonVars = msg.content['text'];
+        var notWellDefined = false;
+        if (msg.content.evalue) 
+            notWellDefined = msg.content.evalue == "name 'var_dic_list' is not defined" || 
+        msg.content.evalue.substr(0,28) == "Error in cat(var_dic_list())"
+        //means that var_dic_list was cleared ==> need to retart the extension
+        if (notWellDefined) varInspector_init() 
+        else $('#varInspector').html(html_table(jsonVars))
+        
+        requirejs(['nbextensions/varInspector/jquery.tablesorter.min'],
+            function() {
+        setTimeout(function() { if ($('#varInspector').length>0)
+            $('#varInspector table').tablesorter()}, 50)
+        });
+    }
+
+    function tableSort() {
+        requirejs(['nbextensions/varInspector/jquery.tablesorter.min'])
+        $('#varInspector table').tablesorter()
+    }
+
+    var varRefresh = function() {
+        var kernelLanguage = Jupyter.notebook.metadata.kernelspec.language.toLowerCase()
+        var kernel_config = cfg.kernels_config[kernelLanguage];
+        requirejs(['nbextensions/varInspector/jquery.tablesorter.min'],
+            function() {
+                Jupyter.notebook.kernel.execute(
+                    kernel_config.varRefreshCmd, { iopub: { output: code_exec_callback } }, { silent: false }
+                );
+            });
+    }
+
+
+    var varInspector_init = function() {
+        // Define code_init
+        // read and execute code_init 
+        function read_code_init(lib) {
+            var libName = Jupyter.notebook.base_url + "nbextensions/varInspector/" + lib;
+            $.get(libName).done(function(data) {
+                st.code_init = data;
+                st.code_init = st.code_init.replace('lenName', cfg.cols.lenName).replace('lenType', cfg.cols.lenType)
+                        .replace('lenVar', cfg.cols.lenVar)
+                        //.replace('types_to_exclude', JSON.stringify(cfg.types_to_exclude).replace(/\"/g, "'"))
+                requirejs(
+                        [
+                            'nbextensions/varInspector/jquery.tablesorter.min'
+                            //'nbextensions/varInspector/colResizable-1.6.min'
+                        ],
+                        function() {
+                            Jupyter.notebook.kernel.execute(st.code_init, { iopub: { output: code_exec_callback } }, { silent: false });
+                        })
+                    variable_inspector(cfg, st);  // create window if not already present      
+                console.log(log_prefix + 'loaded library');
+            }).fail(function() {
+                console.log(log_prefix + 'failed to load ' + lib + ' library')
+            });
+        }
+
+            // read configuration  
+
+            cfg = read_config(cfg, function() {                
+            // Called when config is available
+                if (typeof Jupyter.notebook.kernel !== "undefined" && Jupyter.notebook.kernel !== null) {
+                    var kernelLanguage = Jupyter.notebook.metadata.kernelspec.language.toLowerCase()
+                    var kernel_config = cfg.kernels_config[kernelLanguage];
+                    if (kernel_config === undefined) { // Kernel is not supported
+                        console.warn(log_prefix + " Sorry, can't use kernel language " + kernelLanguage + ".\n" +
+                            "Configurations are currently only defined for the following languages:\n" +
+                            Object.keys(cfg.kernels_config).join(', ') + "\n" +
+                            "See readme for more details.");
+                        if ($("#varInspector_button").length > 0) { // extension was present
+                            $("#varInspector_button").remove(); 
+                            $('#varInspector-wrapper').remove();
+                            // turn off events
+                            events.off('execute.CodeCell', varRefresh); 
+                            events.off('varRefresh', varRefresh);
+                        }
+                        return
+                    }
+                    varInspector_button(); // In case button was removed 
+                    // read and execute code_init (if kernel is supported)
+                    read_code_init(kernel_config.library);
+                    // console.log("code_init-->", st.code_init)
+                    } else {
+                    console.warn(log_prefix + "Kernel not available?");
+                    }
+            }); // called after config is stable  
+
+            // event: on cell execution, update the list of variables 
+            events.on('execute.CodeCell', varRefresh);
+            events.on('varRefresh', varRefresh);
+            }
+
+
+    var create_varInspector_div = function(cfg, st) {
+        function save_position(){
+            Jupyter.notebook.metadata.varInspector.position = {
+                'left': $('#varInspector-wrapper').css('left'),
+                'top': $('#varInspector-wrapper').css('top'),
+                'width': $('#varInspector-wrapper').css('width'),
+                'height': $('#varInspector-wrapper').css('height'),
+                'right': $('#varInspector-wrapper').css('right')
+            };
+        }
+        var varInspector_wrapper = $('<div id="varInspector-wrapper"/>')
+            .append(
+                $('<div id="varInspector-header"/>')
+                .addClass("header")
+                .text("Variable Inspector ")
+                .append(
+                    $("<a/>")
+                    .attr("href", "#")
+                    .text("[x]")
+                    .addClass("kill-btn")
+                    .attr('title', 'Close window')
+                    .click(function() {
+                        toggleVarInspector();
+                        return false;
+                    })
+                )
+                .append(
+                    $("<a/>")
+                    .attr("href", "#")
+                    .addClass("hide-btn")
+                    .attr('title', 'Hide Variable Inspector')
+                    .text("[-]")
+                    .click(function() {
+                        $('#varInspector-wrapper').css('position', 'fixed');
+                        $('#varInspector').slideToggle({
+                            start: function(event, ui) {
+                                // $(this).width($(this).width());
+                            },
+                            'complete': function() {
+                                    Jupyter.notebook.metadata.varInspector['varInspector_section_display'] = $('#varInspector').css('display');
+                                    save_position();
+                                    Jupyter.notebook.set_dirty();
+                            }
+                        });
+                        $('#varInspector-wrapper').toggleClass('closed');
+                        if ($('#varInspector-wrapper').hasClass('closed')) {
+                            cfg.oldHeight = $('#varInspector-wrapper').height(); //.css('height');
+                            $('#varInspector-wrapper').css({ height: 40 });
+                            $('#varInspector-wrapper .hide-btn')
+                                .text('[+]')
+                                .attr('title', 'Show Variable Inspector');
+                        } else {
+                            $('#varInspector-wrapper').height(cfg.oldHeight); //css({ height: cfg.oldHeight });
+                            $('#varInspector').height(cfg.oldHeight - $('#varInspector-header').height() - 30 )
+                            $('#varInspector-wrapper .hide-btn')
+                                .text('[-]')
+                                .attr('title', 'Hide Variable Inspector');
+                        }
+                        return false;
+                    })
+                ).append(
+                    $("<a/>")
+                    .attr("href", "#")
+                    .text("  \u21BB")
+                    .addClass("reload-btn")
+                    .attr('title', 'Reload Variable Inspector')
+                    .click(function() {
+                        //variable_inspector(cfg,st); 
+                        varRefresh();
+                        return false;
+                    })
+                ).append(
+                    $("<span/>")
+                    .html("&nbsp;&nbsp")
+                ).append(
+                    $("<span/>")
+                    .html("&nbsp;&nbsp;")
+                )
+            ).append(
+                $("<div/>").attr("id", "varInspector").addClass('varInspector')
+            )
+
+        $("body").append(varInspector_wrapper);
+        // Ensure position is fixed
+        $('#varInspector-wrapper').css('position', 'fixed');
+
+        // enable dragging and save position on stop moving
+        $('#varInspector-wrapper').draggable({
+            drag: function(event, ui) {}, //end of drag function
+            start: function(event, ui) {
+                $(this).width($(this).width());
+            },
+            stop: function(event, ui) { // on save, store window position
+                    save_position();
+                    Jupyter.notebook.set_dirty();
+                // Ensure position is fixed (again)
+                $('#varInspector-wrapper').css('position', 'fixed');
+            },
+        });
+
+        $('#varInspector-wrapper').resizable({
+            resize: function(event, ui) {
+                $('#varInspector').height($('#varInspector-wrapper').height() - $('#varInspector-header').height());
+            },
+            start: function(event, ui) {
+                //$(this).width($(this).width());
+                $(this).css('position', 'fixed');
+            },
+            stop: function(event, ui) { // on save, store window position
+                    save_position();
+                    $('#varInspector').height($('#varInspector-wrapper').height() - $('#varInspector-header').height())
+                    Jupyter.notebook.set_dirty();
+                // Ensure position is fixed (again)
+                //$(this).css('position', 'fixed');
+            }
+        })
+
+        // restore window position at startup
+            if (Jupyter.notebook.metadata.varInspector.position !== undefined) {
+                $('#varInspector-wrapper').css(Jupyter.notebook.metadata.varInspector.position);
+            }
+        // Ensure position is fixed
+        $('#varInspector-wrapper').css('position', 'fixed');
+
+        // Restore window display 
+            if (Jupyter.notebook.metadata.varInspector !== undefined) {
+                if (Jupyter.notebook.metadata.varInspector['varInspector_section_display'] !== undefined) {
+                    $('#varInspector').css('display', Jupyter.notebook.metadata.varInspector['varInspector_section_display'])
+                    //$('#varInspector').css('height', $('#varInspector-wrapper').height() - $('#varInspector-header').height())
+                    if (Jupyter.notebook.metadata.varInspector['varInspector_section_display'] == 'none') {
+                        $('#varInspector-wrapper').addClass('closed');
+                        $('#varInspector-wrapper').css({ height: 40 });
+                        $('#varInspector-wrapper .hide-btn')
+                            .text('[+]')
+                            .attr('title', 'Show Variable Inspector');
+                    }
+                }
+                if (Jupyter.notebook.metadata.varInspector['window_display'] !== undefined) {
+                    console.log(log_prefix + "Restoring Variable Inspector window");
+                    $('#varInspector-wrapper').css('display', Jupyter.notebook.metadata.varInspector['window_display'] ? 'block' : 'none');
+                    if ($('#varInspector-wrapper').hasClass('closed')){
+                        $('#varInspector').height(cfg.oldHeight - $('#varInspector-header').height())
+                    }else{
+                        $('#varInspector').height($('#varInspector-wrapper').height() - $('#varInspector-header').height()-30)
+                    }
+                    
+                }
+            }
+        // if varInspector-wrapper is undefined (first run(?), then hide it)
+        if ($('#varInspector-wrapper').css('display') == undefined) $('#varInspector-wrapper').css('display', "none") //block
+
+        varInspector_wrapper.addClass('varInspector-float-wrapper');
+    }
+
+    var variable_inspector = function(cfg, st) {
+
+        var varInspector_wrapper = $("#varInspector-wrapper");
+        if (varInspector_wrapper.length === 0) {
+            create_varInspector_div(cfg, st);
+        }
+
+        $(window).resize(function() {
+            $('#varInspector').css({ maxHeight: $(window).height() - 30 });
+            $('#varInspector-wrapper').css({ maxHeight: $(window).height() - 10 });
+        });
+
+        $(window).trigger('resize');
+        varRefresh();
+    };
+
+    var toggle_varInspector = function(cfg, st) {
+        // toggle draw (first because of first-click behavior)
+        $("#varInspector-wrapper").toggle({
+            'progress': function() {},
+            'complete': function() {
+                    Jupyter.notebook.metadata.varInspector['window_display'] = $('#varInspector-wrapper').css('display') == 'block';
+                    Jupyter.notebook.set_dirty();
+                // recompute:
+                variable_inspector(cfg, st);
+            }
+        });
+    };
+
+
+    var load_jupyter_extension = function() {
+        load_css(); //console.log("Loading css")
+        varInspector_button(); //console.log("Adding varInspector_button")
+
+        // If a kernel is available, 
+        if (typeof Jupyter.notebook.kernel !== "undefined" && Jupyter.notebook.kernel !== null) {
+            console.log(log_prefix + "Kernel is available -- varInspector initializing ")
+            varInspector_init();
+        }
+        // if a kernel wasn't available, we still wait for one. Anyway, we will run this for new kernel 
+        // (test if is is a Python kernel and initialize)
+        // on kernel_ready.Kernel, a new kernel has been started and we shall initialize the extension
+        events.on("kernel_ready.Kernel", function(evt, data) {
+            console.log(log_prefix + "Kernel is available -- reading configuration");
+            varInspector_init();
+        });
+    };
+
+    return {
+        load_ipython_extension: load_jupyter_extension,
+        varRefresh: varRefresh
+    };
+
+});

.local/share/jupyter/nbextensions/varInspector/varInspector.yaml

@@ -0,0 +1,45 @@
+Type: Jupyter Notebook Extension
+Name: Variable Inspector
+Description: The Variable Inspector extension collects all defined variables and display them in a floating window. The extension is also draggable, resizable, collapsable. 
+Link: README.md
+Icon: icon.png
+Main: main.js
+Compatibility: 4.x, 5.x
+Parameters:
+- name: varInspector.window_display
+  description: Display window at startup
+  input_type: checkbox
+  default: false
+- name: varInspector.cols.lenName
+  description: Variable name - Maximum number of characters to display
+  input_type: number 
+  default: 16  
+- name: varInspector.cols.lenType
+  description: Variable type - Maximum number of characters to display
+  input_type: number 
+  default: 16  
+- name: varInspector.cols.lenVar
+  description: Variable value/content - Maximum number of characters to display
+  input_type: number 
+  default: 40  
+- name: varInspector.kernels_config
+  description: |
+    json object defining the libraries to load, the delete/remove 
+    kernel commands to delete a variable, and 
+    finally the commands to refresh the list of variables.
+  input_type: json_object
+  default: |
+    {
+      'python': {
+                library: 'var_list.py',
+                delete_cmd_prefix: 'del ',
+                delete_cmd_postfix: '',
+                varRefreshCmd: 'print(var_dic_list())'
+            },
+      'r': {
+                library: 'var_list.r',
+                delete_cmd_prefix: 'rm(',
+                delete_cmd_postfix: ') ',
+                varRefreshCmd: 'cat(var_dic_list()) '
+            }
+    }

.local/share/jupyter/nbextensions/varInspector/var_list.r

@@ -0,0 +1,17 @@
+library(jsonlite)
+var_dic_list = function(){ 
+    ll = ls(.GlobalEnv, all.names = FALSE)
+    varList=list()
+    iter = 1
+        for (k in ll){
+            if (class(get(k))!='function'){
+            class = class(get(k)); rk = capture.output(str(get(k))); size =  object.size(get(k)); sk = substr(get(k),0, 200); 
+            #  [{'varName':v, 'varType': type(eval(v)).__name__, 'varSize': _getsizeof(eval(v)), 'varContent': str(eval(v))[:200]} 
+            l = list(varName = k, varType = class, varSize = size, varContent = sk)
+            varList[[iter]] = l
+            # print(l)
+            iter = iter + 1}
+        }
+return(toJSON(varList, simplifyVector = FALSE, force=TRUE))
+    }
+cat(var_dic_list())

.local/share/jupyter/nbextensions/zenmode/README.md

@@ -0,0 +1,4 @@
+Zenmode
+=======
+
+A little extension to give Zenmode functionality to the IPython notebook

.local/share/jupyter/nbextensions/zenmode/main.css

@@ -0,0 +1,34 @@
+.navbar-inner {
+    opacity: 0.5;
+    -webkit-transition: opacity 0.3s ease-in-out;
+    -moz-transition: opacity 0.3s ease-in-out;
+    -o-transition: opacity 0.3s ease-in-out;
+    transition: opacity 0.3s ease-in-out;
+}
+
+.navbar-inner:hover {
+    opacity: 1.0;
+}
+
+#maintoolbar .navbar-text {
+    display: none !important;
+}
+
+#notebook-container {
+    background-color: rgba(255, 255, 255, 0);
+}
+
+/*
+.cell {
+    background-color: rgb(255, 255, 255);
+}
+
+.CodeMirror {
+    background: #F8FCCF;
+}
+
+div.input_area {
+    margin: 2px;
+    border: none;
+}
+*/

.local/share/jupyter/nbextensions/zenmode/main.js

@@ -0,0 +1,196 @@
+/**
+* ----------------------------------------------------------------------------
+* Copyright (c) 2013 - Damián Avila
+* Copyright (c) 2015 - Joshua Cooke Barnes (jcb91)
+*
+* Distributed under the terms of the Modified BSD License.
+*
+* A little extension to give Zenmode functionality to the IPython notebook.
+* ----------------------------------------------------------------------------
+*/
+
+define([
+    "require",
+    "jquery",
+    "base/js/namespace",
+    "base/js/events"
+], function(
+    requirejs,
+    $,
+    IPython,
+    events
+) {
+    "use_strict";
+
+    var backgrounds = [
+        'back11.jpg', 'back12.jpg', 'back2.jpg', 'back21.jpg', 'back22.jpg',
+        'back3.jpg', 'ipynblogo0.png', 'ipynblogo1.png'
+    ];
+
+     var hide_header = true;
+     var hide_menubar = true
+
+    var getZenModeActive = function() {
+        return ($('link#zenmodecss')[0] !== undefined);
+    };
+
+    // not sure when this changed, so maybe this is the wrong comparison to make
+    var use_layout_manager = (Number(IPython.version.split(".")[0]) < 3);
+    var header_pattern = use_layout_manager ? '#header' : '#header-container';
+    if (use_layout_manager) {
+        // We need to redefine this function because in the IPython codebase
+        // the app_height function does not take into account the possibility
+        // to hide the header and 'menubar' bar.
+        IPython.layout_manager.app_height = function() {
+            var get_height = function(pattern) {
+                var el = $(pattern);
+                return getZenModeActive() ? 0 : el.outerHeight(true);
+            };
+            var h = $(window).height();
+            // content height
+            return h - get_height(header_pattern) - get_height('#menubar') - get_height('#maintoolbar');
+        };
+    }
+
+    var menu_pattern = '#menubar';
+    var oldBgAttrName = "zenmode-old-bg";
+    var toggleZenMode = function (background) {
+        if (getZenModeActive()) {
+            console.log('toggling zenmode off');
+            $('#zenmode-toggle-btn .fa').removeClass("fa-rebel").addClass("fa-empire");
+            $('#zenmodecss').remove();
+
+
+            // Remove zenmode css settings only when changes were made.
+            if (backgrounds.length != 0) {
+                $('body').css({
+                    'background-image': 'none'
+                })
+            }
+            
+            // This should be changed at some point in the future to preserve non-zenmode visibility settings
+            $(menu_pattern).toggle(true);
+            $(header_pattern).toggle(true);
+        }
+        else {
+            console.log('toggling zenmode on');
+            $('#zenmode-toggle-btn .fa').removeClass("fa-empire").addClass("fa-rebel");
+            $('head').append(
+                $('<link id="zenmodecss" rel="stylesheet" type="text/css"/>').attr(
+                    'href', requirejs.toUrl("./main.css"))
+            );
+
+            if (background === undefined){
+                background = backgrounds[Math.floor(Math.random() * backgrounds.length)];
+            }
+            var absolute_url_pat = /^https?:\/\/|^\/\//i;
+            if (!absolute_url_pat.test(background)) {
+                background = requirejs.toUrl("./images/" + background);
+            }
+
+            // Apply zenmode css when there are images to be used.
+            if (backgrounds.length != 0) {
+                $('body').css({
+                    'background-image': 'url(' + background + ')',
+                    'background-repeat': 'no-repeat',
+                    'background-position': 'center center',
+                    'background-attachment': 'fixed',
+                    '-webkit-background-size': 'cover',
+                    '-moz-background-size': 'cover',
+                    '-o-background-size': 'cover',
+                    'background-size': 'cover'
+                });
+            }
+
+            if (hide_menubar)
+                {$(menu_pattern).toggle(false);}
+            if (hide_header)
+                {$(header_pattern).toggle(false);}
+        }
+
+        // Lastly get notebook to do a resize
+        if (use_layout_manager) {
+            IPython.layout_manager.app_height();
+            IPython.layout_manager.do_resize();
+        }
+        else {
+            events.trigger("resize-header.Page");
+        }
+    };
+
+    var setZenModeActive = function(active, background) {
+        if (active === undefined) { active = true; }
+        console.log("zenmode ->", active);
+        if (getZenModeActive() != active) { toggleZenMode(background); }
+    };
+
+    var initialize = function () {
+    	var config = IPython.notebook.config;
+        if (config.data.hasOwnProperty('zenmode_hide_header')) {
+            if (!config.data.zenmode_hide_header) {
+                console.log("not hiding notebook header");
+                hide_header = false;
+            }
+        }
+
+        if (config.data.hasOwnProperty('zenmode_hide_menubar')) {
+            if (!config.data.zenmode_hide_menubar) {
+                console.log("not hiding notebook menubar");
+                hide_menubar = false;
+            }
+        }
+
+        if (config.data.hasOwnProperty('zenmode_use_builtin_backgrounds')) {
+            if (!config.data.zenmode_use_builtin_backgrounds) {
+                console.log("not using builtin zenmode_backgrounds");
+                backgrounds.length = 0;
+            }
+        }
+
+        if (config.data.hasOwnProperty('zenmode_backgrounds')) {
+            if (config.data.zenmode_backgrounds.length > 0) {
+                var new_bg_urls = config.data.zenmode_backgrounds;
+                for (var ii=0; ii < new_bg_urls.length; ii++) {
+                    var bg_url = new_bg_urls[ii].replace(/^\s+|\s+$/g, '');
+                    if (bg_url.length > 0 && bg_url[0] != '#') {
+                        backgrounds.push(bg_url);
+                    }
+                }
+                console.log("additional zenmode backgrounds added");
+            }
+        }
+        console.log("zenmode_backgrounds = ", backgrounds);
+
+        if (config.data.hasOwnProperty('zenmode_set_zenmode_on_load')) {
+            setZenModeActive(
+                config.data.zenmode_set_zenmode_on_load ? true : false
+            );
+        }
+    };
+
+    var load_ipython_extension = function(background) {
+        $(IPython.toolbar.add_buttons_group([
+            IPython.keyboard_manager.actions.register({
+                'help'   : 'Enter/Exit Zenmode',
+                'icon'    : 'fa-empire',
+                'handler': function() {
+                    toggleZenMode(background);
+                    setTimeout(function () {
+                        $('#zenmode-toggle-btn').blur();
+                    }, 500);
+                },
+            }, 'toggle-zenmode', 'zenmode'),
+        ], 'zenmode-btn-grp')).find('.btn').attr('id', 'zenmode-toggle-btn');
+        $("#maintoolbar-container").prepend($('#zenmode-btn-grp'));
+        return IPython.notebook.config.loaded.then(initialize);
+    };
+
+    var extension = {
+        load_ipython_extension : load_ipython_extension,
+        backgrounds : backgrounds,
+        toggleZenMode : toggleZenMode,
+        getZenModeActive : getZenModeActive,
+        setZenModeActive : setZenModeActive
+    };
+    return extension;
+});

.local/share/jupyter/nbextensions/zenmode/zenmode.yaml

@@ -0,0 +1,28 @@
+Type: IPython Notebook Extension
+Name: zenmode
+Link: README.md
+Description: A little extension to give Zenmode functionality to the IPython notebook
+Main: main.js
+Compatibility: 4.x, 5.x
+Parameters:
+- name: zenmode_set_zenmode_on_load
+  description: Set zenmode on when a notebook opens
+  input_type: checkbox
+  default: true
+- name: zenmode_use_builtin_backgrounds
+  description: Use builtin backgrounds in addition to any specified by URL
+  input_type: checkbox
+  default: true
+- name: zenmode_hide_header
+  description: Hide the header in zenmode
+  input_type: checkbox
+  default: true
+- name: zenmode_hide_menubar
+  description: Hide the menubar in zenmode
+  input_type: checkbox
+  default: true
+- name: zenmode_backgrounds
+  description: "Urls to use as backgrounds. Any beginning with # are ignored."
+  input_type: list
+  list_element:
+    input_type: url

.triton/dump/0db70b0f0846c3c6c38c4ccb3ef979e3/triton_.ttir

@@ -0,0 +1,113 @@
+module {
+  tt.func public @triton__0d1d2d3d4d5d6de7de(%arg0: !tt.ptr<i64, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg3: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg4: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg5: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg6: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}, %arg7: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %cst = arith.constant dense<0.000000e+00> : tensor<16x256xbf16>
+    %cst_0 = arith.constant dense<1.000000e+00> : tensor<1x256xf32>
+    %cst_1 = arith.constant dense<0.000000e+00> : tensor<1x256xf32>
+    %cst_2 = arith.constant 0.000000e+00 : f32
+    %cst_3 = arith.constant dense<256> : tensor<16x1xi64>
+    %cst_4 = arith.constant dense<50257> : tensor<16x1xi64>
+    %cst_5 = arith.constant dense<0> : tensor<16x1xi64>
+    %cst_6 = arith.constant dense<9.99999974E-6> : tensor<16x1xf32>
+    %cst_7 = arith.constant dense<2.560000e+02> : tensor<16x1xf32>
+    %cst_8 = arith.constant dense<0.000000e+00> : tensor<16x256xf32>
+    %cst_9 = arith.constant dense<256> : tensor<16x1xi32>
+    %cst_10 = arith.constant dense<256> : tensor<1x256xi32>
+    %cst_11 = arith.constant dense<512> : tensor<16x1xi32>
+    %c16_i32 = arith.constant 16 : i32
+    %0 = tt.get_program_id x : i32
+    %1 = arith.muli %0, %c16_i32 : i32
+    %2 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32>
+    %3 = tt.expand_dims %2 {axis = 1 : i32} : (tensor<16xi32>) -> tensor<16x1xi32>
+    %4 = tt.splat %1 : (i32) -> tensor<16x1xi32>
+    %5 = arith.addi %4, %3 : tensor<16x1xi32>
+    %6 = tt.make_range {end = 256 : i32, start = 0 : i32} : tensor<256xi32>
+    %7 = tt.expand_dims %6 {axis = 0 : i32} : (tensor<256xi32>) -> tensor<1x256xi32>
+    %8 = tt.splat %arg0 : (!tt.ptr<i64, 1>) -> tensor<16x1x!tt.ptr<i64, 1>>
+    %9 = tt.addptr %8, %5 : tensor<16x1x!tt.ptr<i64, 1>>, tensor<16x1xi32>
+    %10 = tt.load %9 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<16x1xi64>
+    %11 = arith.remsi %5, %cst_11 : tensor<16x1xi32>
+    %12 = arith.cmpi slt, %7, %cst_10 : tensor<1x256xi32>
+    %13 = arith.muli %11, %cst_9 : tensor<16x1xi32>
+    %14 = tt.broadcast %7 : (tensor<1x256xi32>) -> tensor<16x256xi32>
+    %15 = tt.broadcast %13 : (tensor<16x1xi32>) -> tensor<16x256xi32>
+    %16 = arith.addi %14, %15 : tensor<16x256xi32>
+    %17 = tt.splat %arg2 : (!tt.ptr<f32, 1>) -> tensor<16x256x!tt.ptr<f32, 1>>
+    %18 = tt.addptr %17, %16 : tensor<16x256x!tt.ptr<f32, 1>>, tensor<16x256xi32>
+    %19 = tt.broadcast %12 : (tensor<1x256xi1>) -> tensor<16x256xi1>
+    %20 = tt.load %18, %19, %cst_8 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<16x256xf32>
+    %21 = arith.muli %5, %cst_9 : tensor<16x1xi32>
+    %22 = tt.broadcast %21 : (tensor<16x1xi32>) -> tensor<16x256xi32>
+    %23 = arith.addi %14, %22 : tensor<16x256xi32>
+    %24 = tt.splat %arg3 : (!tt.ptr<bf16, 1>) -> tensor<16x256x!tt.ptr<bf16, 1>>
+    %25 = tt.addptr %24, %23 : tensor<16x256x!tt.ptr<bf16, 1>>, tensor<16x256xi32>
+    %26 = tt.load %25, %19, %cst {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<16x256xbf16>
+    %27 = arith.extf %26 : tensor<16x256xbf16> to tensor<16x256xf32>
+    %28 = arith.addi %10, %cst_4 : tensor<16x1xi64>
+    %29 = arith.cmpi slt, %10, %cst_5 : tensor<16x1xi64>
+    %30 = arith.select %29, %28, %10 : tensor<16x1xi1>, tensor<16x1xi64>
+    %31 = arith.cmpi sge, %30, %cst_5 : tensor<16x1xi64>
+    %32 = arith.cmpi slt, %30, %cst_4 : tensor<16x1xi64>
+    %33 = arith.andi %31, %32 : tensor<16x1xi1>
+    tt.assert %33, "index out of bounds: 0 <= tmp3 < 50257", "<frozen importlib._bootstrap_external>", "_call_with_frames_removed", 883 : tensor<16x1xi1>
+    %34 = arith.muli %30, %cst_3 : tensor<16x1xi64>
+    %35 = tt.broadcast %34 : (tensor<16x1xi64>) -> tensor<16x256xi64>
+    %36 = arith.extsi %7 : tensor<1x256xi32> to tensor<1x256xi64>
+    %37 = tt.broadcast %36 : (tensor<1x256xi64>) -> tensor<16x256xi64>
+    %38 = arith.addi %37, %35 : tensor<16x256xi64>
+    %39 = tt.splat %arg1 : (!tt.ptr<f32, 1>) -> tensor<16x256x!tt.ptr<f32, 1>>
+    %40 = tt.addptr %39, %38 : tensor<16x256x!tt.ptr<f32, 1>>, tensor<16x256xi64>
+    %41 = tt.load %40, %19, %cst_8 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<16x256xf32>
+    %42 = arith.addf %41, %20 : tensor<16x256xf32>
+    %43 = arith.addf %42, %27 : tensor<16x256xf32>
+    %44 = arith.addf %43, %cst_8 : tensor<16x256xf32>
+    %45 = arith.subf %43, %44 : tensor<16x256xf32>
+    %46 = arith.mulf %43, %45 : tensor<16x256xf32>
+    %47 = arith.addf %46, %cst_8 : tensor<16x256xf32>
+    %48 = arith.select %19, %44, %cst_8 : tensor<16x256xi1>, tensor<16x256xf32>
+    %49 = arith.select %19, %47, %cst_8 : tensor<16x256xi1>, tensor<16x256xf32>
+    %50 = arith.select %12, %cst_0, %cst_1 : tensor<1x256xi1>, tensor<1x256xf32>
+    %51 = tt.broadcast %50 : (tensor<1x256xf32>) -> tensor<16x256xf32>
+    %52:3 = "tt.reduce"(%48, %49, %51) <{axis = 1 : i32}> ({
+    ^bb0(%arg8: f32, %arg9: f32, %arg10: f32, %arg11: f32, %arg12: f32, %arg13: f32):
+      %76 = arith.subf %arg11, %arg8 : f32
+      %77 = arith.addf %arg10, %arg13 : f32
+      %78 = arith.cmpf oeq, %77, %cst_2 : f32
+      %79 = arith.divf %arg13, %77 : f32
+      %80 = arith.select %78, %cst_2, %79 : f32
+      %81 = arith.mulf %76, %80 : f32
+      %82 = arith.addf %arg8, %81 : f32
+      %83 = arith.addf %arg9, %arg12 : f32
+      %84 = arith.mulf %76, %76 : f32
+      %85 = arith.mulf %84, %arg10 : f32
+      %86 = arith.mulf %85, %80 : f32
+      %87 = arith.addf %83, %86 : f32
+      tt.reduce.return %82, %87, %77 : f32, f32, f32
+    }) : (tensor<16x256xf32>, tensor<16x256xf32>, tensor<16x256xf32>) -> (tensor<16xf32>, tensor<16xf32>, tensor<16xf32>)
+    %53 = tt.expand_dims %52#0 {axis = 1 : i32} : (tensor<16xf32>) -> tensor<16x1xf32>
+    %54 = tt.expand_dims %52#1 {axis = 1 : i32} : (tensor<16xf32>) -> tensor<16x1xf32>
+    %55 = tt.load %18, %19, %cst_8 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<16x256xf32>
+    %56 = tt.load %25, %19, %cst {cache = 1 : i32, evict = 2 : i32, isVolatile = false} : tensor<16x256xbf16>
+    %57 = arith.extf %56 : tensor<16x256xbf16> to tensor<16x256xf32>
+    %58 = tt.splat %arg4 : (!tt.ptr<f32, 1>) -> tensor<1x256x!tt.ptr<f32, 1>>
+    %59 = tt.addptr %58, %7 : tensor<1x256x!tt.ptr<f32, 1>>, tensor<1x256xi32>
+    %60 = tt.load %59, %12, %cst_1 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<1x256xf32>
+    tt.assert %33, "index out of bounds: 0 <= tmp16 < 50257", "<frozen importlib._bootstrap_external>", "_call_with_frames_removed", 883 : tensor<16x1xi1>
+    %61 = tt.load %40, %19, %cst_8 {cache = 1 : i32, evict = 2 : i32, isVolatile = false} : tensor<16x256xf32>
+    %62 = arith.addf %61, %55 : tensor<16x256xf32>
+    %63 = arith.addf %62, %57 : tensor<16x256xf32>
+    %64 = tt.broadcast %53 : (tensor<16x1xf32>) -> tensor<16x256xf32>
+    %65 = arith.subf %63, %64 : tensor<16x256xf32>
+    %66 = arith.divf %54, %cst_7 : tensor<16x1xf32>
+    %67 = arith.addf %66, %cst_6 : tensor<16x1xf32>
+    %68 = tt.extern_elementwise %67 {libname = "libdevice", libpath = "/usr/local/lib/python3.10/dist-packages/triton/language/../third_party/cuda/lib/libdevice.10.bc", pure = true, symbol = "__nv_rsqrtf"} : (tensor<16x1xf32>) -> tensor<16x1xf32>
+    %69 = tt.broadcast %68 : (tensor<16x1xf32>) -> tensor<16x256xf32>
+    %70 = arith.mulf %65, %69 : tensor<16x256xf32>
+    %71 = tt.broadcast %60 : (tensor<1x256xf32>) -> tensor<16x256xf32>
+    %72 = arith.mulf %70, %71 : tensor<16x256xf32>
+    %73 = tt.splat %arg5 : (!tt.ptr<bf16, 1>) -> tensor<16x256x!tt.ptr<bf16, 1>>
+    %74 = tt.addptr %73, %23 : tensor<16x256x!tt.ptr<bf16, 1>>, tensor<16x256xi32>
+    %75 = arith.truncf %72 : tensor<16x256xf32> to tensor<16x256xbf16>
+    tt.store %74, %75, %19 {cache = 1 : i32, evict = 1 : i32} : tensor<16x256xbf16>
+    tt.return
+  }
+}

.triton/dump/1c14bdb6903aa6825e214bbdf57fd077/triton_.ptx

@@ -0,0 +1,312 @@
+//
+// Generated by LLVM NVPTX Back-End
+//
+
+.version 8.2
+.target sm_89
+.address_size 64
+
+	// .globl	triton__0d1d2de
+
+.visible .entry triton__0d1d2de(
+	.param .u64 triton__0d1d2de_param_0,
+	.param .u64 triton__0d1d2de_param_1,
+	.param .u32 triton__0d1d2de_param_2
+)
+.maxntid 128, 1, 1
+{
+	.reg .pred 	%p<4>;
+	.reg .b16 	%rs<9>;
+	.reg .b32 	%r<31>;
+	.reg .b64 	%rd<8>;
+	.loc	1 18 0
+$L__func_begin0:
+	.loc	1 18 0
+
+	ld.param.u64 	%rd4, [triton__0d1d2de_param_0];
+	ld.param.u64 	%rd5, [triton__0d1d2de_param_1];
+$L__tmp0:
+	.loc	1 21 36
+	mov.u32 	%r22, %tid.x;
+	shl.b32 	%r23, %r22, 3;
+	and.b32  	%r24, %r23, 1016;
+	.loc	1 20 28
+	mov.u32 %r1, %ctaid.x;
+	.loc	1 20 33
+	shl.b32 	%r25, %r1, 10;
+	.loc	1 21 23
+	or.b32  	%r26, %r25, %r24;
+	.loc	1 24 30
+	mul.wide.s32 	%rd6, %r26, 4;
+	add.s64 	%rd1, %rd4, %rd6;
+	add.s64 	%rd2, %rd1, 16;
+	mov.pred 	%p1, -1;
+	.loc	1 24 35
+	mov.u32 %r10, 0x0;
+	mov.u32 %r11, 0x0;
+	mov.u32 %r12, 0x0;
+	mov.u32 %r13, 0x0;
+	@%p1 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd1 + 0 ];
+	mov.u32 %r14, 0x0;
+	mov.u32 %r15, 0x0;
+	mov.u32 %r16, 0x0;
+	mov.u32 %r17, 0x0;
+	@%p1 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd2 + 0 ];
+	.loc	1 26 25
+	mul.wide.s32 	%rd7, %r26, 2;
+	add.s64 	%rd3, %rd5, %rd7;
+	.loc	1 26 36
+	cvt.rn.bf16.f32 %rs1, %r10;
+	cvt.rn.bf16.f32 %rs2, %r11;
+	cvt.rn.bf16.f32 %rs3, %r12;
+	cvt.rn.bf16.f32 %rs4, %r13;
+	cvt.rn.bf16.f32 %rs5, %r14;
+	cvt.rn.bf16.f32 %rs6, %r15;
+	cvt.rn.bf16.f32 %rs7, %r16;
+	cvt.rn.bf16.f32 %rs8, %r17;
+	mov.b32 	%r27, {%rs1, %rs2};
+	mov.b32 	%r28, {%rs3, %rs4};
+	mov.b32 	%r29, {%rs5, %rs6};
+	mov.b32 	%r30, {%rs7, %rs8};
+	@%p1 st.global.v4.b32 [ %rd3 + 0 ], { %r27, %r28, %r29, %r30 };
+	.loc	1 26 4
+	ret;
+$L__tmp1:
+$L__func_end0:
+
+}
+	.file	1 "/tmp/torchinductor_root/5t/c5tryp5qwkhreijk7s5x327wofz54lwj4kvctuqdzv2vrf2xyons.py"
+	.section	.debug_abbrev
+	{
+.b8 1
+.b8 17
+.b8 1
+.b8 37
+.b8 8
+.b8 19
+.b8 5
+.b8 3
+.b8 8
+.b8 16
+.b8 6
+.b8 27
+.b8 8
+.b8 180
+.b8 66
+.b8 12
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+.b8 0
+.b8 2
+.b8 46
+.b8 0
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 64
+.b8 10
+.b8 135
+.b8 64
+.b8 8
+.b8 3
+.b8 8
+.b8 58
+.b8 11
+.b8 59
+.b8 11
+.b8 63
+.b8 12
+.b8 0
+.b8 0
+.b8 0
+	}
+	.section	.debug_info
+	{
+.b32 176
+.b8 2
+.b8 0
+.b32 .debug_abbrev
+.b8 8
+.b8 1
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 0
+.b8 2
+.b8 0
+.b8 99
+.b8 53
+.b8 116
+.b8 114
+.b8 121
+.b8 112
+.b8 53
+.b8 113
+.b8 119
+.b8 107
+.b8 104
+.b8 114
+.b8 101
+.b8 105
+.b8 106
+.b8 107
+.b8 55
+.b8 115
+.b8 53
+.b8 120
+.b8 51
+.b8 50
+.b8 55
+.b8 119
+.b8 111
+.b8 102
+.b8 122
+.b8 53
+.b8 52
+.b8 108
+.b8 119
+.b8 106
+.b8 52
+.b8 107
+.b8 118
+.b8 99
+.b8 116
+.b8 117
+.b8 113
+.b8 100
+.b8 122
+.b8 118
+.b8 50
+.b8 118
+.b8 114
+.b8 102
+.b8 50
+.b8 120
+.b8 121
+.b8 111
+.b8 110
+.b8 115
+.b8 46
+.b8 112
+.b8 121
+.b8 0
+.b32 .debug_line
+.b8 47
+.b8 116
+.b8 109
+.b8 112
+.b8 47
+.b8 116
+.b8 111
+.b8 114
+.b8 99
+.b8 104
+.b8 105
+.b8 110
+.b8 100
+.b8 117
+.b8 99
+.b8 116
+.b8 111
+.b8 114
+.b8 95
+.b8 114
+.b8 111
+.b8 111
+.b8 116
+.b8 47
+.b8 53
+.b8 116
+.b8 0
+.b8 1
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 2
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 1
+.b8 156
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 101
+.b8 0
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 101
+.b8 0
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+	}
+	.section	.debug_pubnames
+	{
+.b32 $L__pubNames_end0-$L__pubNames_start0
+$L__pubNames_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 180
+.b32 125
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 101
+.b8 0
+.b32 0
+$L__pubNames_end0:
+	}
+	.section	.debug_pubtypes
+	{
+.b32 $L__pubTypes_end0-$L__pubTypes_start0
+$L__pubTypes_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 180
+.b32 0
+$L__pubTypes_end0:
+	}
+	.section	.debug_loc	{	}

.triton/dump/1e922bbbab749da355e4bad9c6b245e6/triton_.llir

@@ -0,0 +1,332 @@
+; ModuleID = 'LLVMDialectModule'
+source_filename = "LLVMDialectModule"
+
+@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1
+
+define void @triton__0d1de(ptr addrspace(1) %0, i32 %1) local_unnamed_addr !dbg !7 {
+  %3 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !10
+  %4 = shl i32 %3, 1, !dbg !10
+  %5 = and i32 %4, 510, !dbg !10
+  %6 = tail call i32 asm "mov.u32 $0, %ctaid.x;", "=r"() #4, !dbg !11
+  %7 = shl i32 %6, 9, !dbg !12
+  %8 = or i32 %7, %5, !dbg !13
+  %9 = sext i32 %8 to i64, !dbg !14
+  %10 = getelementptr i16, ptr addrspace(1) %0, i64 %9, !dbg !14
+  %11 = tail call i32 asm sideeffect "mov.u32 $0, 0x0;\0A\09@$2 ld.global.b32 { $0 }, [ $1 + 0 ];", "=r,l,b"(ptr addrspace(1) %10, i1 true) #4, !dbg !15
+  %12 = trunc i32 %11 to i16, !dbg !15
+  %extelt.offset = lshr i32 %11, 16, !dbg !15
+  %13 = trunc i32 %extelt.offset to i16, !dbg !15
+  %14 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %12) #4, !dbg !16
+  %15 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %13) #4, !dbg !16
+  %16 = fmul float %14, 0x3FE6A09E60000000, !dbg !17
+  %17 = fmul float %15, 0x3FE6A09E60000000, !dbg !17
+  %18 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not.i = icmp eq i32 %18, 0, !dbg !18
+  %19 = tail call float @llvm.nvvm.fabs.ftz.f(float %16) #4, !dbg !18
+  %20 = tail call float @llvm.nvvm.fabs.f(float %16) #4, !dbg !18
+  %.0.i = select i1 %.not.i, float %20, float %19, !dbg !18
+  %21 = fcmp oge float %.0.i, 0x3FF00C1FC0000000, !dbg !18
+  br i1 %21, label %__nv_fabsf.exit1.i, label %23, !dbg !18
+
+__nv_fabsf.exit1.i:                               ; preds = %2
+  %22 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not1.i = icmp eq i32 %22, 0, !dbg !18
+  %.01.i = select i1 %.not1.i, float %20, float %19, !dbg !18
+  br label %__internal_fmad.exit.i, !dbg !18
+
+23:                                               ; preds = %2
+  %24 = fmul float %16, %16, !dbg !18
+  br label %__internal_fmad.exit.i, !dbg !18
+
+__internal_fmad.exit.i:                           ; preds = %23, %__nv_fabsf.exit1.i
+  %25 = phi float [ 0x3FE41B0840000000, %__nv_fabsf.exit1.i ], [ 0x3FC06EBA60000000, %23 ], !dbg !18
+  %26 = phi float [ 0x3FED526FC0000000, %__nv_fabsf.exit1.i ], [ 0xBFD8127580000000, %23 ], !dbg !18
+  %27 = phi float [ 0x3FC39F20C0000000, %__nv_fabsf.exit1.i ], [ 0x3FBCE315E0000000, %23 ], !dbg !18
+  %28 = phi float [ 0xBFA1902C40000000, %__nv_fabsf.exit1.i ], [ 0xBF9B837CE0000000, %23 ], !dbg !18
+  %29 = phi float [ 0x3F75908160000000, %__nv_fabsf.exit1.i ], [ 0x3F755ABD40000000, %23 ], !dbg !18
+  %30 = phi float [ 0xBF3EAC1720000000, %__nv_fabsf.exit1.i ], [ 0xBF4AE9A400000000, %23 ], !dbg !18
+  %31 = phi float [ 0x3EF1394780000000, %__nv_fabsf.exit1.i ], [ 0x3F163D2D40000000, %23 ], !dbg !18
+  %32 = phi float [ %.01.i, %__nv_fabsf.exit1.i ], [ %24, %23 ], !dbg !18
+  %33 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not2.i = icmp eq i32 %33, 0, !dbg !18
+  %34 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %31, float %32, float %30) #4, !dbg !18
+  %35 = tail call float @llvm.nvvm.fma.rn.f(float %31, float %32, float %30) #4, !dbg !18
+  %.02.i = select i1 %.not2.i, float %35, float %34, !dbg !18
+  %36 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not3.i = icmp eq i32 %36, 0, !dbg !18
+  %37 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.02.i, float %32, float %29) #4, !dbg !18
+  %38 = tail call float @llvm.nvvm.fma.rn.f(float %.02.i, float %32, float %29) #4, !dbg !18
+  %.03.i = select i1 %.not3.i, float %38, float %37, !dbg !18
+  %39 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not4.i = icmp eq i32 %39, 0, !dbg !18
+  %40 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.03.i, float %32, float %28) #4, !dbg !18
+  %41 = tail call float @llvm.nvvm.fma.rn.f(float %.03.i, float %32, float %28) #4, !dbg !18
+  %.04.i = select i1 %.not4.i, float %41, float %40, !dbg !18
+  %42 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not5.i = icmp eq i32 %42, 0, !dbg !18
+  %43 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.04.i, float %32, float %27) #4, !dbg !18
+  %44 = tail call float @llvm.nvvm.fma.rn.f(float %.04.i, float %32, float %27) #4, !dbg !18
+  %.05.i = select i1 %.not5.i, float %44, float %43, !dbg !18
+  %45 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not6.i = icmp eq i32 %45, 0, !dbg !18
+  %46 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.05.i, float %32, float %26) #4, !dbg !18
+  %47 = tail call float @llvm.nvvm.fma.rn.f(float %.05.i, float %32, float %26) #4, !dbg !18
+  %.06.i = select i1 %.not6.i, float %47, float %46, !dbg !18
+  %48 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not7.i = icmp eq i32 %48, 0, !dbg !18
+  %49 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.06.i, float %32, float %25) #4, !dbg !18
+  %50 = tail call float @llvm.nvvm.fma.rn.f(float %.06.i, float %32, float %25) #4, !dbg !18
+  %.07.i = select i1 %.not7.i, float %50, float %49, !dbg !18
+  %51 = fneg float %32, !dbg !18
+  %52 = select i1 %21, float %51, float %16, !dbg !18
+  %53 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not8.i = icmp eq i32 %53, 0, !dbg !18
+  %54 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.07.i, float %52, float %52) #4, !dbg !18
+  %55 = tail call float @llvm.nvvm.fma.rn.f(float %.07.i, float %52, float %52) #4, !dbg !18
+  %.08.i = select i1 %.not8.i, float %55, float %54, !dbg !18
+  br i1 %21, label %56, label %__nv_erff.exit, !dbg !18
+
+56:                                               ; preds = %__internal_fmad.exit.i
+  %57 = tail call float @llvm.nvvm.ex2.approx.ftz.f(float %.08.i) #4, !dbg !18
+  %58 = fsub float 1.000000e+00, %57, !dbg !18
+  %59 = bitcast float %58 to i32, !dbg !18
+  %60 = bitcast float %16 to i32, !dbg !18
+  %61 = and i32 %60, -2147483648, !dbg !18
+  %62 = or i32 %61, %59, !dbg !18
+  %63 = bitcast i32 %62 to float, !dbg !18
+  br label %__nv_erff.exit, !dbg !18
+
+__nv_erff.exit:                                   ; preds = %__internal_fmad.exit.i, %56
+  %r.0.i = phi float [ %63, %56 ], [ %.08.i, %__internal_fmad.exit.i ], !dbg !18
+  %64 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not.i1 = icmp eq i32 %64, 0, !dbg !18
+  %65 = tail call float @llvm.nvvm.fabs.ftz.f(float %17) #4, !dbg !18
+  %66 = tail call float @llvm.nvvm.fabs.f(float %17) #4, !dbg !18
+  %.0.i2 = select i1 %.not.i1, float %66, float %65, !dbg !18
+  %67 = fcmp oge float %.0.i2, 0x3FF00C1FC0000000, !dbg !18
+  br i1 %67, label %__nv_fabsf.exit1.i19, label %69, !dbg !18
+
+__nv_fabsf.exit1.i19:                             ; preds = %__nv_erff.exit
+  %68 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not1.i20 = icmp eq i32 %68, 0, !dbg !18
+  %.01.i21 = select i1 %.not1.i20, float %66, float %65, !dbg !18
+  br label %__internal_fmad.exit.i3, !dbg !18
+
+69:                                               ; preds = %__nv_erff.exit
+  %70 = fmul float %17, %17, !dbg !18
+  br label %__internal_fmad.exit.i3, !dbg !18
+
+__internal_fmad.exit.i3:                          ; preds = %69, %__nv_fabsf.exit1.i19
+  %71 = phi float [ 0x3FE41B0840000000, %__nv_fabsf.exit1.i19 ], [ 0x3FC06EBA60000000, %69 ], !dbg !18
+  %72 = phi float [ 0x3FED526FC0000000, %__nv_fabsf.exit1.i19 ], [ 0xBFD8127580000000, %69 ], !dbg !18
+  %73 = phi float [ 0x3FC39F20C0000000, %__nv_fabsf.exit1.i19 ], [ 0x3FBCE315E0000000, %69 ], !dbg !18
+  %74 = phi float [ 0xBFA1902C40000000, %__nv_fabsf.exit1.i19 ], [ 0xBF9B837CE0000000, %69 ], !dbg !18
+  %75 = phi float [ 0x3F75908160000000, %__nv_fabsf.exit1.i19 ], [ 0x3F755ABD40000000, %69 ], !dbg !18
+  %76 = phi float [ 0xBF3EAC1720000000, %__nv_fabsf.exit1.i19 ], [ 0xBF4AE9A400000000, %69 ], !dbg !18
+  %77 = phi float [ 0x3EF1394780000000, %__nv_fabsf.exit1.i19 ], [ 0x3F163D2D40000000, %69 ], !dbg !18
+  %78 = phi float [ %.01.i21, %__nv_fabsf.exit1.i19 ], [ %70, %69 ], !dbg !18
+  %79 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not2.i4 = icmp eq i32 %79, 0, !dbg !18
+  %80 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %77, float %78, float %76) #4, !dbg !18
+  %81 = tail call float @llvm.nvvm.fma.rn.f(float %77, float %78, float %76) #4, !dbg !18
+  %.02.i5 = select i1 %.not2.i4, float %81, float %80, !dbg !18
+  %82 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not3.i6 = icmp eq i32 %82, 0, !dbg !18
+  %83 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.02.i5, float %78, float %75) #4, !dbg !18
+  %84 = tail call float @llvm.nvvm.fma.rn.f(float %.02.i5, float %78, float %75) #4, !dbg !18
+  %.03.i7 = select i1 %.not3.i6, float %84, float %83, !dbg !18
+  %85 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not4.i8 = icmp eq i32 %85, 0, !dbg !18
+  %86 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.03.i7, float %78, float %74) #4, !dbg !18
+  %87 = tail call float @llvm.nvvm.fma.rn.f(float %.03.i7, float %78, float %74) #4, !dbg !18
+  %.04.i9 = select i1 %.not4.i8, float %87, float %86, !dbg !18
+  %88 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not5.i10 = icmp eq i32 %88, 0, !dbg !18
+  %89 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.04.i9, float %78, float %73) #4, !dbg !18
+  %90 = tail call float @llvm.nvvm.fma.rn.f(float %.04.i9, float %78, float %73) #4, !dbg !18
+  %.05.i11 = select i1 %.not5.i10, float %90, float %89, !dbg !18
+  %91 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not6.i12 = icmp eq i32 %91, 0, !dbg !18
+  %92 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.05.i11, float %78, float %72) #4, !dbg !18
+  %93 = tail call float @llvm.nvvm.fma.rn.f(float %.05.i11, float %78, float %72) #4, !dbg !18
+  %.06.i13 = select i1 %.not6.i12, float %93, float %92, !dbg !18
+  %94 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not7.i14 = icmp eq i32 %94, 0, !dbg !18
+  %95 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.06.i13, float %78, float %71) #4, !dbg !18
+  %96 = tail call float @llvm.nvvm.fma.rn.f(float %.06.i13, float %78, float %71) #4, !dbg !18
+  %.07.i15 = select i1 %.not7.i14, float %96, float %95, !dbg !18
+  %97 = fneg float %78, !dbg !18
+  %98 = select i1 %67, float %97, float %17, !dbg !18
+  %99 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !18
+  %.not8.i16 = icmp eq i32 %99, 0, !dbg !18
+  %100 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.07.i15, float %98, float %98) #4, !dbg !18
+  %101 = tail call float @llvm.nvvm.fma.rn.f(float %.07.i15, float %98, float %98) #4, !dbg !18
+  %.08.i17 = select i1 %.not8.i16, float %101, float %100, !dbg !18
+  br i1 %67, label %102, label %__nv_erff.exit22, !dbg !18
+
+102:                                              ; preds = %__internal_fmad.exit.i3
+  %103 = tail call float @llvm.nvvm.ex2.approx.ftz.f(float %.08.i17) #4, !dbg !18
+  %104 = fsub float 1.000000e+00, %103, !dbg !18
+  %105 = bitcast float %104 to i32, !dbg !18
+  %106 = bitcast float %17 to i32, !dbg !18
+  %107 = and i32 %106, -2147483648, !dbg !18
+  %108 = or i32 %107, %105, !dbg !18
+  %109 = bitcast i32 %108 to float, !dbg !18
+  br label %__nv_erff.exit22, !dbg !18
+
+__nv_erff.exit22:                                 ; preds = %__internal_fmad.exit.i3, %102
+  %r.0.i18 = phi float [ %109, %102 ], [ %.08.i17, %__internal_fmad.exit.i3 ], !dbg !18
+  %110 = fmul float %15, 5.000000e-01, !dbg !19
+  %111 = fmul float %14, 5.000000e-01, !dbg !19
+  %112 = fadd float %r.0.i, 1.000000e+00, !dbg !20
+  %113 = fadd float %r.0.i18, 1.000000e+00, !dbg !20
+  %114 = fmul float %111, %112, !dbg !21
+  %115 = fmul float %110, %113, !dbg !21
+  %116 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %114) #4, !dbg !22
+  %117 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %115) #4, !dbg !22
+  %118 = insertelement <2 x i16> undef, i16 %116, i64 0, !dbg !22
+  %119 = insertelement <2 x i16> %118, i16 %117, i64 1, !dbg !22
+  %120 = bitcast <2 x i16> %119 to i32, !dbg !22
+  tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %120, ptr addrspace(1) %10, i1 true) #4, !dbg !22
+  ret void, !dbg !23
+}
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare noundef i32 @llvm.nvvm.read.ptx.sreg.tid.x() #0
+
+; Function Attrs: alwaysinline nounwind
+define float @__nv_erff(float %a) local_unnamed_addr #1 {
+__nv_fabsf.exit:
+  %0 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not = icmp eq i32 %0, 0
+  %1 = tail call float @llvm.nvvm.fabs.ftz.f(float %a) #4
+  %2 = tail call float @llvm.nvvm.fabs.f(float %a) #4
+  %.0 = select i1 %.not, float %2, float %1
+  %3 = fcmp oge float %.0, 0x3FF00C1FC0000000
+  br i1 %3, label %__nv_fabsf.exit1, label %5
+
+__nv_fabsf.exit1:                                 ; preds = %__nv_fabsf.exit
+  %4 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not1 = icmp eq i32 %4, 0
+  %.01 = select i1 %.not1, float %2, float %1
+  br label %__internal_fmad.exit
+
+5:                                                ; preds = %__nv_fabsf.exit
+  %6 = fmul float %a, %a
+  br label %__internal_fmad.exit
+
+__internal_fmad.exit:                             ; preds = %5, %__nv_fabsf.exit1
+  %7 = phi float [ 0x3FE41B0840000000, %__nv_fabsf.exit1 ], [ 0x3FC06EBA60000000, %5 ]
+  %8 = phi float [ 0x3FED526FC0000000, %__nv_fabsf.exit1 ], [ 0xBFD8127580000000, %5 ]
+  %9 = phi float [ 0x3FC39F20C0000000, %__nv_fabsf.exit1 ], [ 0x3FBCE315E0000000, %5 ]
+  %10 = phi float [ 0xBFA1902C40000000, %__nv_fabsf.exit1 ], [ 0xBF9B837CE0000000, %5 ]
+  %11 = phi float [ 0x3F75908160000000, %__nv_fabsf.exit1 ], [ 0x3F755ABD40000000, %5 ]
+  %12 = phi float [ 0xBF3EAC1720000000, %__nv_fabsf.exit1 ], [ 0xBF4AE9A400000000, %5 ]
+  %13 = phi float [ 0x3EF1394780000000, %__nv_fabsf.exit1 ], [ 0x3F163D2D40000000, %5 ]
+  %14 = phi float [ %.01, %__nv_fabsf.exit1 ], [ %6, %5 ]
+  %15 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not2 = icmp eq i32 %15, 0
+  %16 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %13, float %14, float %12) #4
+  %17 = tail call float @llvm.nvvm.fma.rn.f(float %13, float %14, float %12) #4
+  %.02 = select i1 %.not2, float %17, float %16
+  %18 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not3 = icmp eq i32 %18, 0
+  %19 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.02, float %14, float %11) #4
+  %20 = tail call float @llvm.nvvm.fma.rn.f(float %.02, float %14, float %11) #4
+  %.03 = select i1 %.not3, float %20, float %19
+  %21 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not4 = icmp eq i32 %21, 0
+  %22 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.03, float %14, float %10) #4
+  %23 = tail call float @llvm.nvvm.fma.rn.f(float %.03, float %14, float %10) #4
+  %.04 = select i1 %.not4, float %23, float %22
+  %24 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not5 = icmp eq i32 %24, 0
+  %25 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.04, float %14, float %9) #4
+  %26 = tail call float @llvm.nvvm.fma.rn.f(float %.04, float %14, float %9) #4
+  %.05 = select i1 %.not5, float %26, float %25
+  %27 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not6 = icmp eq i32 %27, 0
+  %28 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.05, float %14, float %8) #4
+  %29 = tail call float @llvm.nvvm.fma.rn.f(float %.05, float %14, float %8) #4
+  %.06 = select i1 %.not6, float %29, float %28
+  %30 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not7 = icmp eq i32 %30, 0
+  %31 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.06, float %14, float %7) #4
+  %32 = tail call float @llvm.nvvm.fma.rn.f(float %.06, float %14, float %7) #4
+  %.07 = select i1 %.not7, float %32, float %31
+  %33 = fneg float %14
+  %34 = select i1 %3, float %33, float %a
+  %35 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4
+  %.not8 = icmp eq i32 %35, 0
+  %36 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.07, float %34, float %34) #4
+  %37 = tail call float @llvm.nvvm.fma.rn.f(float %.07, float %34, float %34) #4
+  %.08 = select i1 %.not8, float %37, float %36
+  br i1 %3, label %38, label %46
+
+38:                                               ; preds = %__internal_fmad.exit
+  %39 = tail call float @llvm.nvvm.ex2.approx.ftz.f(float %.08) #4
+  %40 = fsub float 1.000000e+00, %39
+  %41 = bitcast float %40 to i32
+  %42 = bitcast float %a to i32
+  %43 = and i32 %42, -2147483648
+  %44 = or i32 %43, %41
+  %45 = bitcast i32 %44 to float
+  br label %46
+
+46:                                               ; preds = %38, %__internal_fmad.exit
+  %r.0 = phi float [ %45, %38 ], [ %.08, %__internal_fmad.exit ]
+  ret float %r.0
+}
+
+declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #2
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.nvvm.fabs.ftz.f(float) #0
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.nvvm.fabs.f(float) #0
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.nvvm.fma.rn.ftz.f(float, float, float) #0
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.nvvm.fma.rn.f(float, float, float) #0
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none)
+declare float @llvm.nvvm.ex2.approx.ftz.f(float) #3
+
+attributes #0 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #1 = { alwaysinline nounwind "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #3 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) }
+attributes #4 = { nounwind }
+
+!llvm.module.flags = !{!0, !1}
+!llvm.dbg.cu = !{!2}
+!nvvm.annotations = !{!4, !5, !5, !4}
+!llvm.ident = !{!6}
+
+!0 = !{i32 2, !"Debug Info Version", i32 3}
+!1 = !{i32 4, !"nvvm-reflect-ftz", i32 1}
+!2 = distinct !DICompileUnit(language: DW_LANG_C, file: !3, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug)
+!3 = !DIFile(filename: "cafucwnmq4o436kwzkmrinerrnocxll7q6wsadcl726g6cradipo.py", directory: "/tmp/torchinductor_root/af")
+!4 = !{ptr @triton__0d1de, !"kernel", i32 1}
+!5 = !{ptr @triton__0d1de, !"maxntidx", i32 256}
+!6 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"}
+!7 = distinct !DISubprogram(name: "triton__0d1de", linkageName: "triton__0d1de", scope: !3, file: !3, line: 18, type: !8, scopeLine: 18, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
+!8 = !DISubroutineType(cc: DW_CC_normal, types: !9)
+!9 = !{}
+!10 = !DILocation(line: 21, column: 36, scope: !7)
+!11 = !DILocation(line: 20, column: 28, scope: !7)
+!12 = !DILocation(line: 20, column: 33, scope: !7)
+!13 = !DILocation(line: 21, column: 23, scope: !7)
+!14 = !DILocation(line: 24, column: 34, scope: !7)
+!15 = !DILocation(line: 24, column: 39, scope: !7)
+!16 = !DILocation(line: 24, column: 48, scope: !7)
+!17 = !DILocation(line: 29, column: 18, scope: !7)
+!18 = !DILocation(line: 30, column: 23, scope: !7)
+!19 = !DILocation(line: 27, column: 18, scope: !7)
+!20 = !DILocation(line: 32, column: 18, scope: !7)
+!21 = !DILocation(line: 33, column: 18, scope: !7)
+!22 = !DILocation(line: 35, column: 40, scope: !7)
+!23 = !DILocation(line: 35, column: 4, scope: !7)

.triton/dump/1e922bbbab749da355e4bad9c6b245e6/triton_.ptx

@@ -0,0 +1,446 @@
+//
+// Generated by LLVM NVPTX Back-End
+//
+
+.version 8.2
+.target sm_89
+.address_size 64
+
+	// .globl	triton__0d1de
+.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90, 0};
+
+.visible .entry triton__0d1de(
+	.param .u64 triton__0d1de_param_0,
+	.param .u32 triton__0d1de_param_1
+)
+.maxntid 256, 1, 1
+{
+	.reg .pred 	%p<9>;
+	.reg .b16 	%rs<5>;
+	.reg .b32 	%r<22>;
+	.reg .f32 	%f<113>;
+	.reg .b64 	%rd<6>;
+	.loc	1 18 0
+$L__func_begin0:
+	.loc	1 18 0
+
+	ld.param.u64 	%rd3, [triton__0d1de_param_0];
+$L__tmp0:
+	.loc	1 21 36
+	mov.u32 	%r5, %tid.x;
+	shl.b32 	%r6, %r5, 1;
+	and.b32  	%r7, %r6, 510;
+	.loc	1 20 28
+	mov.u32 %r1, %ctaid.x;
+	.loc	1 20 33
+	shl.b32 	%r8, %r1, 9;
+	.loc	1 21 23
+	or.b32  	%r9, %r8, %r7;
+	.loc	1 24 34
+	mul.wide.s32 	%rd4, %r9, 2;
+	add.s64 	%rd5, %rd3, %rd4;
+	mov.pred 	%p1, -1;
+	.loc	1 24 39
+	mov.u32 %r2, 0x0;
+	@%p1 ld.global.b32 { %r2 }, [ %rd5 + 0 ];
+	cvt.u16.u32 	%rs1, %r2;
+	{ .reg .b16 tmp; mov.b32 {tmp, %rs2}, %r2; }
+	.loc	1 24 48
+	cvt.f32.bf16 %r3, %rs1;
+	mov.b32 	%f1, %r3;
+	cvt.f32.bf16 %r4, %rs2;
+	mov.b32 	%f2, %r4;
+	.loc	1 29 18
+	mul.f32 	%f3, %f1, 0f3F3504F3;
+	.loc	1 30 23
+	abs.ftz.f32 	%f5, %f3;
+	setp.ge.f32 	%p2, %f5, 0f3F8060FE;
+	mov.f32 	%f101, 0f3789CA3C;
+	mov.f32 	%f100, 0fB9F560B9;
+	mov.f32 	%f99, 0f3BAC840B;
+	mov.f32 	%f98, 0fBD0C8162;
+	mov.f32 	%f97, 0f3E1CF906;
+	mov.f32 	%f96, 0f3F6A937E;
+	mov.f32 	%f95, 0f3F20D842;
+	mov.f32 	%f102, %f5;
+	@%p2 bra 	$L__BB0_2;
+	.loc	1 0 23
+	mov.f32 	%f101, 0f38B1E96A;
+	mov.f32 	%f100, 0fBA574D20;
+	mov.f32 	%f99, 0f3BAAD5EA;
+	mov.f32 	%f98, 0fBCDC1BE7;
+	mov.f32 	%f97, 0f3DE718AF;
+	mov.f32 	%f96, 0fBEC093AC;
+	mov.f32 	%f95, 0f3E0375D3;
+	.loc	1 30 23
+	mul.f32 	%f102, %f3, %f3;
+$L__BB0_2:
+	.loc	1 0 0
+	mul.f32 	%f4, %f2, 0f3F3504F3;
+	.loc	1 30 23
+	setp.ltu.f32 	%p3, %f5, 0f3F8060FE;
+	fma.rn.ftz.f32 	%f45, %f101, %f102, %f100;
+	fma.rn.ftz.f32 	%f46, %f45, %f102, %f99;
+	fma.rn.ftz.f32 	%f47, %f46, %f102, %f98;
+	fma.rn.ftz.f32 	%f48, %f47, %f102, %f97;
+	fma.rn.ftz.f32 	%f49, %f48, %f102, %f96;
+	fma.rn.ftz.f32 	%f50, %f49, %f102, %f95;
+	neg.f32 	%f51, %f102;
+	selp.f32 	%f52, %f51, %f3, %p2;
+	fma.rn.ftz.f32 	%f103, %f50, %f52, %f52;
+	mov.f32 	%f94, 0f3F800000;
+	@%p3 bra 	$L__BB0_4;
+	ex2.approx.ftz.f32 	%f53, %f103;
+	sub.f32 	%f55, %f94, %f53;
+	mov.b32 	%r10, %f55;
+	mov.b32 	%r11, %f3;
+	and.b32  	%r12, %r11, -2147483648;
+	or.b32  	%r13, %r12, %r10;
+	mov.b32 	%f103, %r13;
+$L__BB0_4:
+	abs.ftz.f32 	%f18, %f4;
+	setp.ge.f32 	%p5, %f18, 0f3F8060FE;
+	mov.f32 	%f110, 0f3789CA3C;
+	mov.f32 	%f109, 0fB9F560B9;
+	mov.f32 	%f108, 0f3BAC840B;
+	mov.f32 	%f107, 0fBD0C8162;
+	mov.f32 	%f106, 0f3E1CF906;
+	mov.f32 	%f105, 0f3F6A937E;
+	mov.f32 	%f104, 0f3F20D842;
+	mov.f32 	%f111, %f18;
+	@%p5 bra 	$L__BB0_6;
+	mul.f32 	%f111, %f4, %f4;
+	mov.f32 	%f110, 0f38B1E96A;
+	mov.f32 	%f109, 0fBA574D20;
+	mov.f32 	%f108, 0f3BAAD5EA;
+	mov.f32 	%f107, 0fBCDC1BE7;
+	mov.f32 	%f106, 0f3DE718AF;
+	mov.f32 	%f105, 0fBEC093AC;
+	mov.f32 	%f104, 0f3E0375D3;
+$L__BB0_6:
+	setp.ltu.f32 	%p6, %f18, 0f3F8060FE;
+	fma.rn.ftz.f32 	%f70, %f110, %f111, %f109;
+	fma.rn.ftz.f32 	%f71, %f70, %f111, %f108;
+	fma.rn.ftz.f32 	%f72, %f71, %f111, %f107;
+	fma.rn.ftz.f32 	%f73, %f72, %f111, %f106;
+	fma.rn.ftz.f32 	%f74, %f73, %f111, %f105;
+	fma.rn.ftz.f32 	%f75, %f74, %f111, %f104;
+	neg.f32 	%f76, %f111;
+	selp.f32 	%f77, %f76, %f4, %p5;
+	fma.rn.ftz.f32 	%f112, %f75, %f77, %f77;
+	@%p6 bra 	$L__BB0_8;
+	ex2.approx.ftz.f32 	%f78, %f112;
+	sub.f32 	%f80, %f94, %f78;
+	mov.b32 	%r14, %f80;
+	mov.b32 	%r15, %f4;
+	and.b32  	%r16, %r15, -2147483648;
+	or.b32  	%r17, %r16, %r14;
+	mov.b32 	%f112, %r17;
+$L__BB0_8:
+	.loc	1 27 18
+	mul.f32 	%f81, %f2, 0f3F000000;
+	mul.f32 	%f82, %f1, 0f3F000000;
+	.loc	1 32 18
+	add.f32 	%f83, %f103, 0f3F800000;
+	add.f32 	%f84, %f112, 0f3F800000;
+	.loc	1 33 18
+	mul.f32 	%f85, %f82, %f83;
+	mul.f32 	%f86, %f81, %f84;
+	.loc	1 35 40
+	mov.b32 	%r18, %f85;
+	cvt.rn.bf16.f32 %rs3, %r18;
+	mov.b32 	%r19, %f86;
+	cvt.rn.bf16.f32 %rs4, %r19;
+	mov.b32 	%r21, {%rs3, %rs4};
+	@%p1 st.global.b32 [ %rd5 + 0 ], { %r21 };
+	.loc	1 35 4
+	ret;
+$L__tmp1:
+$L__func_end0:
+
+}
+	// .globl	__nv_erff
+.visible .func  (.param .b32 func_retval0) __nv_erff(
+	.param .b32 __nv_erff_param_0
+)
+{
+	.reg .pred 	%p<4>;
+	.reg .b32 	%r<5>;
+	.reg .f32 	%f<49>;
+$L__func_begin1:
+
+	ld.param.f32 	%f14, [__nv_erff_param_0];
+	abs.ftz.f32 	%f1, %f14;
+	setp.ge.f32 	%p1, %f1, 0f3F8060FE;
+	mov.f32 	%f46, 0f3789CA3C;
+	mov.f32 	%f45, 0fB9F560B9;
+	mov.f32 	%f44, 0f3BAC840B;
+	mov.f32 	%f43, 0fBD0C8162;
+	mov.f32 	%f42, 0f3E1CF906;
+	mov.f32 	%f41, 0f3F6A937E;
+	mov.f32 	%f40, 0f3F20D842;
+	mov.f32 	%f47, %f1;
+	@%p1 bra 	$L__BB1_2;
+	mul.f32 	%f47, %f14, %f14;
+	mov.f32 	%f46, 0f38B1E96A;
+	mov.f32 	%f45, 0fBA574D20;
+	mov.f32 	%f44, 0f3BAAD5EA;
+	mov.f32 	%f43, 0fBCDC1BE7;
+	mov.f32 	%f42, 0f3DE718AF;
+	mov.f32 	%f41, 0fBEC093AC;
+	mov.f32 	%f40, 0f3E0375D3;
+$L__BB1_2:
+	setp.ltu.f32 	%p2, %f1, 0f3F8060FE;
+	fma.rn.ftz.f32 	%f29, %f46, %f47, %f45;
+	fma.rn.ftz.f32 	%f30, %f29, %f47, %f44;
+	fma.rn.ftz.f32 	%f31, %f30, %f47, %f43;
+	fma.rn.ftz.f32 	%f32, %f31, %f47, %f42;
+	fma.rn.ftz.f32 	%f33, %f32, %f47, %f41;
+	fma.rn.ftz.f32 	%f34, %f33, %f47, %f40;
+	neg.f32 	%f35, %f47;
+	selp.f32 	%f36, %f35, %f14, %p1;
+	fma.rn.ftz.f32 	%f48, %f34, %f36, %f36;
+	@%p2 bra 	$L__BB1_4;
+	ex2.approx.ftz.f32 	%f37, %f48;
+	mov.f32 	%f38, 0f3F800000;
+	sub.f32 	%f39, %f38, %f37;
+	mov.b32 	%r1, %f39;
+	mov.b32 	%r2, %f14;
+	and.b32  	%r3, %r2, -2147483648;
+	or.b32  	%r4, %r3, %r1;
+	mov.b32 	%f48, %r4;
+$L__BB1_4:
+	st.param.f32 	[func_retval0+0], %f48;
+	ret;
+$L__func_end1:
+
+}
+	.file	1 "/tmp/torchinductor_root/af/cafucwnmq4o436kwzkmrinerrnocxll7q6wsadcl726g6cradipo.py"
+	.section	.debug_abbrev
+	{
+.b8 1
+.b8 17
+.b8 1
+.b8 37
+.b8 8
+.b8 19
+.b8 5
+.b8 3
+.b8 8
+.b8 16
+.b8 6
+.b8 27
+.b8 8
+.b8 180
+.b8 66
+.b8 12
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+.b8 0
+.b8 2
+.b8 46
+.b8 0
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 64
+.b8 10
+.b8 135
+.b8 64
+.b8 8
+.b8 3
+.b8 8
+.b8 58
+.b8 11
+.b8 59
+.b8 11
+.b8 63
+.b8 12
+.b8 0
+.b8 0
+.b8 0
+	}
+	.section	.debug_info
+	{
+.b32 172
+.b8 2
+.b8 0
+.b32 .debug_abbrev
+.b8 8
+.b8 1
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 0
+.b8 2
+.b8 0
+.b8 99
+.b8 97
+.b8 102
+.b8 117
+.b8 99
+.b8 119
+.b8 110
+.b8 109
+.b8 113
+.b8 52
+.b8 111
+.b8 52
+.b8 51
+.b8 54
+.b8 107
+.b8 119
+.b8 122
+.b8 107
+.b8 109
+.b8 114
+.b8 105
+.b8 110
+.b8 101
+.b8 114
+.b8 114
+.b8 110
+.b8 111
+.b8 99
+.b8 120
+.b8 108
+.b8 108
+.b8 55
+.b8 113
+.b8 54
+.b8 119
+.b8 115
+.b8 97
+.b8 100
+.b8 99
+.b8 108
+.b8 55
+.b8 50
+.b8 54
+.b8 103
+.b8 54
+.b8 99
+.b8 114
+.b8 97
+.b8 100
+.b8 105
+.b8 112
+.b8 111
+.b8 46
+.b8 112
+.b8 121
+.b8 0
+.b32 .debug_line
+.b8 47
+.b8 116
+.b8 109
+.b8 112
+.b8 47
+.b8 116
+.b8 111
+.b8 114
+.b8 99
+.b8 104
+.b8 105
+.b8 110
+.b8 100
+.b8 117
+.b8 99
+.b8 116
+.b8 111
+.b8 114
+.b8 95
+.b8 114
+.b8 111
+.b8 111
+.b8 116
+.b8 47
+.b8 97
+.b8 102
+.b8 0
+.b8 1
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 2
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 1
+.b8 156
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 101
+.b8 0
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 101
+.b8 0
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+	}
+	.section	.debug_pubnames
+	{
+.b32 $L__pubNames_end0-$L__pubNames_start0
+$L__pubNames_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 176
+.b32 125
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 101
+.b8 0
+.b32 0
+$L__pubNames_end0:
+	}
+	.section	.debug_pubtypes
+	{
+.b32 $L__pubTypes_end0-$L__pubTypes_start0
+$L__pubTypes_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 176
+.b32 0
+$L__pubTypes_end0:
+	}
+	.section	.debug_loc	{	}

.triton/dump/1e922bbbab749da355e4bad9c6b245e6/triton_.ttgir

@@ -0,0 +1,26 @@
+#blocked = #triton_gpu.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [8], order = [0], CTAsPerCGA = [1], CTASplitNum = [1], CTAOrder = [0]}>
+module attributes {"triton_gpu.compute-capability" = 89 : i32, "triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-warp" = 32 : i32} {
+  tt.func public @triton__0d1de(%arg0: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg1: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %cst = arith.constant dense<1.000000e+00> : tensor<512xf32, #blocked>
+    %cst_0 = arith.constant dense<0.707106769> : tensor<512xf32, #blocked>
+    %cst_1 = arith.constant dense<5.000000e-01> : tensor<512xf32, #blocked>
+    %c512_i32 = arith.constant 512 : i32
+    %0 = tt.get_program_id x : i32
+    %1 = arith.muli %0, %c512_i32 : i32
+    %2 = tt.make_range {end = 512 : i32, start = 0 : i32} : tensor<512xi32, #blocked>
+    %3 = tt.splat %1 : (i32) -> tensor<512xi32, #blocked>
+    %4 = arith.addi %3, %2 : tensor<512xi32, #blocked>
+    %5 = tt.splat %arg0 : (!tt.ptr<bf16, 1>) -> tensor<512x!tt.ptr<bf16, 1>, #blocked>
+    %6 = tt.addptr %5, %4 : tensor<512x!tt.ptr<bf16, 1>, #blocked>, tensor<512xi32, #blocked>
+    %7 = tt.load %6 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<512xbf16, #blocked>
+    %8 = arith.extf %7 : tensor<512xbf16, #blocked> to tensor<512xf32, #blocked>
+    %9 = arith.mulf %8, %cst_1 : tensor<512xf32, #blocked>
+    %10 = arith.mulf %8, %cst_0 : tensor<512xf32, #blocked>
+    %11 = tt.extern_elementwise %10 {libname = "libdevice", libpath = "/usr/local/lib/python3.10/dist-packages/triton/language/../third_party/cuda/lib/libdevice.10.bc", pure = true, symbol = "__nv_erff"} : (tensor<512xf32, #blocked>) -> tensor<512xf32, #blocked>
+    %12 = arith.addf %11, %cst : tensor<512xf32, #blocked>
+    %13 = arith.mulf %9, %12 : tensor<512xf32, #blocked>
+    %14 = arith.truncf %13 : tensor<512xf32, #blocked> to tensor<512xbf16, #blocked>
+    tt.store %6, %14 {cache = 1 : i32, evict = 1 : i32} : tensor<512xbf16, #blocked>
+    tt.return
+  }
+}

.triton/dump/3cd3b6d7993c56f7d0340d40c84f737c/triton_.ptx

@@ -0,0 +1,809 @@
+//
+// Generated by LLVM NVPTX Back-End
+//
+
+.version 8.2
+.target sm_89
+.address_size 64
+
+	// .globl	triton__0d1d2d3d4d5d6de7de
+.extern .func __assertfail
+(
+	.param .b64 __assertfail_param_0,
+	.param .b64 __assertfail_param_1,
+	.param .b32 __assertfail_param_2,
+	.param .b64 __assertfail_param_3,
+	.param .b64 __assertfail_param_4
+)
+;
+.global .align 1 .b8 assertFunc_1[25] = {95, 99, 97, 108, 108, 95, 119, 105, 116, 104, 95, 102, 114, 97, 109, 101, 115, 95, 114, 101, 109, 111, 118, 101, 100};
+.global .align 1 .b8 assertFile_1[38] = {60, 102, 114, 111, 122, 101, 110, 32, 105, 109, 112, 111, 114, 116, 108, 105, 98, 46, 95, 98, 111, 111, 116, 115, 116, 114, 97, 112, 95, 101, 120, 116, 101, 114, 110, 97, 108, 62};
+.global .align 1 .b8 assertMessage_1[39] = {105, 110, 100, 101, 120, 32, 111, 117, 116, 32, 111, 102, 32, 98, 111, 117, 110, 100, 115, 58, 32, 48, 32, 60, 61, 32, 116, 109, 112, 49, 54, 32, 60, 32, 53, 48, 50, 53, 55};
+.global .align 1 .b8 assertFunc_0[25] = {95, 99, 97, 108, 108, 95, 119, 105, 116, 104, 95, 102, 114, 97, 109, 101, 115, 95, 114, 101, 109, 111, 118, 101, 100};
+.global .align 1 .b8 assertFile_0[38] = {60, 102, 114, 111, 122, 101, 110, 32, 105, 109, 112, 111, 114, 116, 108, 105, 98, 46, 95, 98, 111, 111, 116, 115, 116, 114, 97, 112, 95, 101, 120, 116, 101, 114, 110, 97, 108, 62};
+.global .align 1 .b8 assertMessage_0[38] = {105, 110, 100, 101, 120, 32, 111, 117, 116, 32, 111, 102, 32, 98, 111, 117, 110, 100, 115, 58, 32, 48, 32, 60, 61, 32, 116, 109, 112, 51, 32, 60, 32, 53, 48, 50, 53, 55};
+.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90, 0};
+
+.visible .entry triton__0d1d2d3d4d5d6de7de(
+	.param .u64 triton__0d1d2d3d4d5d6de7de_param_0,
+	.param .u64 triton__0d1d2d3d4d5d6de7de_param_1,
+	.param .u64 triton__0d1d2d3d4d5d6de7de_param_2,
+	.param .u64 triton__0d1d2d3d4d5d6de7de_param_3,
+	.param .u64 triton__0d1d2d3d4d5d6de7de_param_4,
+	.param .u64 triton__0d1d2d3d4d5d6de7de_param_5,
+	.param .u32 triton__0d1d2d3d4d5d6de7de_param_6,
+	.param .u32 triton__0d1d2d3d4d5d6de7de_param_7
+)
+.maxntid 256, 1, 1
+{
+	.reg .pred 	%p<33>;
+	.reg .b16 	%rs<13>;
+	.reg .b32 	%r<93>;
+	.reg .f32 	%f<79>;
+	.reg .b64 	%rd<92>;
+	.loc	1 18 0
+$L__func_begin0:
+	.loc	1 18 0
+
+	ld.param.u64 	%rd37, [triton__0d1d2d3d4d5d6de7de_param_4];
+	ld.param.u64 	%rd36, [triton__0d1d2d3d4d5d6de7de_param_3];
+	ld.param.u64 	%rd35, [triton__0d1d2d3d4d5d6de7de_param_2];
+	ld.param.u64 	%rd34, [triton__0d1d2d3d4d5d6de7de_param_1];
+	ld.param.u64 	%rd43, [triton__0d1d2d3d4d5d6de7de_param_0];
+$L__tmp0:
+	.loc	1 22 44
+	mov.u32 	%r1, %tid.x;
+	bfe.u32 	%r2, %r1, 2, 6;
+	and.b32  	%r16, %r1, 63;
+	.loc	1 24 33
+	and.b32  	%r3, %r1, 3;
+	.loc	1 21 28
+	mov.u32 %r15, %ctaid.x;
+	.loc	1 21 33
+	shl.b32 	%r17, %r15, 6;
+	.loc	1 22 23
+	or.b32  	%r18, %r17, %r2;
+	or.b32  	%r19, %r17, %r16;
+	.loc	1 26 30
+	mul.wide.s32 	%rd44, %r18, 8;
+	add.s64 	%rd40, %rd43, %rd44;
+	mul.wide.s32 	%rd45, %r19, 8;
+	add.s64 	%rd42, %rd43, %rd45;
+	mov.pred 	%p13, -1;
+	.loc	1 26 35
+	mov.u64 %rd39, 0x0;
+	@%p13 ld.global.L1::evict_last.b64 { %rd39 }, [ %rd40 + 0 ];
+	mov.u64 %rd41, 0x0;
+	@%p13 ld.global.L1::evict_last.b64 { %rd41 }, [ %rd42 + 0 ];
+	.loc	1 27 18
+	bfe.s32 	%r20, %r15, 25, 1;
+	shr.u32 	%r21, %r20, 23;
+	add.s32 	%r22, %r18, %r21;
+	and.b32  	%r23, %r22, 16776704;
+	sub.s32 	%r24, %r18, %r23;
+	.loc	1 35 44
+	shl.b32 	%r5, %r24, 8;
+	.loc	1 37 22
+	add.s64 	%rd46, %rd41, 50257;
+	.loc	1 38 22
+	setp.lt.s64 	%p3, %rd39, 0;
+	setp.lt.s64 	%p4, %rd41, 0;
+	.loc	1 39 36
+	selp.b64 	%rd47, %rd46, %rd41, %p4;
+	.loc	1 40 40
+	setp.gt.u64 	%p5, %rd47, 50256;
+	.loc	1 41 44
+	shl.b64 	%rd48, %rd39, 8;
+	add.s64 	%rd49, %rd48, 12865792;
+	selp.b64 	%rd2, %rd49, %rd48, %p3;
+	mov.u16 	%rs12, 0;
+	mov.b32 	%r76, 0;
+	mov.b32 	%r88, 883;
+	mov.u64 	%rd81, 1;
+	.loc	1 40 55
+	@%p5 bra 	$L__BB0_3;
+	bra.uni 	$L__BB0_1;
+$L__BB0_3:
+	.loc	1 31 36
+	shl.b64 	%rd55, %rd2, 2;
+	mul.wide.u32 	%rd88, %r3, 4;
+	add.s64 	%rd87, %rd55, %rd88;
+	add.s64 	%rd83, %rd34, %rd87;
+	shl.b32 	%r42, %r15, 14;
+	shl.b32 	%r43, %r2, 8;
+	or.b32  	%r44, %r42, %r43;
+	or.b32  	%r91, %r44, %r3;
+	add.s32 	%r45, %r5, %r3;
+	mul.wide.s32 	%rd86, %r45, 4;
+	add.s64 	%rd82, %rd35, %rd86;
+	mov.f32 	%f78, 0f00000000;
+	mov.b32 	%r89, -4;
+	mov.f32 	%f77, %f78;
+	mov.f32 	%f76, %f78;
+$L__BB0_4:
+	.loc	1 35 50
+	mov.u32 %r46, 0x0;
+	@%p13 ld.global.L1::evict_last.b32 { %r46 }, [ %rd82 + 0 ];
+	@!%p13 mov.u32 %r46, %r76;
+	mov.b32 	%f31, %r46;
+	.loc	1 31 36
+	add.s32 	%r89, %r89, 4;
+	.loc	1 36 34
+	add.s32 	%r54, %r89, %r91;
+	mul.wide.s32 	%rd59, %r54, 2;
+	add.s64 	%rd57, %rd36, %rd59;
+	.loc	1 36 50
+	mov.u16 %rs4, 0x0;
+	@%p13 ld.global.L1::evict_last.b16 { %rs4 }, [ %rd57 + 0 ];
+	@!%p13 mov.u16 %rs4, %rs12;
+	.loc	1 36 101
+	cvt.f32.bf16 %r48, %rs4;
+	mov.b32 	%f32, %r48;
+	.loc	1 40 55
+	mov.u64 	%rd60, assertMessage_0;
+	cvta.global.u64 	%rd61, %rd60;
+	mov.u64 	%rd62, assertFile_0;
+	cvta.global.u64 	%rd63, %rd62;
+	mov.u64 	%rd64, assertFunc_0;
+	cvta.global.u64 	%rd65, %rd64;
+	{ // callseq 10, 0
+	.reg .b32 temp_param_reg;
+	.param .b64 param0;
+	st.param.b64 	[param0+0], %rd61;
+	.param .b64 param1;
+	st.param.b64 	[param1+0], %rd63;
+	.param .b32 param2;
+	st.param.b32 	[param2+0], %r88;
+	.param .b64 param3;
+	st.param.b64 	[param3+0], %rd65;
+	.param .b64 param4;
+	st.param.b64 	[param4+0], %rd81;
+	call.uni 
+	__assertfail, 
+	(
+	param0, 
+	param1, 
+	param2, 
+	param3, 
+	param4
+	);
+	} // callseq 10
+	.loc	1 41 52
+	mov.u32 %r49, 0x0;
+	@%p13 ld.global.L1::evict_last.b32 { %r49 }, [ %rd83 + 0 ];
+	@!%p13 mov.u32 %r49, %r76;
+	mov.b32 	%f33, %r49;
+	.loc	1 42 22
+	add.f32 	%f34, %f31, %f33;
+	.loc	1 44 22
+	add.f32 	%f35, %f32, %f34;
+$L__tmp1:
+	.loc	2 96 20
+	sub.f32 	%f36, %f35, %f76;
+	.loc	2 97 26
+	add.f32 	%f78, %f78, 0f3F800000;
+	.loc	2 98 30
+	mov.b32 	%r52, %f36;
+	mov.b32 	%r53, %f78;
+	div.full.f32 %r51, %r52, %r53;
+	mov.b32 	%f37, %r51;
+	.loc	2 98 22
+	add.f32 	%f76, %f76, %f37;
+	.loc	2 101 30
+	sub.f32 	%f38, %f35, %f76;
+$L__tmp2:
+	.loc	1 50 50
+	fma.rn.f32 	%f77, %f36, %f38, %f77;
+	.loc	1 31 36
+	add.s64 	%rd83, %rd83, 16;
+	add.s64 	%rd82, %rd82, 16;
+	setp.lt.u32 	%p19, %r89, 252;
+	@%p19 bra 	$L__BB0_4;
+	bra.uni 	$L__BB0_5;
+$L__BB0_1:
+	.loc	1 0 36
+	mov.b32 	%r90, -4;
+	.loc	1 31 36
+	shl.b64 	%rd50, %rd2, 2;
+	mul.wide.u32 	%rd88, %r3, 4;
+	add.s64 	%rd87, %rd50, %rd88;
+	add.s64 	%rd85, %rd34, %rd87;
+	shl.b32 	%r27, %r15, 14;
+	shl.b32 	%r28, %r2, 8;
+	or.b32  	%r29, %r27, %r28;
+	or.b32  	%r91, %r29, %r3;
+	add.s32 	%r30, %r5, %r3;
+	mul.wide.s32 	%rd86, %r30, 4;
+	add.s64 	%rd84, %rd35, %rd86;
+	mov.f32 	%f78, 0f00000000;
+	mov.f32 	%f77, %f78;
+	mov.f32 	%f76, %f78;
+$L__BB0_2:
+	.loc	1 35 50
+	mov.u32 %r31, 0x0;
+	@%p13 ld.global.L1::evict_last.b32 { %r31 }, [ %rd84 + 0 ];
+	@!%p13 mov.u32 %r31, %r76;
+	mov.b32 	%f22, %r31;
+	.loc	1 31 36
+	add.s32 	%r90, %r90, 4;
+	.loc	1 36 34
+	add.s32 	%r39, %r90, %r91;
+	mul.wide.s32 	%rd54, %r39, 2;
+	add.s64 	%rd52, %rd36, %rd54;
+	.loc	1 36 50
+	mov.u16 %rs1, 0x0;
+	@%p13 ld.global.L1::evict_last.b16 { %rs1 }, [ %rd52 + 0 ];
+	@!%p13 mov.u16 %rs1, %rs12;
+	.loc	1 36 101
+	cvt.f32.bf16 %r33, %rs1;
+	mov.b32 	%f23, %r33;
+	.loc	1 41 52
+	mov.u32 %r34, 0x0;
+	@%p13 ld.global.L1::evict_last.b32 { %r34 }, [ %rd85 + 0 ];
+	@!%p13 mov.u32 %r34, %r76;
+	mov.b32 	%f24, %r34;
+	.loc	1 42 22
+	add.f32 	%f25, %f22, %f24;
+	.loc	1 44 22
+	add.f32 	%f26, %f23, %f25;
+$L__tmp3:
+	.loc	2 96 20
+	sub.f32 	%f27, %f26, %f76;
+	.loc	2 97 26
+	add.f32 	%f78, %f78, 0f3F800000;
+	.loc	2 98 30
+	mov.b32 	%r37, %f27;
+	mov.b32 	%r38, %f78;
+	div.full.f32 %r36, %r37, %r38;
+	mov.b32 	%f28, %r36;
+	.loc	2 98 22
+	add.f32 	%f76, %f76, %f28;
+	.loc	2 101 30
+	sub.f32 	%f29, %f26, %f76;
+$L__tmp4:
+	.loc	1 50 50
+	fma.rn.f32 	%f77, %f27, %f29, %f77;
+	.loc	1 31 36
+	add.s64 	%rd85, %rd85, 16;
+	add.s64 	%rd84, %rd84, 16;
+	setp.lt.u32 	%p12, %r90, 252;
+	@%p12 bra 	$L__BB0_2;
+$L__BB0_5:
+	.loc	1 0 36
+	ld.param.u64 	%rd38, [triton__0d1d2d3d4d5d6de7de_param_5];
+$L__tmp5:
+	.loc	2 120 46
+	mov.b32 	%r66, %f76;
+	shfl.sync.bfly.b32	%r67, %r66, 2, 31, -1;
+	mov.b32 	%f39, %r67;
+	mov.b32 	%r68, %f77;
+	shfl.sync.bfly.b32	%r69, %r68, 2, 31, -1;
+	mov.b32 	%f40, %r69;
+	mov.b32 	%r70, %f78;
+	shfl.sync.bfly.b32	%r57, %r70, 2, 31, -1;
+	mov.b32 	%f41, %r57;
+$L__tmp6:
+	.loc	2 108 21
+	sub.f32 	%f42, %f39, %f76;
+	.loc	2 109 28
+	add.f32 	%f43, %f78, %f41;
+	.loc	2 110 39
+	setp.eq.f32 	%p20, %f43, 0f00000000;
+	.loc	2 110 60
+	mov.b32 	%r58, %f43;
+	div.full.f32 %r56, %r57, %r58;
+	mov.b32 	%f44, %r56;
+	.loc	2 110 49
+	selp.f32 	%f45, 0f00000000, %f44, %p20;
+	.loc	2 112 17
+	fma.rn.f32 	%f46, %f42, %f45, %f76;
+	.loc	2 113 15
+	add.f32 	%f47, %f77, %f40;
+	.loc	2 113 30
+	mul.f32 	%f48, %f42, %f42;
+	.loc	2 113 38
+	mul.f32 	%f49, %f78, %f48;
+	.loc	2 113 22
+	fma.rn.f32 	%f50, %f49, %f45, %f47;
+$L__tmp7:
+	.loc	2 120 46
+	mov.b32 	%r71, %f46;
+	shfl.sync.bfly.b32	%r72, %r71, 1, 31, -1;
+	mov.b32 	%f51, %r72;
+	mov.b32 	%r73, %f50;
+	shfl.sync.bfly.b32	%r74, %r73, 1, 31, -1;
+	mov.b32 	%f52, %r74;
+	shfl.sync.bfly.b32	%r60, %r58, 1, 31, -1;
+	mov.b32 	%f53, %r60;
+$L__tmp8:
+	.loc	2 108 21
+	sub.f32 	%f54, %f51, %f46;
+	.loc	2 109 28
+	add.f32 	%f55, %f43, %f53;
+	.loc	2 110 39
+	setp.eq.f32 	%p21, %f55, 0f00000000;
+	.loc	2 110 60
+	mov.b32 	%r61, %f55;
+	div.full.f32 %r59, %r60, %r61;
+	mov.b32 	%f56, %r59;
+	.loc	2 110 49
+	selp.f32 	%f57, 0f00000000, %f56, %p21;
+	.loc	2 112 17
+	fma.rn.f32 	%f16, %f54, %f57, %f46;
+	.loc	2 113 15
+	add.f32 	%f58, %f50, %f52;
+	.loc	2 113 30
+	mul.f32 	%f59, %f54, %f54;
+	.loc	2 113 38
+	mul.f32 	%f60, %f43, %f59;
+	.loc	2 113 22
+	fma.rn.f32 	%f61, %f57, %f60, %f58;
+$L__tmp9:
+	.loc	1 75 24
+	mov.b32 	%r63, %f61;
+	mov.b32 	%r64, 1132462080;
+	div.full.f32 %r62, %r63, %r64;
+	mov.b32 	%f62, %r62;
+	.loc	1 77 24
+	add.f32 	%f17, %f62, 0f3727C5AC;
+	.loc	1 58 36
+	add.s64 	%rd91, %rd34, %rd87;
+	add.s64 	%rd90, %rd37, %rd88;
+	add.s64 	%rd89, %rd35, %rd86;
+	mov.b32 	%r92, -4;
+	setp.lt.u64 	%p28, %rd47, 50257;
+	rsqrt.approx.ftz.f32 	%f67, %f17;
+	bra.uni 	$L__BB0_6;
+$L__BB0_8:
+	.loc	1 0 0
+	mov.b32 	%f18, %r75;
+	cvt.s64.s32 	%rd30, %r81;
+	cvt.f32.bf16 %r77, %rs7;
+	mov.b32 	%f19, %r77;
+	mov.b32 	%f20, %r78;
+	.loc	1 69 54
+	mov.u32 %r83, 0x0;
+	@%p13 ld.global.L1::evict_first.b32 { %r83 }, [ %rd91 + 0 ];
+	@!%p13 mov.u32 %r83, %r76;
+	mov.b32 	%f63, %r83;
+	.loc	1 70 24
+	add.f32 	%f64, %f18, %f63;
+	.loc	1 72 24
+	add.f32 	%f65, %f19, %f64;
+	.loc	1 73 24
+	sub.f32 	%f66, %f65, %f16;
+	.loc	1 79 24
+	mul.f32 	%f68, %f66, %f67;
+	.loc	1 80 24
+	mul.f32 	%f69, %f68, %f20;
+	.loc	1 82 29
+	shl.b64 	%rd80, %rd30, 1;
+	add.s64 	%rd79, %rd38, %rd80;
+	.loc	1 82 52
+	mov.b32 	%r85, %f69;
+	cvt.rn.bf16.f32 %rs10, %r85;
+	@%p13 st.global.b16 [ %rd79 + 0 ], { %rs10 };
+	.loc	1 58 36
+	add.s32 	%r92, %r92, 4;
+	add.s64 	%rd91, %rd91, 16;
+	add.s64 	%rd90, %rd90, 16;
+	add.s64 	%rd89, %rd89, 16;
+	setp.lt.u32 	%p32, %r92, 252;
+	@%p32 bra 	$L__BB0_6;
+	bra.uni 	$L__BB0_9;
+$L__BB0_6:
+	.loc	1 62 51
+	mov.u32 %r75, 0x0;
+	@%p13 ld.global.L1::evict_last.b32 { %r75 }, [ %rd89 + 0 ];
+	@!%p13 mov.u32 %r75, %r76;
+	.loc	1 63 35
+	add.s32 	%r80, %r91, %r92;
+	add.s32 	%r81, %r80, 4;
+	mul.wide.s32 	%rd70, %r81, 2;
+	add.s64 	%rd68, %rd36, %rd70;
+	.loc	1 63 51
+	mov.u16 %rs7, 0x0;
+	@%p13 ld.global.L1::evict_first.b16 { %rs7 }, [ %rd68 + 0 ];
+	@!%p13 mov.u16 %rs7, %rs12;
+	.loc	1 64 40
+	mov.u32 %r78, 0x0;
+	@%p13 ld.global.L1::evict_last.b32 { %r78 }, [ %rd90 + 0 ];
+	@!%p13 mov.u32 %r78, %r76;
+	.loc	1 68 57
+	@%p28 bra 	$L__BB0_8;
+	mov.u64 	%rd71, assertMessage_1;
+	cvta.global.u64 	%rd72, %rd71;
+	mov.u64 	%rd73, assertFile_1;
+	cvta.global.u64 	%rd74, %rd73;
+	mov.u64 	%rd75, assertFunc_1;
+	cvta.global.u64 	%rd76, %rd75;
+	{ // callseq 11, 0
+	.reg .b32 temp_param_reg;
+	.param .b64 param0;
+	st.param.b64 	[param0+0], %rd72;
+	.param .b64 param1;
+	st.param.b64 	[param1+0], %rd74;
+	.param .b32 param2;
+	st.param.b32 	[param2+0], %r88;
+	.param .b64 param3;
+	st.param.b64 	[param3+0], %rd76;
+	.param .b64 param4;
+	st.param.b64 	[param4+0], %rd81;
+	call.uni 
+	__assertfail, 
+	(
+	param0, 
+	param1, 
+	param2, 
+	param3, 
+	param4
+	);
+	} // callseq 11
+	bra.uni 	$L__BB0_8;
+$L__BB0_9:
+	.loc	1 58 4
+	ret;
+$L__tmp10:
+$L__func_end0:
+
+}
+	// .globl	__nv_rsqrtf
+.visible .func  (.param .b32 func_retval0) __nv_rsqrtf(
+	.param .b32 __nv_rsqrtf_param_0
+)
+{
+	.reg .f32 	%f<3>;
+$L__func_begin1:
+
+	ld.param.f32 	%f1, [__nv_rsqrtf_param_0];
+	rsqrt.approx.ftz.f32 	%f2, %f1;
+	st.param.f32 	[func_retval0+0], %f2;
+	ret;
+$L__func_end1:
+
+}
+	.file	1 "/tmp/torchinductor_root/ci/ccig6fki6p4lxrdmgg6eudahiexcvueeol2p4qp532pvve2y463y.py"
+	.file	2 "/usr/local/lib/python3.10/dist-packages/torch/_inductor/triton_helpers.py"
+	.section	.debug_abbrev
+	{
+.b8 1
+.b8 17
+.b8 1
+.b8 37
+.b8 8
+.b8 19
+.b8 5
+.b8 3
+.b8 8
+.b8 16
+.b8 6
+.b8 27
+.b8 8
+.b8 180
+.b8 66
+.b8 12
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+.b8 0
+.b8 2
+.b8 46
+.b8 0
+.b8 135
+.b8 64
+.b8 8
+.b8 3
+.b8 8
+.b8 58
+.b8 11
+.b8 59
+.b8 11
+.b8 63
+.b8 12
+.b8 32
+.b8 11
+.b8 0
+.b8 0
+.b8 3
+.b8 46
+.b8 1
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 64
+.b8 10
+.b8 49
+.b8 19
+.b8 0
+.b8 0
+.b8 4
+.b8 29
+.b8 0
+.b8 49
+.b8 19
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 88
+.b8 11
+.b8 89
+.b8 11
+.b8 87
+.b8 11
+.b8 0
+.b8 0
+.b8 5
+.b8 29
+.b8 1
+.b8 49
+.b8 19
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 88
+.b8 11
+.b8 89
+.b8 11
+.b8 87
+.b8 11
+.b8 0
+.b8 0
+.b8 0
+	}
+	.section	.debug_info
+	{
+.b32 302
+.b8 2
+.b8 0
+.b32 .debug_abbrev
+.b8 8
+.b8 1
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 0
+.b8 2
+.b8 0
+.b8 99
+.b8 99
+.b8 105
+.b8 103
+.b8 54
+.b8 102
+.b8 107
+.b8 105
+.b8 54
+.b8 112
+.b8 52
+.b8 108
+.b8 120
+.b8 114
+.b8 100
+.b8 109
+.b8 103
+.b8 103
+.b8 54
+.b8 101
+.b8 117
+.b8 100
+.b8 97
+.b8 104
+.b8 105
+.b8 101
+.b8 120
+.b8 99
+.b8 118
+.b8 117
+.b8 101
+.b8 101
+.b8 111
+.b8 108
+.b8 50
+.b8 112
+.b8 52
+.b8 113
+.b8 112
+.b8 53
+.b8 51
+.b8 50
+.b8 112
+.b8 118
+.b8 118
+.b8 101
+.b8 50
+.b8 121
+.b8 52
+.b8 54
+.b8 51
+.b8 121
+.b8 46
+.b8 112
+.b8 121
+.b8 0
+.b32 .debug_line
+.b8 47
+.b8 116
+.b8 109
+.b8 112
+.b8 47
+.b8 116
+.b8 111
+.b8 114
+.b8 99
+.b8 104
+.b8 105
+.b8 110
+.b8 100
+.b8 117
+.b8 99
+.b8 116
+.b8 111
+.b8 114
+.b8 95
+.b8 114
+.b8 111
+.b8 111
+.b8 116
+.b8 47
+.b8 99
+.b8 105
+.b8 0
+.b8 1
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 2
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 51
+.b8 100
+.b8 52
+.b8 100
+.b8 53
+.b8 100
+.b8 54
+.b8 100
+.b8 101
+.b8 55
+.b8 100
+.b8 101
+.b8 0
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 51
+.b8 100
+.b8 52
+.b8 100
+.b8 53
+.b8 100
+.b8 54
+.b8 100
+.b8 101
+.b8 55
+.b8 100
+.b8 101
+.b8 0
+.b8 1
+.b8 18
+.b8 1
+.b8 1
+.b8 3
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 1
+.b8 156
+.b32 125
+.b8 4
+.b32 125
+.b64 $L__tmp1
+.b64 $L__tmp4
+.b8 2
+.b8 47
+.b8 41
+.b8 4
+.b32 125
+.b64 $L__tmp5
+.b64 $L__tmp8
+.b8 2
+.b8 53
+.b8 44
+.b8 5
+.b32 125
+.b64 $L__tmp6
+.b64 $L__tmp9
+.b8 2
+.b8 53
+.b8 44
+.b8 4
+.b32 125
+.b64 $L__tmp6
+.b64 $L__tmp9
+.b8 2
+.b8 120
+.b8 46
+.b8 0
+.b8 0
+.b8 0
+	}
+	.section	.debug_pubnames
+	{
+.b32 $L__pubNames_end0-$L__pubNames_start0
+$L__pubNames_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 306
+.b32 125
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 51
+.b8 100
+.b8 52
+.b8 100
+.b8 53
+.b8 100
+.b8 54
+.b8 100
+.b8 101
+.b8 55
+.b8 100
+.b8 101
+.b8 0
+.b32 0
+$L__pubNames_end0:
+	}
+	.section	.debug_pubtypes
+	{
+.b32 $L__pubTypes_end0-$L__pubTypes_start0
+$L__pubTypes_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 306
+.b32 0
+$L__pubTypes_end0:
+	}
+	.section	.debug_loc	{	}

.triton/dump/3cd3b6d7993c56f7d0340d40c84f737c/triton_.ttgir

@@ -0,0 +1,152 @@
+#blocked = #triton_gpu.blocked<{sizePerThread = [1, 1], threadsPerWarp = [8, 4], warpsPerCTA = [8, 1], order = [1, 0], CTAsPerCGA = [1, 1], CTASplitNum = [1, 1], CTAOrder = [1, 0]}>
+#blocked1 = #triton_gpu.blocked<{sizePerThread = [1, 1], threadsPerWarp = [32, 1], warpsPerCTA = [8, 1], order = [1, 0], CTAsPerCGA = [1, 1], CTASplitNum = [1, 1], CTAOrder = [1, 0]}>
+module attributes {"triton_gpu.compute-capability" = 89 : i32, "triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-warp" = 32 : i32} {
+  tt.func public @triton__0d1d2d3d4d5d6de7de(%arg0: !tt.ptr<i64, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg3: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg4: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg5: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg6: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}, %arg7: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %cst = arith.constant dense<512> : tensor<64x1xi32, #blocked>
+    %cst_0 = arith.constant dense<256> : tensor<64x1xi32, #blocked>
+    %cst_1 = arith.constant dense<256> : tensor<64x1xi64, #blocked>
+    %cst_2 = arith.constant dense<0> : tensor<64x1xi64, #blocked>
+    %cst_3 = arith.constant dense<50257> : tensor<64x1xi64, #blocked>
+    %cst_4 = arith.constant dense<50257> : tensor<64x1xi64, #blocked1>
+    %cst_5 = arith.constant dense<0> : tensor<64x1xi64, #blocked1>
+    %c0_i32 = arith.constant 0 : i32
+    %c4_i32 = arith.constant 4 : i32
+    %c256_i32 = arith.constant 256 : i32
+    %cst_6 = arith.constant dense<1.000000e+00> : tensor<64x4xf32, #blocked>
+    %cst_7 = arith.constant 0.000000e+00 : f32
+    %cst_8 = arith.constant dense<0.000000e+00> : tensor<64x4xbf16, #blocked>
+    %cst_9 = arith.constant dense<0.000000e+00> : tensor<1x4xf32, #blocked>
+    %cst_10 = arith.constant dense<0.000000e+00> : tensor<64x4xf32, #blocked>
+    %cst_11 = arith.constant dense<256> : tensor<1x4xi32, #blocked>
+    %cst_12 = arith.constant dense<9.99999974E-6> : tensor<64x1xf32, #blocked>
+    %cst_13 = arith.constant dense<2.560000e+02> : tensor<64x1xf32, #blocked>
+    %c64_i32 = arith.constant 64 : i32
+    %0 = tt.get_program_id x : i32
+    %1 = arith.muli %0, %c64_i32 : i32
+    %2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #triton_gpu.slice<{dim = 1, parent = #blocked}>>
+    %3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #triton_gpu.slice<{dim = 1, parent = #blocked1}>>
+    %4 = tt.expand_dims %2 {axis = 1 : i32} : (tensor<64xi32, #triton_gpu.slice<{dim = 1, parent = #blocked}>>) -> tensor<64x1xi32, #blocked>
+    %5 = tt.expand_dims %3 {axis = 1 : i32} : (tensor<64xi32, #triton_gpu.slice<{dim = 1, parent = #blocked1}>>) -> tensor<64x1xi32, #blocked1>
+    %6 = tt.splat %1 : (i32) -> tensor<64x1xi32, #blocked>
+    %7 = tt.splat %1 : (i32) -> tensor<64x1xi32, #blocked1>
+    %8 = arith.addi %6, %4 : tensor<64x1xi32, #blocked>
+    %9 = arith.addi %7, %5 : tensor<64x1xi32, #blocked1>
+    %10 = tt.make_range {end = 4 : i32, start = 0 : i32} : tensor<4xi32, #triton_gpu.slice<{dim = 0, parent = #blocked}>>
+    %11 = tt.expand_dims %10 {axis = 0 : i32} : (tensor<4xi32, #triton_gpu.slice<{dim = 0, parent = #blocked}>>) -> tensor<1x4xi32, #blocked>
+    %12 = tt.splat %arg0 : (!tt.ptr<i64, 1>) -> tensor<64x1x!tt.ptr<i64, 1>, #blocked>
+    %13 = tt.splat %arg0 : (!tt.ptr<i64, 1>) -> tensor<64x1x!tt.ptr<i64, 1>, #blocked1>
+    %14 = tt.addptr %12, %8 : tensor<64x1x!tt.ptr<i64, 1>, #blocked>, tensor<64x1xi32, #blocked>
+    %15 = tt.addptr %13, %9 : tensor<64x1x!tt.ptr<i64, 1>, #blocked1>, tensor<64x1xi32, #blocked1>
+    %16 = tt.load %14 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<64x1xi64, #blocked>
+    %17 = tt.load %15 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<64x1xi64, #blocked1>
+    %18 = arith.remsi %8, %cst : tensor<64x1xi32, #blocked>
+    %19 = arith.muli %18, %cst_0 : tensor<64x1xi32, #blocked>
+    %20 = tt.broadcast %19 : (tensor<64x1xi32, #blocked>) -> tensor<64x4xi32, #blocked>
+    %21 = tt.splat %arg2 : (!tt.ptr<f32, 1>) -> tensor<64x4x!tt.ptr<f32, 1>, #blocked>
+    %22 = arith.muli %8, %cst_0 : tensor<64x1xi32, #blocked>
+    %23 = tt.broadcast %22 : (tensor<64x1xi32, #blocked>) -> tensor<64x4xi32, #blocked>
+    %24 = tt.splat %arg3 : (!tt.ptr<bf16, 1>) -> tensor<64x4x!tt.ptr<bf16, 1>, #blocked>
+    %25 = arith.addi %16, %cst_3 : tensor<64x1xi64, #blocked>
+    %26 = arith.addi %17, %cst_4 : tensor<64x1xi64, #blocked1>
+    %27 = arith.cmpi slt, %16, %cst_2 : tensor<64x1xi64, #blocked>
+    %28 = arith.cmpi slt, %17, %cst_5 : tensor<64x1xi64, #blocked1>
+    %29 = arith.select %27, %25, %16 : tensor<64x1xi1, #blocked>, tensor<64x1xi64, #blocked>
+    %30 = arith.select %28, %26, %17 : tensor<64x1xi1, #blocked1>, tensor<64x1xi64, #blocked1>
+    %31 = arith.cmpi sge, %30, %cst_5 : tensor<64x1xi64, #blocked1>
+    %32 = arith.cmpi slt, %30, %cst_4 : tensor<64x1xi64, #blocked1>
+    %33 = arith.andi %31, %32 : tensor<64x1xi1, #blocked1>
+    %34 = arith.muli %29, %cst_1 : tensor<64x1xi64, #blocked>
+    %35 = tt.broadcast %34 : (tensor<64x1xi64, #blocked>) -> tensor<64x4xi64, #blocked>
+    %36 = tt.splat %arg1 : (!tt.ptr<f32, 1>) -> tensor<64x4x!tt.ptr<f32, 1>, #blocked>
+    %37:3 = scf.for %arg8 = %c0_i32 to %c256_i32 step %c4_i32 iter_args(%arg9 = %cst_10, %arg10 = %cst_10, %arg11 = %cst_10) -> (tensor<64x4xf32, #blocked>, tensor<64x4xf32, #blocked>, tensor<64x4xf32, #blocked>)  : i32 {
+      %46 = tt.splat %arg8 : (i32) -> tensor<1x4xi32, #blocked>
+      %47 = arith.addi %46, %11 : tensor<1x4xi32, #blocked>
+      %48 = arith.cmpi slt, %47, %cst_11 : tensor<1x4xi32, #blocked>
+      %49 = tt.broadcast %47 : (tensor<1x4xi32, #blocked>) -> tensor<64x4xi32, #blocked>
+      %50 = arith.addi %49, %20 : tensor<64x4xi32, #blocked>
+      %51 = tt.addptr %21, %50 : tensor<64x4x!tt.ptr<f32, 1>, #blocked>, tensor<64x4xi32, #blocked>
+      %52 = tt.broadcast %48 : (tensor<1x4xi1, #blocked>) -> tensor<64x4xi1, #blocked>
+      %53 = tt.load %51, %52, %cst_10 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<64x4xf32, #blocked>
+      %54 = arith.addi %49, %23 : tensor<64x4xi32, #blocked>
+      %55 = tt.addptr %24, %54 : tensor<64x4x!tt.ptr<bf16, 1>, #blocked>, tensor<64x4xi32, #blocked>
+      %56 = tt.load %55, %52, %cst_8 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<64x4xbf16, #blocked>
+      %57 = arith.extf %56 : tensor<64x4xbf16, #blocked> to tensor<64x4xf32, #blocked>
+      tt.assert %33, "index out of bounds: 0 <= tmp3 < 50257", "<frozen importlib._bootstrap_external>", "_call_with_frames_removed", 883 : tensor<64x1xi1, #blocked1>
+      %58 = arith.extsi %47 : tensor<1x4xi32, #blocked> to tensor<1x4xi64, #blocked>
+      %59 = tt.broadcast %58 : (tensor<1x4xi64, #blocked>) -> tensor<64x4xi64, #blocked>
+      %60 = arith.addi %59, %35 : tensor<64x4xi64, #blocked>
+      %61 = tt.addptr %36, %60 : tensor<64x4x!tt.ptr<f32, 1>, #blocked>, tensor<64x4xi64, #blocked>
+      %62 = tt.load %61, %52, %cst_10 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<64x4xf32, #blocked>
+      %63 = arith.addf %62, %53 : tensor<64x4xf32, #blocked>
+      %64 = arith.addf %63, %57 : tensor<64x4xf32, #blocked>
+      %65 = arith.subf %64, %arg9 : tensor<64x4xf32, #blocked>
+      %66 = arith.addf %arg11, %cst_6 : tensor<64x4xf32, #blocked>
+      %67 = arith.divf %65, %66 : tensor<64x4xf32, #blocked>
+      %68 = arith.addf %arg9, %67 : tensor<64x4xf32, #blocked>
+      %69 = arith.subf %64, %68 : tensor<64x4xf32, #blocked>
+      %70 = arith.mulf %65, %69 : tensor<64x4xf32, #blocked>
+      %71 = arith.addf %arg10, %70 : tensor<64x4xf32, #blocked>
+      %72 = arith.select %52, %68, %arg9 : tensor<64x4xi1, #blocked>, tensor<64x4xf32, #blocked>
+      %73 = arith.select %52, %71, %arg10 : tensor<64x4xi1, #blocked>, tensor<64x4xf32, #blocked>
+      %74 = arith.select %52, %66, %arg11 : tensor<64x4xi1, #blocked>, tensor<64x4xf32, #blocked>
+      scf.yield %72, %73, %74 : tensor<64x4xf32, #blocked>, tensor<64x4xf32, #blocked>, tensor<64x4xf32, #blocked>
+    }
+    %38:3 = "tt.reduce"(%37#0, %37#1, %37#2) <{axis = 1 : i32}> ({
+    ^bb0(%arg8: f32, %arg9: f32, %arg10: f32, %arg11: f32, %arg12: f32, %arg13: f32):
+      %46 = arith.subf %arg11, %arg8 : f32
+      %47 = arith.addf %arg10, %arg13 : f32
+      %48 = arith.cmpf oeq, %47, %cst_7 : f32
+      %49 = arith.divf %arg13, %47 : f32
+      %50 = arith.select %48, %cst_7, %49 : f32
+      %51 = arith.mulf %46, %50 : f32
+      %52 = arith.addf %arg8, %51 : f32
+      %53 = arith.addf %arg9, %arg12 : f32
+      %54 = arith.mulf %46, %46 : f32
+      %55 = arith.mulf %54, %arg10 : f32
+      %56 = arith.mulf %55, %50 : f32
+      %57 = arith.addf %53, %56 : f32
+      tt.reduce.return %52, %57, %47 : f32, f32, f32
+    }) : (tensor<64x4xf32, #blocked>, tensor<64x4xf32, #blocked>, tensor<64x4xf32, #blocked>) -> (tensor<64xf32, #triton_gpu.slice<{dim = 1, parent = #blocked}>>, tensor<64xf32, #triton_gpu.slice<{dim = 1, parent = #blocked}>>, tensor<64xf32, #triton_gpu.slice<{dim = 1, parent = #blocked}>>)
+    %39 = tt.expand_dims %38#0 {axis = 1 : i32} : (tensor<64xf32, #triton_gpu.slice<{dim = 1, parent = #blocked}>>) -> tensor<64x1xf32, #blocked>
+    %40 = tt.expand_dims %38#1 {axis = 1 : i32} : (tensor<64xf32, #triton_gpu.slice<{dim = 1, parent = #blocked}>>) -> tensor<64x1xf32, #blocked>
+    %41 = tt.splat %arg4 : (!tt.ptr<f32, 1>) -> tensor<1x4x!tt.ptr<f32, 1>, #blocked>
+    %42 = tt.broadcast %39 : (tensor<64x1xf32, #blocked>) -> tensor<64x4xf32, #blocked>
+    %43 = arith.divf %40, %cst_13 : tensor<64x1xf32, #blocked>
+    %44 = arith.addf %43, %cst_12 : tensor<64x1xf32, #blocked>
+    %45 = tt.splat %arg5 : (!tt.ptr<bf16, 1>) -> tensor<64x4x!tt.ptr<bf16, 1>, #blocked>
+    scf.for %arg8 = %c0_i32 to %c256_i32 step %c4_i32  : i32 {
+      %46 = tt.splat %arg8 : (i32) -> tensor<1x4xi32, #blocked>
+      %47 = arith.addi %46, %11 : tensor<1x4xi32, #blocked>
+      %48 = arith.cmpi slt, %47, %cst_11 : tensor<1x4xi32, #blocked>
+      %49 = tt.broadcast %47 : (tensor<1x4xi32, #blocked>) -> tensor<64x4xi32, #blocked>
+      %50 = arith.addi %49, %20 : tensor<64x4xi32, #blocked>
+      %51 = tt.addptr %21, %50 : tensor<64x4x!tt.ptr<f32, 1>, #blocked>, tensor<64x4xi32, #blocked>
+      %52 = tt.broadcast %48 : (tensor<1x4xi1, #blocked>) -> tensor<64x4xi1, #blocked>
+      %53 = tt.load %51, %52, %cst_10 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<64x4xf32, #blocked>
+      %54 = arith.addi %49, %23 : tensor<64x4xi32, #blocked>
+      %55 = tt.addptr %24, %54 : tensor<64x4x!tt.ptr<bf16, 1>, #blocked>, tensor<64x4xi32, #blocked>
+      %56 = tt.load %55, %52, %cst_8 {cache = 1 : i32, evict = 2 : i32, isVolatile = false} : tensor<64x4xbf16, #blocked>
+      %57 = arith.extf %56 : tensor<64x4xbf16, #blocked> to tensor<64x4xf32, #blocked>
+      %58 = tt.addptr %41, %47 : tensor<1x4x!tt.ptr<f32, 1>, #blocked>, tensor<1x4xi32, #blocked>
+      %59 = tt.load %58, %48, %cst_9 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<1x4xf32, #blocked>
+      tt.assert %33, "index out of bounds: 0 <= tmp16 < 50257", "<frozen importlib._bootstrap_external>", "_call_with_frames_removed", 883 : tensor<64x1xi1, #blocked1>
+      %60 = arith.extsi %47 : tensor<1x4xi32, #blocked> to tensor<1x4xi64, #blocked>
+      %61 = tt.broadcast %60 : (tensor<1x4xi64, #blocked>) -> tensor<64x4xi64, #blocked>
+      %62 = arith.addi %61, %35 : tensor<64x4xi64, #blocked>
+      %63 = tt.addptr %36, %62 : tensor<64x4x!tt.ptr<f32, 1>, #blocked>, tensor<64x4xi64, #blocked>
+      %64 = tt.load %63, %52, %cst_10 {cache = 1 : i32, evict = 2 : i32, isVolatile = false} : tensor<64x4xf32, #blocked>
+      %65 = arith.addf %64, %53 : tensor<64x4xf32, #blocked>
+      %66 = arith.addf %65, %57 : tensor<64x4xf32, #blocked>
+      %67 = arith.subf %66, %42 : tensor<64x4xf32, #blocked>
+      %68 = tt.extern_elementwise %44 {libname = "libdevice", libpath = "/usr/local/lib/python3.10/dist-packages/triton/language/../third_party/cuda/lib/libdevice.10.bc", pure = true, symbol = "__nv_rsqrtf"} : (tensor<64x1xf32, #blocked>) -> tensor<64x1xf32, #blocked>
+      %69 = tt.broadcast %68 : (tensor<64x1xf32, #blocked>) -> tensor<64x4xf32, #blocked>
+      %70 = arith.mulf %67, %69 : tensor<64x4xf32, #blocked>
+      %71 = tt.broadcast %59 : (tensor<1x4xf32, #blocked>) -> tensor<64x4xf32, #blocked>
+      %72 = arith.mulf %70, %71 : tensor<64x4xf32, #blocked>
+      %73 = tt.addptr %45, %54 : tensor<64x4x!tt.ptr<bf16, 1>, #blocked>, tensor<64x4xi32, #blocked>
+      %74 = arith.truncf %72 : tensor<64x4xf32, #blocked> to tensor<64x4xbf16, #blocked>
+      tt.store %73, %74, %52 {cache = 1 : i32, evict = 1 : i32} : tensor<64x4xbf16, #blocked>
+    }
+    tt.return
+  }
+}

.triton/dump/4993935f9a0e5939755cfb42600362cf/triton_.ptx

@@ -0,0 +1,295 @@
+//
+// Generated by LLVM NVPTX Back-End
+//
+
+.version 8.2
+.target sm_89
+.address_size 64
+
+	// .globl	triton__0d1d2de
+
+.visible .entry triton__0d1d2de(
+	.param .u64 triton__0d1d2de_param_0,
+	.param .u64 triton__0d1d2de_param_1,
+	.param .u32 triton__0d1d2de_param_2
+)
+.maxntid 256, 1, 1
+{
+	.reg .pred 	%p<3>;
+	.reg .b16 	%rs<3>;
+	.reg .b32 	%r<13>;
+	.reg .b64 	%rd<7>;
+	.loc	1 18 0
+$L__func_begin0:
+	.loc	1 18 0
+
+	ld.param.u64 	%rd3, [triton__0d1d2de_param_0];
+	ld.param.u64 	%rd4, [triton__0d1d2de_param_1];
+$L__tmp0:
+	.loc	1 21 36
+	mov.u32 	%r7, %tid.x;
+	shl.b32 	%r8, %r7, 1;
+	and.b32  	%r9, %r8, 510;
+	.loc	1 20 28
+	mov.u32 %r1, %ctaid.x;
+	.loc	1 20 33
+	shl.b32 	%r10, %r1, 9;
+	.loc	1 21 23
+	or.b32  	%r11, %r10, %r9;
+	.loc	1 24 30
+	mul.wide.s32 	%rd5, %r11, 4;
+	add.s64 	%rd1, %rd3, %rd5;
+	mov.pred 	%p1, -1;
+	.loc	1 24 35
+	mov.u32 %r4, 0x0;
+	mov.u32 %r5, 0x0;
+	@%p1 ld.global.v2.b32 { %r4, %r5 }, [ %rd1 + 0 ];
+	.loc	1 26 25
+	mul.wide.s32 	%rd6, %r11, 2;
+	add.s64 	%rd2, %rd4, %rd6;
+	.loc	1 26 36
+	cvt.rn.bf16.f32 %rs1, %r4;
+	cvt.rn.bf16.f32 %rs2, %r5;
+	mov.b32 	%r12, {%rs1, %rs2};
+	@%p1 st.global.b32 [ %rd2 + 0 ], { %r12 };
+	.loc	1 26 4
+	ret;
+$L__tmp1:
+$L__func_end0:
+
+}
+	.file	1 "/tmp/torchinductor_root/zj/czjxjqxojsyyr4zmce6q6twysnucw6p4l5ujgp6ts2ecrm3ue3ex.py"
+	.section	.debug_abbrev
+	{
+.b8 1
+.b8 17
+.b8 1
+.b8 37
+.b8 8
+.b8 19
+.b8 5
+.b8 3
+.b8 8
+.b8 16
+.b8 6
+.b8 27
+.b8 8
+.b8 180
+.b8 66
+.b8 12
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+.b8 0
+.b8 2
+.b8 46
+.b8 0
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 64
+.b8 10
+.b8 135
+.b8 64
+.b8 8
+.b8 3
+.b8 8
+.b8 58
+.b8 11
+.b8 59
+.b8 11
+.b8 63
+.b8 12
+.b8 0
+.b8 0
+.b8 0
+	}
+	.section	.debug_info
+	{
+.b32 176
+.b8 2
+.b8 0
+.b32 .debug_abbrev
+.b8 8
+.b8 1
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 0
+.b8 2
+.b8 0
+.b8 99
+.b8 122
+.b8 106
+.b8 120
+.b8 106
+.b8 113
+.b8 120
+.b8 111
+.b8 106
+.b8 115
+.b8 121
+.b8 121
+.b8 114
+.b8 52
+.b8 122
+.b8 109
+.b8 99
+.b8 101
+.b8 54
+.b8 113
+.b8 54
+.b8 116
+.b8 119
+.b8 121
+.b8 115
+.b8 110
+.b8 117
+.b8 99
+.b8 119
+.b8 54
+.b8 112
+.b8 52
+.b8 108
+.b8 53
+.b8 117
+.b8 106
+.b8 103
+.b8 112
+.b8 54
+.b8 116
+.b8 115
+.b8 50
+.b8 101
+.b8 99
+.b8 114
+.b8 109
+.b8 51
+.b8 117
+.b8 101
+.b8 51
+.b8 101
+.b8 120
+.b8 46
+.b8 112
+.b8 121
+.b8 0
+.b32 .debug_line
+.b8 47
+.b8 116
+.b8 109
+.b8 112
+.b8 47
+.b8 116
+.b8 111
+.b8 114
+.b8 99
+.b8 104
+.b8 105
+.b8 110
+.b8 100
+.b8 117
+.b8 99
+.b8 116
+.b8 111
+.b8 114
+.b8 95
+.b8 114
+.b8 111
+.b8 111
+.b8 116
+.b8 47
+.b8 122
+.b8 106
+.b8 0
+.b8 1
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 2
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 1
+.b8 156
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 101
+.b8 0
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 101
+.b8 0
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+	}
+	.section	.debug_pubnames
+	{
+.b32 $L__pubNames_end0-$L__pubNames_start0
+$L__pubNames_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 180
+.b32 125
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 101
+.b8 0
+.b32 0
+$L__pubNames_end0:
+	}
+	.section	.debug_pubtypes
+	{
+.b32 $L__pubTypes_end0-$L__pubTypes_start0
+$L__pubTypes_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 180
+.b32 0
+$L__pubTypes_end0:
+	}
+	.section	.debug_loc	{	}

.triton/dump/4c6ad48573c74d55ed79384f6b432d50/triton_.ttir

@@ -0,0 +1,18 @@
+module {
+  tt.func public @triton__0d1d2de(%arg0: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg2: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %c1024_i32 = arith.constant 1024 : i32
+    %0 = tt.get_program_id x : i32
+    %1 = arith.muli %0, %c1024_i32 : i32
+    %2 = tt.make_range {end = 1024 : i32, start = 0 : i32} : tensor<1024xi32>
+    %3 = tt.splat %1 : (i32) -> tensor<1024xi32>
+    %4 = arith.addi %3, %2 : tensor<1024xi32>
+    %5 = tt.splat %arg0 : (!tt.ptr<f32, 1>) -> tensor<1024x!tt.ptr<f32, 1>>
+    %6 = tt.addptr %5, %4 : tensor<1024x!tt.ptr<f32, 1>>, tensor<1024xi32>
+    %7 = tt.load %6 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<1024xf32>
+    %8 = tt.splat %arg1 : (!tt.ptr<bf16, 1>) -> tensor<1024x!tt.ptr<bf16, 1>>
+    %9 = tt.addptr %8, %4 : tensor<1024x!tt.ptr<bf16, 1>>, tensor<1024xi32>
+    %10 = arith.truncf %7 : tensor<1024xf32> to tensor<1024xbf16>
+    tt.store %9, %10 {cache = 1 : i32, evict = 1 : i32} : tensor<1024xbf16>
+    tt.return
+  }
+}

.triton/dump/4ce9eb7fe63f19e54893f0c74df91471/triton_.ttgir

@@ -0,0 +1,28 @@
+#blocked = #triton_gpu.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [8], order = [0], CTAsPerCGA = [1], CTASplitNum = [1], CTAOrder = [0]}>
+module attributes {"triton_gpu.compute-capability" = 89 : i32, "triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-warp" = 32 : i32} {
+  tt.func public @triton__0d1d2de(%arg0: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg2: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %cst = arith.constant dense<1.000000e+00> : tensor<512xf32, #blocked>
+    %cst_0 = arith.constant dense<0.707106769> : tensor<512xf32, #blocked>
+    %cst_1 = arith.constant dense<5.000000e-01> : tensor<512xf32, #blocked>
+    %c512_i32 = arith.constant 512 : i32
+    %0 = tt.get_program_id x : i32
+    %1 = arith.muli %0, %c512_i32 : i32
+    %2 = tt.make_range {end = 512 : i32, start = 0 : i32} : tensor<512xi32, #blocked>
+    %3 = tt.splat %1 : (i32) -> tensor<512xi32, #blocked>
+    %4 = arith.addi %3, %2 : tensor<512xi32, #blocked>
+    %5 = tt.splat %arg0 : (!tt.ptr<bf16, 1>) -> tensor<512x!tt.ptr<bf16, 1>, #blocked>
+    %6 = tt.addptr %5, %4 : tensor<512x!tt.ptr<bf16, 1>, #blocked>, tensor<512xi32, #blocked>
+    %7 = tt.load %6 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<512xbf16, #blocked>
+    %8 = arith.extf %7 : tensor<512xbf16, #blocked> to tensor<512xf32, #blocked>
+    %9 = arith.mulf %8, %cst_1 : tensor<512xf32, #blocked>
+    %10 = arith.mulf %8, %cst_0 : tensor<512xf32, #blocked>
+    %11 = tt.extern_elementwise %10 {libname = "libdevice", libpath = "/usr/local/lib/python3.10/dist-packages/triton/language/../third_party/cuda/lib/libdevice.10.bc", pure = true, symbol = "__nv_erff"} : (tensor<512xf32, #blocked>) -> tensor<512xf32, #blocked>
+    %12 = arith.addf %11, %cst : tensor<512xf32, #blocked>
+    %13 = arith.mulf %9, %12 : tensor<512xf32, #blocked>
+    %14 = tt.splat %arg1 : (!tt.ptr<bf16, 1>) -> tensor<512x!tt.ptr<bf16, 1>, #blocked>
+    %15 = tt.addptr %14, %4 : tensor<512x!tt.ptr<bf16, 1>, #blocked>, tensor<512xi32, #blocked>
+    %16 = arith.truncf %13 : tensor<512xf32, #blocked> to tensor<512xbf16, #blocked>
+    tt.store %15, %16 {cache = 1 : i32, evict = 1 : i32} : tensor<512xbf16, #blocked>
+    tt.return
+  }
+}

.triton/dump/4ce9eb7fe63f19e54893f0c74df91471/triton_.ttir

@@ -0,0 +1,27 @@
+module {
+  tt.func public @triton__0d1d2de(%arg0: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg2: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %cst = arith.constant dense<1.000000e+00> : tensor<512xf32>
+    %cst_0 = arith.constant dense<0.707106769> : tensor<512xf32>
+    %cst_1 = arith.constant dense<5.000000e-01> : tensor<512xf32>
+    %c512_i32 = arith.constant 512 : i32
+    %0 = tt.get_program_id x : i32
+    %1 = arith.muli %0, %c512_i32 : i32
+    %2 = tt.make_range {end = 512 : i32, start = 0 : i32} : tensor<512xi32>
+    %3 = tt.splat %1 : (i32) -> tensor<512xi32>
+    %4 = arith.addi %3, %2 : tensor<512xi32>
+    %5 = tt.splat %arg0 : (!tt.ptr<bf16, 1>) -> tensor<512x!tt.ptr<bf16, 1>>
+    %6 = tt.addptr %5, %4 : tensor<512x!tt.ptr<bf16, 1>>, tensor<512xi32>
+    %7 = tt.load %6 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<512xbf16>
+    %8 = arith.extf %7 : tensor<512xbf16> to tensor<512xf32>
+    %9 = arith.mulf %8, %cst_1 : tensor<512xf32>
+    %10 = arith.mulf %8, %cst_0 : tensor<512xf32>
+    %11 = tt.extern_elementwise %10 {libname = "libdevice", libpath = "/usr/local/lib/python3.10/dist-packages/triton/language/../third_party/cuda/lib/libdevice.10.bc", pure = true, symbol = "__nv_erff"} : (tensor<512xf32>) -> tensor<512xf32>
+    %12 = arith.addf %11, %cst : tensor<512xf32>
+    %13 = arith.mulf %9, %12 : tensor<512xf32>
+    %14 = tt.splat %arg1 : (!tt.ptr<bf16, 1>) -> tensor<512x!tt.ptr<bf16, 1>>
+    %15 = tt.addptr %14, %4 : tensor<512x!tt.ptr<bf16, 1>>, tensor<512xi32>
+    %16 = arith.truncf %13 : tensor<512xf32> to tensor<512xbf16>
+    tt.store %15, %16 {cache = 1 : i32, evict = 1 : i32} : tensor<512xbf16>
+    tt.return
+  }
+}

.triton/dump/76fb48b96c75cb8e388c291a18ef9b02/triton_.llir

@@ -0,0 +1,600 @@
+; ModuleID = 'LLVMDialectModule'
+source_filename = "LLVMDialectModule"
+
+@assertFunc_1 = internal constant [8 x i8] c"<module>"
+@assertFile_1 = internal constant [68 x i8] c"/usr/local/lib/python3.10/dist-packages/torch/_inductor/codecache.py"
+@assertMessage_1 = internal constant [39 x i8] c"index out of bounds: 0 <= tmp16 < 50257"
+@assertFunc_0 = internal constant [8 x i8] c"<module>"
+@assertFile_0 = internal constant [68 x i8] c"/usr/local/lib/python3.10/dist-packages/torch/_inductor/codecache.py"
+@assertMessage_0 = internal constant [38 x i8] c"index out of bounds: 0 <= tmp3 < 50257"
+@global_smem = external addrspace(3) global [0 x i8]
+@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1
+
+declare void @__assertfail(ptr, ptr, i32, ptr, i64) local_unnamed_addr
+
+define void @triton__0d1d2d3d4d5d6de7de(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, ptr addrspace(1) %4, ptr addrspace(1) %5, i32 %6, i32 %7) local_unnamed_addr !dbg !7 {
+  %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !10
+  %10 = and i32 %9, 31, !dbg !10
+  %11 = lshr i32 %9, 6, !dbg !10
+  %12 = and i32 %11, 1, !dbg !10
+  %13 = and i32 %9, 1, !dbg !10
+  %urem = shl i32 %9, 1, !dbg !11
+  %14 = and i32 %urem, 126, !dbg !11
+  %15 = tail call i32 asm "mov.u32 $0, %ctaid.x;", "=r"() #6, !dbg !12
+  %16 = shl i32 %15, 1, !dbg !13
+  %17 = or i32 %16, %12, !dbg !14
+  %18 = or i32 %16, %13, !dbg !14
+  %19 = sext i32 %17 to i64, !dbg !15
+  %20 = getelementptr i64, ptr addrspace(1) %0, i64 %19, !dbg !15
+  %21 = sext i32 %18 to i64, !dbg !15
+  %22 = getelementptr i64, ptr addrspace(1) %0, i64 %21, !dbg !15
+  %23 = tail call i64 asm sideeffect "mov.u64 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b64 { $0 }, [ $1 + 0 ];", "=l,l,b"(ptr addrspace(1) %20, i1 true) #6, !dbg !16
+  %24 = tail call i64 asm sideeffect "mov.u64 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b64 { $0 }, [ $1 + 0 ];", "=l,l,b"(ptr addrspace(1) %20, i1 true) #6, !dbg !16
+  %25 = tail call i64 asm sideeffect "mov.u64 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b64 { $0 }, [ $1 + 0 ];", "=l,l,b"(ptr addrspace(1) %22, i1 true) #6, !dbg !16
+  %26 = srem i32 %17, 512, !dbg !17
+  %27 = shl nsw i32 %26, 8, !dbg !18
+  %28 = shl i32 %17, 8, !dbg !19
+  %29 = add i64 %25, 50257, !dbg !20
+  %30 = icmp slt i64 %23, 0, !dbg !21
+  %31 = icmp slt i64 %25, 0, !dbg !21
+  %32 = select i1 %31, i64 %29, i64 %25, !dbg !22
+  %33 = icmp ugt i64 %32, 50256, !dbg !23
+  %34 = shl i64 %23, 8, !dbg !24
+  %35 = add i64 %34, 12865792, !dbg !24
+  %36 = select i1 %30, i64 %35, i64 %34, !dbg !24
+  %37 = getelementptr float, ptr addrspace(1) %1, i64 %36
+  %38 = or i32 %14, %27, !dbg !25
+  %39 = sext i32 %38 to i64, !dbg !26
+  %40 = getelementptr float, ptr addrspace(1) %2, i64 %39, !dbg !26
+  %41 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_last.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %40, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !27
+  %42 = extractvalue { i32, i32 } %41, 0, !dbg !27
+  %43 = extractvalue { i32, i32 } %41, 1, !dbg !27
+  %44 = insertelement <2 x i32> poison, i32 %42, i64 0, !dbg !27
+  %45 = insertelement <2 x i32> %44, i32 %43, i64 1, !dbg !27
+  %46 = bitcast <2 x i32> %45 to <2 x float>, !dbg !27
+  %47 = or i32 %14, %28, !dbg !28
+  %48 = sext i32 %47 to i64, !dbg !29
+  %49 = getelementptr i16, ptr addrspace(1) %3, i64 %48, !dbg !29
+  %50 = tail call i32 asm sideeffect "mov.u32 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b32 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u32 $0, $3;", "=r,l,b,r,b"(ptr addrspace(1) %49, i1 true, i32 0, i1 true) #6, !dbg !30
+  %51 = trunc i32 %50 to i16, !dbg !30
+  %extelt.offset2 = lshr i32 %50, 16, !dbg !30
+  %52 = trunc i32 %extelt.offset2 to i16, !dbg !30
+  %53 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %51) #6, !dbg !31
+  %54 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %52) #6, !dbg !31
+  br i1 %33, label %55, label %56, !dbg !32
+
+55:                                               ; preds = %8
+  tail call void @__assertfail(ptr nonnull @assertMessage_0, ptr nonnull @assertFile_0, i32 1892, ptr nonnull @assertFunc_0, i64 1), !dbg !32
+  br label %56, !dbg !32
+
+56:                                               ; preds = %55, %8
+  %57 = zext nneg i32 %14 to i64, !dbg !33
+  %58 = getelementptr float, ptr addrspace(1) %37, i64 %57, !dbg !34
+  %59 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_last.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %58, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !35
+  %60 = extractvalue { i32, i32 } %59, 0, !dbg !35
+  %61 = extractvalue { i32, i32 } %59, 1, !dbg !35
+  %62 = insertelement <2 x i32> poison, i32 %60, i64 0, !dbg !35
+  %63 = insertelement <2 x i32> %62, i32 %61, i64 1, !dbg !35
+  %64 = bitcast <2 x i32> %63 to <2 x float>, !dbg !35
+  %65 = fadd <2 x float> %46, %64, !dbg !36
+  %66 = insertelement <2 x float> poison, float %53, i64 0, !dbg !37
+  %67 = insertelement <2 x float> %66, float %54, i64 1, !dbg !37
+  %68 = fadd <2 x float> %67, %65, !dbg !37
+  %69 = extractelement <2 x float> %68, i64 0, !dbg !38
+  %70 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %69, float 1.000000e+00) #6, !dbg !38
+  %71 = extractelement <2 x float> %68, i64 1, !dbg !38
+  %72 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %71, float 1.000000e+00) #6, !dbg !38
+  %73 = insertelement <2 x float> poison, float %70, i64 0, !dbg !42
+  %74 = insertelement <2 x float> %73, float %72, i64 1, !dbg !42
+  %75 = fadd <2 x float> %74, zeroinitializer, !dbg !42
+  %76 = fsub <2 x float> %68, %75, !dbg !43
+  %77 = fmul <2 x float> %68, %76, !dbg !44
+  %78 = fadd <2 x float> %77, zeroinitializer, !dbg !45
+  %79 = or i32 %14, 128, !dbg !46
+  %80 = or i32 %79, %27, !dbg !25
+  %81 = sext i32 %80 to i64, !dbg !26
+  %82 = getelementptr float, ptr addrspace(1) %2, i64 %81, !dbg !26
+  %83 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_last.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %82, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !27
+  %84 = extractvalue { i32, i32 } %83, 0, !dbg !27
+  %85 = extractvalue { i32, i32 } %83, 1, !dbg !27
+  %86 = insertelement <2 x i32> poison, i32 %84, i64 0, !dbg !27
+  %87 = insertelement <2 x i32> %86, i32 %85, i64 1, !dbg !27
+  %88 = bitcast <2 x i32> %87 to <2 x float>, !dbg !27
+  %89 = or i32 %79, %28, !dbg !28
+  %90 = sext i32 %89 to i64, !dbg !29
+  %91 = getelementptr i16, ptr addrspace(1) %3, i64 %90, !dbg !29
+  %92 = tail call i32 asm sideeffect "mov.u32 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b32 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u32 $0, $3;", "=r,l,b,r,b"(ptr addrspace(1) %91, i1 true, i32 0, i1 true) #6, !dbg !30
+  %93 = trunc i32 %92 to i16, !dbg !30
+  %extelt.offset2.1 = lshr i32 %92, 16, !dbg !30
+  %94 = trunc i32 %extelt.offset2.1 to i16, !dbg !30
+  %95 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %93) #6, !dbg !31
+  %96 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %94) #6, !dbg !31
+  br i1 %33, label %97, label %98, !dbg !32
+
+97:                                               ; preds = %56
+  tail call void @__assertfail(ptr nonnull @assertMessage_0, ptr nonnull @assertFile_0, i32 1892, ptr nonnull @assertFunc_0, i64 1), !dbg !32
+  br label %98, !dbg !32
+
+98:                                               ; preds = %97, %56
+  %99 = zext nneg i32 %79 to i64, !dbg !33
+  %100 = getelementptr float, ptr addrspace(1) %37, i64 %99, !dbg !34
+  %101 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_last.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %100, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !35
+  %102 = extractvalue { i32, i32 } %101, 0, !dbg !35
+  %103 = extractvalue { i32, i32 } %101, 1, !dbg !35
+  %104 = insertelement <2 x i32> poison, i32 %102, i64 0, !dbg !35
+  %105 = insertelement <2 x i32> %104, i32 %103, i64 1, !dbg !35
+  %106 = bitcast <2 x i32> %105 to <2 x float>, !dbg !35
+  %107 = fadd <2 x float> %88, %106, !dbg !36
+  %108 = insertelement <2 x float> poison, float %95, i64 0, !dbg !37
+  %109 = insertelement <2 x float> %108, float %96, i64 1, !dbg !37
+  %110 = fadd <2 x float> %109, %107, !dbg !37
+  %111 = fsub <2 x float> %110, %75, !dbg !47
+  %112 = extractelement <2 x float> %111, i64 0, !dbg !38
+  %113 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %112, float 2.000000e+00) #6, !dbg !38
+  %114 = extractelement <2 x float> %111, i64 1, !dbg !38
+  %115 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %114, float 2.000000e+00) #6, !dbg !38
+  %116 = insertelement <2 x float> poison, float %113, i64 0, !dbg !42
+  %117 = insertelement <2 x float> %116, float %115, i64 1, !dbg !42
+  %118 = fadd <2 x float> %75, %117, !dbg !42
+  %119 = fsub <2 x float> %110, %118, !dbg !43
+  %120 = fmul <2 x float> %111, %119, !dbg !44
+  %121 = fadd <2 x float> %78, %120, !dbg !45
+  %122 = lshr i32 %9, 5, !dbg !10
+  %123 = and i32 %122, 1, !dbg !11
+  %124 = and i32 %9, 127, !dbg !11
+  %125 = zext nneg i32 %124 to i64, !dbg !48
+  %126 = getelementptr float, ptr addrspace(3) @global_smem, i64 %125, !dbg !48
+  store <1 x float> <float 2.000000e+00>, ptr addrspace(3) %126, align 4, !dbg !48
+  %127 = add nuw nsw i32 %124, 130, !dbg !48
+  %128 = zext nneg i32 %127 to i64, !dbg !48
+  %129 = getelementptr float, ptr addrspace(3) @global_smem, i64 %128, !dbg !48
+  store <1 x float> <float 2.000000e+00>, ptr addrspace(3) %129, align 4, !dbg !48
+  tail call void @llvm.nvvm.barrier0(), !dbg !48
+  %130 = mul nuw nsw i32 %12, 130, !dbg !48
+  %131 = add nuw nsw i32 %130, %14, !dbg !48
+  %132 = zext nneg i32 %131 to i64, !dbg !48
+  %133 = getelementptr float, ptr addrspace(3) @global_smem, i64 %132, !dbg !48
+  %134 = load float, ptr addrspace(3) %133, align 8, !dbg !48
+  %135 = getelementptr inbounds <2 x float>, ptr addrspace(3) %133, i64 0, i64 1, !dbg !48
+  %136 = load float, ptr addrspace(3) %135, align 4, !dbg !48
+  tail call void @llvm.nvvm.barrier0(), !dbg !49
+  %137 = extractelement <2 x float> %118, i64 0, !dbg !51
+  %138 = extractelement <2 x float> %118, i64 1, !dbg !55
+  %139 = fsub float %138, %137, !dbg !55
+  %140 = fadd float %134, %136, !dbg !56
+  %141 = fcmp oeq float %140, 0.000000e+00, !dbg !57
+  %142 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %136, float %140) #6, !dbg !58
+  %143 = select i1 %141, float 0.000000e+00, float %142, !dbg !59
+  %144 = fmul float %139, %143, !dbg !60
+  %145 = fadd float %137, %144, !dbg !51
+  %shift = shufflevector <2 x float> %121, <2 x float> poison, <2 x i32> <i32 1, i32 poison>, !dbg !61
+  %146 = fadd <2 x float> %121, %shift, !dbg !61
+  %147 = extractelement <2 x float> %146, i64 0, !dbg !61
+  %148 = fmul float %139, %139, !dbg !62
+  %149 = fmul float %148, %134, !dbg !63
+  %150 = fmul float %149, %143, !dbg !64
+  %151 = fadd float %147, %150, !dbg !65
+  %152 = bitcast float %145 to i32, !dbg !49
+  %153 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %152, i32 16, i32 31), !dbg !49
+  %154 = bitcast i32 %153 to float, !dbg !49
+  %155 = bitcast float %151 to i32, !dbg !49
+  %156 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %155, i32 16, i32 31), !dbg !49
+  %157 = bitcast i32 %156 to float, !dbg !49
+  %158 = bitcast float %140 to i32, !dbg !49
+  %159 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %158, i32 16, i32 31), !dbg !49
+  %160 = bitcast i32 %159 to float, !dbg !49
+  %161 = fsub float %154, %145, !dbg !55
+  %162 = fadd float %140, %160, !dbg !56
+  %163 = fcmp oeq float %162, 0.000000e+00, !dbg !57
+  %164 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %160, float %162) #6, !dbg !58
+  %165 = select i1 %163, float 0.000000e+00, float %164, !dbg !59
+  %166 = fmul float %161, %165, !dbg !60
+  %167 = fadd float %145, %166, !dbg !51
+  %168 = fadd float %151, %157, !dbg !61
+  %169 = fmul float %161, %161, !dbg !62
+  %170 = fmul float %140, %169, !dbg !63
+  %171 = fmul float %170, %165, !dbg !64
+  %172 = fadd float %168, %171, !dbg !65
+  %173 = bitcast float %167 to i32, !dbg !49
+  %174 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %173, i32 8, i32 31), !dbg !49
+  %175 = bitcast i32 %174 to float, !dbg !49
+  %176 = bitcast float %172 to i32, !dbg !49
+  %177 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %176, i32 8, i32 31), !dbg !49
+  %178 = bitcast i32 %177 to float, !dbg !49
+  %179 = bitcast float %162 to i32, !dbg !49
+  %180 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %179, i32 8, i32 31), !dbg !49
+  %181 = bitcast i32 %180 to float, !dbg !49
+  %182 = fsub float %175, %167, !dbg !55
+  %183 = fadd float %162, %181, !dbg !56
+  %184 = fcmp oeq float %183, 0.000000e+00, !dbg !57
+  %185 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %181, float %183) #6, !dbg !58
+  %186 = select i1 %184, float 0.000000e+00, float %185, !dbg !59
+  %187 = fmul float %182, %186, !dbg !60
+  %188 = fadd float %167, %187, !dbg !51
+  %189 = fadd float %172, %178, !dbg !61
+  %190 = fmul float %182, %182, !dbg !62
+  %191 = fmul float %162, %190, !dbg !63
+  %192 = fmul float %186, %191, !dbg !64
+  %193 = fadd float %189, %192, !dbg !65
+  %194 = bitcast float %188 to i32, !dbg !49
+  %195 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %194, i32 4, i32 31), !dbg !49
+  %196 = bitcast i32 %195 to float, !dbg !49
+  %197 = bitcast float %193 to i32, !dbg !49
+  %198 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %197, i32 4, i32 31), !dbg !49
+  %199 = bitcast i32 %198 to float, !dbg !49
+  %200 = bitcast float %183 to i32, !dbg !49
+  %201 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %200, i32 4, i32 31), !dbg !49
+  %202 = bitcast i32 %201 to float, !dbg !49
+  %203 = fsub float %196, %188, !dbg !55
+  %204 = fadd float %183, %202, !dbg !56
+  %205 = fcmp oeq float %204, 0.000000e+00, !dbg !57
+  %206 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %202, float %204) #6, !dbg !58
+  %207 = select i1 %205, float 0.000000e+00, float %206, !dbg !59
+  %208 = fmul float %203, %207, !dbg !60
+  %209 = fadd float %188, %208, !dbg !51
+  %210 = fadd float %193, %199, !dbg !61
+  %211 = fmul float %203, %203, !dbg !62
+  %212 = fmul float %183, %211, !dbg !63
+  %213 = fmul float %207, %212, !dbg !64
+  %214 = fadd float %210, %213, !dbg !65
+  %215 = bitcast float %209 to i32, !dbg !49
+  %216 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %215, i32 2, i32 31), !dbg !49
+  %217 = bitcast i32 %216 to float, !dbg !49
+  %218 = bitcast float %214 to i32, !dbg !49
+  %219 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %218, i32 2, i32 31), !dbg !49
+  %220 = bitcast i32 %219 to float, !dbg !49
+  %221 = bitcast float %204 to i32, !dbg !49
+  %222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %221, i32 2, i32 31), !dbg !49
+  %223 = bitcast i32 %222 to float, !dbg !49
+  %224 = fsub float %217, %209, !dbg !55
+  %225 = fadd float %204, %223, !dbg !56
+  %226 = fcmp oeq float %225, 0.000000e+00, !dbg !57
+  %227 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %223, float %225) #6, !dbg !58
+  %228 = select i1 %226, float 0.000000e+00, float %227, !dbg !59
+  %229 = fmul float %224, %228, !dbg !60
+  %230 = fadd float %209, %229, !dbg !51
+  %231 = fadd float %214, %220, !dbg !61
+  %232 = fmul float %224, %224, !dbg !62
+  %233 = fmul float %204, %232, !dbg !63
+  %234 = fmul float %228, %233, !dbg !64
+  %235 = fadd float %231, %234, !dbg !65
+  %236 = bitcast float %230 to i32, !dbg !49
+  %237 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %236, i32 1, i32 31), !dbg !49
+  %238 = bitcast i32 %237 to float, !dbg !49
+  %239 = bitcast float %235 to i32, !dbg !49
+  %240 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %239, i32 1, i32 31), !dbg !49
+  %241 = bitcast i32 %240 to float, !dbg !49
+  %242 = bitcast float %225 to i32, !dbg !49
+  %243 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %242, i32 1, i32 31), !dbg !49
+  %244 = bitcast i32 %243 to float, !dbg !49
+  %245 = fsub float %238, %230, !dbg !55
+  %246 = fadd float %225, %244, !dbg !56
+  %247 = fcmp oeq float %246, 0.000000e+00, !dbg !57
+  %248 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %244, float %246) #6, !dbg !58
+  %249 = select i1 %247, float 0.000000e+00, float %248, !dbg !59
+  %250 = fmul float %245, %249, !dbg !60
+  %251 = fadd float %230, %250, !dbg !51
+  %252 = fadd float %235, %241, !dbg !61
+  %253 = fmul float %245, %245, !dbg !62
+  %254 = fmul float %225, %253, !dbg !63
+  %255 = fmul float %249, %254, !dbg !64
+  %256 = fadd float %252, %255, !dbg !65
+  %257 = icmp eq i32 %10, 0, !dbg !49
+  %258 = shl nuw nsw i32 %12, 1, !dbg !49
+  %259 = or i32 %258, %123, !dbg !49
+  %260 = zext nneg i32 %259 to i64, !dbg !49
+  %261 = getelementptr float, ptr addrspace(3) @global_smem, i64 %260, !dbg !49
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %261, float %251, i1 %257) #6, !dbg !49
+  %262 = getelementptr float, ptr addrspace(3) getelementptr ([0 x i8], ptr addrspace(3) @global_smem, i64 0, i64 16), i64 %260, !dbg !49
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %262, float %256, i1 %257) #6, !dbg !49
+  %263 = getelementptr float, ptr addrspace(3) getelementptr ([0 x i8], ptr addrspace(3) @global_smem, i64 0, i64 32), i64 %260, !dbg !49
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %263, float %246, i1 %257) #6, !dbg !49
+  tail call void @llvm.nvvm.barrier0(), !dbg !49
+  %264 = icmp slt i32 %9, 4, !dbg !49
+  %265 = sext i32 %9 to i64, !dbg !49
+  %266 = getelementptr float, ptr addrspace(3) @global_smem, i64 %265, !dbg !49
+  %267 = tail call float asm sideeffect "@$2 ld.shared.b32 $0, [ $1 + 0 ];", "=r,r,b"(ptr addrspace(3) %266, i1 %264) #6, !dbg !49
+  %268 = getelementptr float, ptr addrspace(3) getelementptr ([0 x i8], ptr addrspace(3) @global_smem, i64 0, i64 16), i64 %265, !dbg !49
+  %269 = tail call float asm sideeffect "@$2 ld.shared.b32 $0, [ $1 + 0 ];", "=r,r,b"(ptr addrspace(3) %268, i1 %264) #6, !dbg !49
+  %270 = getelementptr float, ptr addrspace(3) getelementptr ([0 x i8], ptr addrspace(3) @global_smem, i64 0, i64 32), i64 %265, !dbg !49
+  %271 = tail call float asm sideeffect "@$2 ld.shared.b32 $0, [ $1 + 0 ];", "=r,r,b"(ptr addrspace(3) %270, i1 %264) #6, !dbg !49
+  %272 = bitcast float %267 to i32, !dbg !49
+  %273 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %272, i32 1, i32 31), !dbg !49
+  %274 = bitcast i32 %273 to float, !dbg !49
+  %275 = bitcast float %269 to i32, !dbg !49
+  %276 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %275, i32 1, i32 31), !dbg !49
+  %277 = bitcast i32 %276 to float, !dbg !49
+  %278 = bitcast float %271 to i32, !dbg !49
+  %279 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %278, i32 1, i32 31), !dbg !49
+  %280 = bitcast i32 %279 to float, !dbg !49
+  %281 = fsub float %274, %267, !dbg !55
+  %282 = fadd float %271, %280, !dbg !56
+  %283 = fcmp oeq float %282, 0.000000e+00, !dbg !57
+  %284 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %280, float %282) #6, !dbg !58
+  %285 = select i1 %283, float 0.000000e+00, float %284, !dbg !59
+  %286 = fmul float %281, %285, !dbg !60
+  %287 = fadd float %267, %286, !dbg !51
+  %288 = fadd float %269, %277, !dbg !61
+  %289 = fmul float %281, %281, !dbg !62
+  %290 = fmul float %271, %289, !dbg !63
+  %291 = fmul float %290, %285, !dbg !64
+  %292 = fadd float %288, %291, !dbg !65
+  %293 = icmp eq i32 %13, 0, !dbg !49
+  %294 = and i1 %264, %293, !dbg !49
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %266, float %287, i1 %294) #6, !dbg !49
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %268, float %292, i1 %294) #6, !dbg !49
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %270, float %282, i1 %294) #6, !dbg !49
+  tail call void @llvm.nvvm.barrier0(), !dbg !49
+  %295 = zext nneg i32 %258 to i64, !dbg !49
+  %296 = getelementptr float, ptr addrspace(3) @global_smem, i64 %295, !dbg !49
+  %297 = load float, ptr addrspace(3) %296, align 4, !dbg !49
+  %298 = getelementptr float, ptr addrspace(3) getelementptr ([0 x i8], ptr addrspace(3) @global_smem, i64 0, i64 16), i64 %295, !dbg !49
+  %299 = load float, ptr addrspace(3) %298, align 4, !dbg !49
+  %300 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %299, float 2.560000e+02) #6, !dbg !66
+  %301 = tail call float asm "div.full.f32 $0, $1, $2;", "=r,r,r"(float %299, float 2.560000e+02) #6, !dbg !66
+  %302 = fadd float %300, 0x3EE4F8B580000000, !dbg !67
+  %303 = getelementptr float, ptr addrspace(3) @global_smem, i64 %57
+  %304 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_last.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %40, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !68
+  %305 = extractvalue { i32, i32 } %304, 0, !dbg !68
+  %306 = extractvalue { i32, i32 } %304, 1, !dbg !68
+  %307 = bitcast i32 %305 to float, !dbg !68
+  %308 = bitcast i32 %306 to float, !dbg !68
+  %309 = tail call i32 asm sideeffect "mov.u32 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b32 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u32 $0, $3;", "=r,l,b,r,b"(ptr addrspace(1) %49, i1 true, i32 0, i1 true) #6, !dbg !69
+  %310 = trunc i32 %309 to i16, !dbg !69
+  %extelt.offset = lshr i32 %309, 16, !dbg !69
+  %311 = trunc i32 %extelt.offset to i16, !dbg !69
+  %312 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %310) #6, !dbg !70
+  %313 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %311) #6, !dbg !70
+  %314 = getelementptr float, ptr addrspace(1) %4, i64 %125, !dbg !71
+  %315 = tail call i32 asm sideeffect "mov.u32 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b32 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u32 $0, $3;", "=r,l,b,r,b"(ptr addrspace(1) %314, i1 true, i32 0, i1 true) #6, !dbg !72
+  br i1 %33, label %316, label %317, !dbg !73
+
+316:                                              ; preds = %98
+  tail call void @__assertfail(ptr nonnull @assertMessage_1, ptr nonnull @assertFile_1, i32 1892, ptr nonnull @assertFunc_1, i64 1), !dbg !73
+  br label %317, !dbg !73
+
+317:                                              ; preds = %316, %98
+  %318 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_first.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %58, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !74
+  %319 = extractvalue { i32, i32 } %318, 0, !dbg !74
+  %320 = extractvalue { i32, i32 } %318, 1, !dbg !74
+  %321 = bitcast i32 %319 to float, !dbg !74
+  %322 = bitcast i32 %320 to float, !dbg !74
+  %323 = fadd float %307, %321, !dbg !75
+  %324 = fadd float %308, %322, !dbg !75
+  %325 = fadd float %312, %323, !dbg !76
+  %326 = fadd float %313, %324, !dbg !76
+  %327 = fsub float %325, %297, !dbg !77
+  %328 = fsub float %326, %297, !dbg !77
+  %329 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !78
+  %.not.i = icmp eq i32 %329, 0, !dbg !78
+  br i1 %.not.i, label %332, label %330, !dbg !78
+
+330:                                              ; preds = %317
+  %331 = tail call float @llvm.nvvm.rsqrt.approx.ftz.f(float %302), !dbg !78
+  br label %__nv_rsqrtf.exit, !dbg !78
+
+332:                                              ; preds = %317
+  %333 = tail call float @llvm.nvvm.rsqrt.approx.f(float %302), !dbg !78
+  br label %__nv_rsqrtf.exit, !dbg !78
+
+__nv_rsqrtf.exit:                                 ; preds = %330, %332
+  %.0.i = phi float [ %331, %330 ], [ %333, %332 ], !dbg !78
+  %334 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !78
+  %335 = fmul float %327, %.0.i, !dbg !79
+  %336 = fmul float %328, %.0.i, !dbg !79
+  tail call void @llvm.nvvm.barrier0(), !dbg !80
+  store i32 %315, ptr addrspace(3) %126, align 4, !dbg !80
+  tail call void @llvm.nvvm.barrier0(), !dbg !80
+  %337 = load float, ptr addrspace(3) %303, align 8, !dbg !80
+  %338 = getelementptr inbounds <2 x float>, ptr addrspace(3) %303, i64 0, i64 1, !dbg !80
+  %339 = load float, ptr addrspace(3) %338, align 4, !dbg !80
+  %340 = fmul float %335, %337, !dbg !80
+  %341 = fmul float %336, %339, !dbg !80
+  %342 = getelementptr i16, ptr addrspace(1) %5, i64 %48, !dbg !81
+  %343 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %340) #6, !dbg !82
+  %344 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %341) #6, !dbg !82
+  %345 = insertelement <2 x i16> undef, i16 %343, i64 0, !dbg !82
+  %346 = insertelement <2 x i16> %345, i16 %344, i64 1, !dbg !82
+  %347 = bitcast <2 x i16> %346 to i32, !dbg !82
+  tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %347, ptr addrspace(1) %342, i1 true) #6, !dbg !82
+  %348 = or i32 %124, 128, !dbg !83
+  %349 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_last.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %82, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !68
+  %350 = tail call i32 asm sideeffect "mov.u32 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b32 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u32 $0, $3;", "=r,l,b,r,b"(ptr addrspace(1) %91, i1 true, i32 0, i1 true) #6, !dbg !69
+  %351 = trunc i32 %350 to i16, !dbg !69
+  %extelt.offset.1 = lshr i32 %350, 16, !dbg !69
+  %352 = trunc i32 %extelt.offset.1 to i16, !dbg !69
+  %353 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %351) #6, !dbg !70
+  %354 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %352) #6, !dbg !70
+  %355 = zext nneg i32 %348 to i64, !dbg !71
+  %356 = getelementptr float, ptr addrspace(1) %4, i64 %355, !dbg !71
+  %357 = tail call i32 asm sideeffect "mov.u32 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b32 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u32 $0, $3;", "=r,l,b,r,b"(ptr addrspace(1) %356, i1 true, i32 0, i1 true) #6, !dbg !72
+  br i1 %33, label %358, label %359, !dbg !73
+
+358:                                              ; preds = %__nv_rsqrtf.exit
+  tail call void @__assertfail(ptr nonnull @assertMessage_1, ptr nonnull @assertFile_1, i32 1892, ptr nonnull @assertFunc_1, i64 1), !dbg !73
+  br label %359, !dbg !73
+
+359:                                              ; preds = %358, %__nv_rsqrtf.exit
+  %360 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, 0x0;\0A\09mov.u32 $1, 0x0;\0A\09@$3 ld.global.L1::evict_first.v2.b32 { $0, $1 }, [ $2 + 0 ];\0A\09@!$5 mov.u32 $0, $4;\0A\09@!$7 mov.u32 $1, $6;", "=r,=r,l,b,r,b,r,b"(ptr addrspace(1) %100, i1 true, i32 0, i1 true, i32 0, i1 true) #6, !dbg !74
+  %361 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !78
+  %.not.i.1 = icmp eq i32 %361, 0, !dbg !78
+  br i1 %.not.i.1, label %364, label %362, !dbg !78
+
+362:                                              ; preds = %359
+  %363 = tail call float @llvm.nvvm.rsqrt.approx.ftz.f(float %302), !dbg !78
+  br label %__nv_rsqrtf.exit.1, !dbg !78
+
+364:                                              ; preds = %359
+  %365 = tail call float @llvm.nvvm.rsqrt.approx.f(float %302), !dbg !78
+  br label %__nv_rsqrtf.exit.1, !dbg !78
+
+__nv_rsqrtf.exit.1:                               ; preds = %364, %362
+  %.0.i.1 = phi float [ %363, %362 ], [ %365, %364 ], !dbg !78
+  %366 = extractvalue { i32, i32 } %349, 1, !dbg !68
+  %367 = bitcast i32 %366 to float, !dbg !68
+  %368 = extractvalue { i32, i32 } %360, 1, !dbg !74
+  %369 = bitcast i32 %368 to float, !dbg !74
+  %370 = fadd float %367, %369, !dbg !75
+  %371 = fadd float %354, %370, !dbg !76
+  %372 = fsub float %371, %297, !dbg !77
+  %373 = extractvalue { i32, i32 } %349, 0, !dbg !68
+  %374 = bitcast i32 %373 to float, !dbg !68
+  %375 = extractvalue { i32, i32 } %360, 0, !dbg !74
+  %376 = bitcast i32 %375 to float, !dbg !74
+  %377 = fadd float %374, %376, !dbg !75
+  %378 = fadd float %353, %377, !dbg !76
+  %379 = fsub float %378, %297, !dbg !77
+  %380 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !78
+  %381 = fmul float %379, %.0.i.1, !dbg !79
+  %382 = fmul float %372, %.0.i.1, !dbg !79
+  tail call void @llvm.nvvm.barrier0(), !dbg !80
+  store i32 %357, ptr addrspace(3) %126, align 4, !dbg !80
+  tail call void @llvm.nvvm.barrier0(), !dbg !80
+  %383 = load float, ptr addrspace(3) %303, align 8, !dbg !80
+  %384 = load float, ptr addrspace(3) %338, align 4, !dbg !80
+  %385 = fmul float %381, %383, !dbg !80
+  %386 = fmul float %382, %384, !dbg !80
+  %387 = getelementptr i16, ptr addrspace(1) %5, i64 %90, !dbg !81
+  %388 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %385) #6, !dbg !82
+  %389 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %386) #6, !dbg !82
+  %390 = insertelement <2 x i16> undef, i16 %388, i64 0, !dbg !82
+  %391 = insertelement <2 x i16> %390, i16 %389, i64 1, !dbg !82
+  %392 = bitcast <2 x i16> %391 to i32, !dbg !82
+  tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %392, ptr addrspace(1) %387, i1 true) #6, !dbg !82
+  ret void, !dbg !84
+}
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare noundef i32 @llvm.nvvm.read.ptx.sreg.tid.x() #0
+
+; Function Attrs: convergent nocallback nounwind
+declare void @llvm.nvvm.barrier0() #1
+
+; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite)
+declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2
+
+; Function Attrs: alwaysinline nounwind
+define float @__nv_rsqrtf(float %x) local_unnamed_addr #3 {
+  %1 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6
+  %.not = icmp eq i32 %1, 0
+  br i1 %.not, label %4, label %2
+
+2:                                                ; preds = %0
+  %3 = tail call float @llvm.nvvm.rsqrt.approx.ftz.f(float %x)
+  br label %6
+
+4:                                                ; preds = %0
+  %5 = tail call float @llvm.nvvm.rsqrt.approx.f(float %x)
+  br label %6
+
+6:                                                ; preds = %4, %2
+  %.0 = phi float [ %3, %2 ], [ %5, %4 ]
+  ret float %.0
+}
+
+declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #4
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none)
+declare float @llvm.nvvm.rsqrt.approx.ftz.f(float) #5
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none)
+declare float @llvm.nvvm.rsqrt.approx.f(float) #5
+
+attributes #0 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #1 = { convergent nocallback nounwind }
+attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) }
+attributes #3 = { alwaysinline nounwind "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #4 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #5 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0, !1}
+!llvm.dbg.cu = !{!2}
+!nvvm.annotations = !{!4, !5, !5, !4}
+!llvm.ident = !{!6}
+
+!0 = !{i32 2, !"Debug Info Version", i32 3}
+!1 = !{i32 4, !"nvvm-reflect-ftz", i32 1}
+!2 = distinct !DICompileUnit(language: DW_LANG_C, file: !3, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug)
+!3 = !DIFile(filename: "ccig6fki6p4lxrdmgg6eudahiexcvueeol2p4qp532pvve2y463y.py", directory: "/tmp/torchinductor_root/ci")
+!4 = !{ptr @triton__0d1d2d3d4d5d6de7de, !"kernel", i32 1}
+!5 = !{ptr @triton__0d1d2d3d4d5d6de7de, !"maxntidx", i32 128}
+!6 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"}
+!7 = distinct !DISubprogram(name: "triton__0d1d2d3d4d5d6de7de", linkageName: "triton__0d1d2d3d4d5d6de7de", scope: !3, file: !3, line: 18, type: !8, scopeLine: 18, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
+!8 = !DISubroutineType(cc: DW_CC_normal, types: !9)
+!9 = !{}
+!10 = !DILocation(line: 22, column: 44, scope: !7)
+!11 = !DILocation(line: 24, column: 33, scope: !7)
+!12 = !DILocation(line: 21, column: 28, scope: !7)
+!13 = !DILocation(line: 21, column: 33, scope: !7)
+!14 = !DILocation(line: 22, column: 23, scope: !7)
+!15 = !DILocation(line: 26, column: 30, scope: !7)
+!16 = !DILocation(line: 26, column: 35, scope: !7)
+!17 = !DILocation(line: 27, column: 18, scope: !7)
+!18 = !DILocation(line: 35, column: 44, scope: !7)
+!19 = !DILocation(line: 36, column: 44, scope: !7)
+!20 = !DILocation(line: 37, column: 22, scope: !7)
+!21 = !DILocation(line: 38, column: 22, scope: !7)
+!22 = !DILocation(line: 39, column: 36, scope: !7)
+!23 = !DILocation(line: 40, column: 40, scope: !7)
+!24 = !DILocation(line: 41, column: 44, scope: !7)
+!25 = !DILocation(line: 35, column: 40, scope: !7)
+!26 = !DILocation(line: 35, column: 34, scope: !7)
+!27 = !DILocation(line: 35, column: 50, scope: !7)
+!28 = !DILocation(line: 36, column: 40, scope: !7)
+!29 = !DILocation(line: 36, column: 34, scope: !7)
+!30 = !DILocation(line: 36, column: 50, scope: !7)
+!31 = !DILocation(line: 36, column: 101, scope: !7)
+!32 = !DILocation(line: 40, column: 55, scope: !7)
+!33 = !DILocation(line: 41, column: 40, scope: !7)
+!34 = !DILocation(line: 41, column: 34, scope: !7)
+!35 = !DILocation(line: 41, column: 52, scope: !7)
+!36 = !DILocation(line: 42, column: 22, scope: !7)
+!37 = !DILocation(line: 44, column: 22, scope: !7)
+!38 = !DILocation(line: 98, column: 30, scope: !39, inlinedAt: !41)
+!39 = distinct !DILexicalBlockFile(scope: !7, file: !40, discriminator: 0)
+!40 = !DIFile(filename: "triton_helpers.py", directory: "/usr/local/lib/python3.10/dist-packages/torch/_inductor")
+!41 = !DILocation(line: 47, column: 41, scope: !39)
+!42 = !DILocation(line: 98, column: 22, scope: !39, inlinedAt: !41)
+!43 = !DILocation(line: 101, column: 30, scope: !39, inlinedAt: !41)
+!44 = !DILocation(line: 101, column: 22, scope: !39, inlinedAt: !41)
+!45 = !DILocation(line: 50, column: 50, scope: !7)
+!46 = !DILocation(line: 32, column: 27, scope: !7)
+!47 = !DILocation(line: 96, column: 20, scope: !39, inlinedAt: !41)
+!48 = !DILocation(line: 31, column: 36, scope: !7)
+!49 = !DILocation(line: 120, column: 46, scope: !39, inlinedAt: !50)
+!50 = !DILocation(line: 53, column: 44, scope: !39)
+!51 = !DILocation(line: 112, column: 17, scope: !52, inlinedAt: !53)
+!52 = distinct !DILexicalBlockFile(scope: !39, file: !40, discriminator: 0)
+!53 = !DILocation(line: 120, column: 46, scope: !52, inlinedAt: !54)
+!54 = !DILocation(line: 53, column: 44, scope: !52)
+!55 = !DILocation(line: 108, column: 21, scope: !52, inlinedAt: !53)
+!56 = !DILocation(line: 109, column: 28, scope: !52, inlinedAt: !53)
+!57 = !DILocation(line: 110, column: 39, scope: !52, inlinedAt: !53)
+!58 = !DILocation(line: 110, column: 60, scope: !52, inlinedAt: !53)
+!59 = !DILocation(line: 110, column: 49, scope: !52, inlinedAt: !53)
+!60 = !DILocation(line: 112, column: 25, scope: !52, inlinedAt: !53)
+!61 = !DILocation(line: 113, column: 15, scope: !52, inlinedAt: !53)
+!62 = !DILocation(line: 113, column: 30, scope: !52, inlinedAt: !53)
+!63 = !DILocation(line: 113, column: 38, scope: !52, inlinedAt: !53)
+!64 = !DILocation(line: 113, column: 49, scope: !52, inlinedAt: !53)
+!65 = !DILocation(line: 113, column: 22, scope: !52, inlinedAt: !53)
+!66 = !DILocation(line: 75, column: 24, scope: !7)
+!67 = !DILocation(line: 77, column: 24, scope: !7)
+!68 = !DILocation(line: 62, column: 51, scope: !7)
+!69 = !DILocation(line: 63, column: 51, scope: !7)
+!70 = !DILocation(line: 63, column: 103, scope: !7)
+!71 = !DILocation(line: 64, column: 35, scope: !7)
+!72 = !DILocation(line: 64, column: 40, scope: !7)
+!73 = !DILocation(line: 68, column: 57, scope: !7)
+!74 = !DILocation(line: 69, column: 54, scope: !7)
+!75 = !DILocation(line: 70, column: 24, scope: !7)
+!76 = !DILocation(line: 72, column: 24, scope: !7)
+!77 = !DILocation(line: 73, column: 24, scope: !7)
+!78 = !DILocation(line: 78, column: 30, scope: !7)
+!79 = !DILocation(line: 79, column: 24, scope: !7)
+!80 = !DILocation(line: 80, column: 24, scope: !7)
+!81 = !DILocation(line: 82, column: 29, scope: !7)
+!82 = !DILocation(line: 82, column: 52, scope: !7)
+!83 = !DILocation(line: 59, column: 27, scope: !7)
+!84 = !DILocation(line: 58, column: 4, scope: !7)

.triton/dump/76fb48b96c75cb8e388c291a18ef9b02/triton_.ttir

@@ -0,0 +1,153 @@
+module {
+  tt.func public @triton__0d1d2d3d4d5d6de7de(%arg0: !tt.ptr<i64, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg3: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg4: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg5: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg6: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}, %arg7: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %cst = arith.constant dense<0.000000e+00> : tensor<2x128xbf16>
+    %cst_0 = arith.constant 0.000000e+00 : f32
+    %cst_1 = arith.constant dense<1.000000e+00> : tensor<2x128xf32>
+    %c256_i32 = arith.constant 256 : i32
+    %c128_i32 = arith.constant 128 : i32
+    %c0_i32 = arith.constant 0 : i32
+    %cst_2 = arith.constant dense<256> : tensor<2x1xi64>
+    %cst_3 = arith.constant dense<0> : tensor<2x1xi64>
+    %cst_4 = arith.constant dense<50257> : tensor<2x1xi64>
+    %cst_5 = arith.constant dense<9.99999974E-6> : tensor<2x1xf32>
+    %cst_6 = arith.constant dense<2.560000e+02> : tensor<2x1xf32>
+    %cst_7 = arith.constant dense<0.000000e+00> : tensor<1x128xf32>
+    %cst_8 = arith.constant dense<0.000000e+00> : tensor<2x128xf32>
+    %cst_9 = arith.constant dense<256> : tensor<2x1xi32>
+    %cst_10 = arith.constant dense<256> : tensor<1x128xi32>
+    %cst_11 = arith.constant dense<512> : tensor<2x1xi32>
+    %c2_i32 = arith.constant 2 : i32
+    %0 = tt.get_program_id x : i32
+    %1 = arith.muli %0, %c2_i32 : i32
+    %2 = tt.make_range {end = 2 : i32, start = 0 : i32} : tensor<2xi32>
+    %3 = tt.expand_dims %2 {axis = 1 : i32} : (tensor<2xi32>) -> tensor<2x1xi32>
+    %4 = tt.splat %1 : (i32) -> tensor<2x1xi32>
+    %5 = arith.addi %4, %3 : tensor<2x1xi32>
+    %6 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32>
+    %7 = tt.expand_dims %6 {axis = 0 : i32} : (tensor<128xi32>) -> tensor<1x128xi32>
+    %8 = tt.splat %arg0 : (!tt.ptr<i64, 1>) -> tensor<2x1x!tt.ptr<i64, 1>>
+    %9 = tt.addptr %8, %5 : tensor<2x1x!tt.ptr<i64, 1>>, tensor<2x1xi32>
+    %10 = tt.load %9 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<2x1xi64>
+    %11 = arith.remsi %5, %cst_11 : tensor<2x1xi32>
+    %12 = arith.muli %11, %cst_9 : tensor<2x1xi32>
+    %13 = tt.broadcast %12 : (tensor<2x1xi32>) -> tensor<2x128xi32>
+    %14 = tt.splat %arg2 : (!tt.ptr<f32, 1>) -> tensor<2x128x!tt.ptr<f32, 1>>
+    %15 = arith.muli %5, %cst_9 : tensor<2x1xi32>
+    %16 = tt.broadcast %15 : (tensor<2x1xi32>) -> tensor<2x128xi32>
+    %17 = tt.splat %arg3 : (!tt.ptr<bf16, 1>) -> tensor<2x128x!tt.ptr<bf16, 1>>
+    %18 = arith.addi %10, %cst_4 : tensor<2x1xi64>
+    %19 = arith.cmpi slt, %10, %cst_3 : tensor<2x1xi64>
+    %20 = arith.select %19, %18, %10 : tensor<2x1xi1>, tensor<2x1xi64>
+    %21 = arith.cmpi sge, %20, %cst_3 : tensor<2x1xi64>
+    %22 = arith.cmpi slt, %20, %cst_4 : tensor<2x1xi64>
+    %23 = arith.andi %21, %22 : tensor<2x1xi1>
+    %24 = arith.muli %20, %cst_2 : tensor<2x1xi64>
+    %25 = tt.broadcast %24 : (tensor<2x1xi64>) -> tensor<2x128xi64>
+    %26 = tt.splat %arg1 : (!tt.ptr<f32, 1>) -> tensor<2x128x!tt.ptr<f32, 1>>
+    %27:3 = scf.for %arg8 = %c0_i32 to %c256_i32 step %c128_i32 iter_args(%arg9 = %cst_8, %arg10 = %cst_8, %arg11 = %cst_8) -> (tensor<2x128xf32>, tensor<2x128xf32>, tensor<2x128xf32>)  : i32 {
+      %51 = tt.splat %arg8 : (i32) -> tensor<1x128xi32>
+      %52 = arith.addi %51, %7 : tensor<1x128xi32>
+      %53 = arith.cmpi slt, %52, %cst_10 : tensor<1x128xi32>
+      %54 = tt.broadcast %52 : (tensor<1x128xi32>) -> tensor<2x128xi32>
+      %55 = arith.addi %54, %13 : tensor<2x128xi32>
+      %56 = tt.addptr %14, %55 : tensor<2x128x!tt.ptr<f32, 1>>, tensor<2x128xi32>
+      %57 = tt.broadcast %53 : (tensor<1x128xi1>) -> tensor<2x128xi1>
+      %58 = tt.load %56, %57, %cst_8 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<2x128xf32>
+      %59 = arith.addi %54, %16 : tensor<2x128xi32>
+      %60 = tt.addptr %17, %59 : tensor<2x128x!tt.ptr<bf16, 1>>, tensor<2x128xi32>
+      %61 = tt.load %60, %57, %cst {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<2x128xbf16>
+      %62 = arith.extf %61 : tensor<2x128xbf16> to tensor<2x128xf32>
+      tt.assert %23, "index out of bounds: 0 <= tmp3 < 50257", "/usr/local/lib/python3.10/dist-packages/torch/_inductor/codecache.py", "<module>", 1892 : tensor<2x1xi1>
+      %63 = arith.extsi %52 : tensor<1x128xi32> to tensor<1x128xi64>
+      %64 = tt.broadcast %63 : (tensor<1x128xi64>) -> tensor<2x128xi64>
+      %65 = arith.addi %64, %25 : tensor<2x128xi64>
+      %66 = tt.addptr %26, %65 : tensor<2x128x!tt.ptr<f32, 1>>, tensor<2x128xi64>
+      %67 = tt.load %66, %57, %cst_8 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<2x128xf32>
+      %68 = arith.addf %67, %58 : tensor<2x128xf32>
+      %69 = arith.addf %68, %62 : tensor<2x128xf32>
+      %70 = arith.subf %69, %arg9 : tensor<2x128xf32>
+      %71 = arith.addf %arg11, %cst_1 : tensor<2x128xf32>
+      %72 = arith.divf %70, %71 : tensor<2x128xf32>
+      %73 = arith.addf %arg9, %72 : tensor<2x128xf32>
+      %74 = arith.subf %69, %73 : tensor<2x128xf32>
+      %75 = arith.mulf %70, %74 : tensor<2x128xf32>
+      %76 = arith.addf %arg10, %75 : tensor<2x128xf32>
+      %77 = arith.select %57, %73, %arg9 : tensor<2x128xi1>, tensor<2x128xf32>
+      %78 = arith.select %57, %76, %arg10 : tensor<2x128xi1>, tensor<2x128xf32>
+      %79 = arith.select %57, %71, %arg11 : tensor<2x128xi1>, tensor<2x128xf32>
+      scf.yield %77, %78, %79 : tensor<2x128xf32>, tensor<2x128xf32>, tensor<2x128xf32>
+    }
+    %28:3 = "tt.reduce"(%27#0, %27#1, %27#2) <{axis = 1 : i32}> ({
+    ^bb0(%arg8: f32, %arg9: f32, %arg10: f32, %arg11: f32, %arg12: f32, %arg13: f32):
+      %51 = arith.subf %arg11, %arg8 : f32
+      %52 = arith.addf %arg10, %arg13 : f32
+      %53 = arith.cmpf oeq, %52, %cst_0 : f32
+      %54 = arith.divf %arg13, %52 : f32
+      %55 = arith.select %53, %cst_0, %54 : f32
+      %56 = arith.mulf %51, %55 : f32
+      %57 = arith.addf %arg8, %56 : f32
+      %58 = arith.addf %arg9, %arg12 : f32
+      %59 = arith.mulf %51, %51 : f32
+      %60 = arith.mulf %59, %arg10 : f32
+      %61 = arith.mulf %60, %55 : f32
+      %62 = arith.addf %58, %61 : f32
+      tt.reduce.return %57, %62, %52 : f32, f32, f32
+    }) : (tensor<2x128xf32>, tensor<2x128xf32>, tensor<2x128xf32>) -> (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>)
+    %29 = tt.expand_dims %28#0 {axis = 1 : i32} : (tensor<2xf32>) -> tensor<2x1xf32>
+    %30 = tt.expand_dims %28#1 {axis = 1 : i32} : (tensor<2xf32>) -> tensor<2x1xf32>
+    %31 = arith.muli %11, %cst_9 : tensor<2x1xi32>
+    %32 = tt.broadcast %31 : (tensor<2x1xi32>) -> tensor<2x128xi32>
+    %33 = tt.splat %arg2 : (!tt.ptr<f32, 1>) -> tensor<2x128x!tt.ptr<f32, 1>>
+    %34 = arith.muli %5, %cst_9 : tensor<2x1xi32>
+    %35 = tt.broadcast %34 : (tensor<2x1xi32>) -> tensor<2x128xi32>
+    %36 = tt.splat %arg3 : (!tt.ptr<bf16, 1>) -> tensor<2x128x!tt.ptr<bf16, 1>>
+    %37 = tt.splat %arg4 : (!tt.ptr<f32, 1>) -> tensor<1x128x!tt.ptr<f32, 1>>
+    %38 = arith.addi %10, %cst_4 : tensor<2x1xi64>
+    %39 = arith.cmpi slt, %10, %cst_3 : tensor<2x1xi64>
+    %40 = arith.select %39, %38, %10 : tensor<2x1xi1>, tensor<2x1xi64>
+    %41 = arith.cmpi sge, %40, %cst_3 : tensor<2x1xi64>
+    %42 = arith.cmpi slt, %40, %cst_4 : tensor<2x1xi64>
+    %43 = arith.andi %41, %42 : tensor<2x1xi1>
+    %44 = arith.muli %40, %cst_2 : tensor<2x1xi64>
+    %45 = tt.broadcast %44 : (tensor<2x1xi64>) -> tensor<2x128xi64>
+    %46 = tt.splat %arg1 : (!tt.ptr<f32, 1>) -> tensor<2x128x!tt.ptr<f32, 1>>
+    %47 = tt.broadcast %29 : (tensor<2x1xf32>) -> tensor<2x128xf32>
+    %48 = arith.divf %30, %cst_6 : tensor<2x1xf32>
+    %49 = arith.addf %48, %cst_5 : tensor<2x1xf32>
+    %50 = tt.splat %arg5 : (!tt.ptr<bf16, 1>) -> tensor<2x128x!tt.ptr<bf16, 1>>
+    scf.for %arg8 = %c0_i32 to %c256_i32 step %c128_i32  : i32 {
+      %51 = tt.splat %arg8 : (i32) -> tensor<1x128xi32>
+      %52 = arith.addi %51, %7 : tensor<1x128xi32>
+      %53 = arith.cmpi slt, %52, %cst_10 : tensor<1x128xi32>
+      %54 = tt.broadcast %52 : (tensor<1x128xi32>) -> tensor<2x128xi32>
+      %55 = arith.addi %54, %32 : tensor<2x128xi32>
+      %56 = tt.addptr %33, %55 : tensor<2x128x!tt.ptr<f32, 1>>, tensor<2x128xi32>
+      %57 = tt.broadcast %53 : (tensor<1x128xi1>) -> tensor<2x128xi1>
+      %58 = tt.load %56, %57, %cst_8 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<2x128xf32>
+      %59 = arith.addi %54, %35 : tensor<2x128xi32>
+      %60 = tt.addptr %36, %59 : tensor<2x128x!tt.ptr<bf16, 1>>, tensor<2x128xi32>
+      %61 = tt.load %60, %57, %cst {cache = 1 : i32, evict = 2 : i32, isVolatile = false} : tensor<2x128xbf16>
+      %62 = arith.extf %61 : tensor<2x128xbf16> to tensor<2x128xf32>
+      %63 = tt.addptr %37, %52 : tensor<1x128x!tt.ptr<f32, 1>>, tensor<1x128xi32>
+      %64 = tt.load %63, %53, %cst_7 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<1x128xf32>
+      tt.assert %43, "index out of bounds: 0 <= tmp16 < 50257", "/usr/local/lib/python3.10/dist-packages/torch/_inductor/codecache.py", "<module>", 1892 : tensor<2x1xi1>
+      %65 = arith.extsi %52 : tensor<1x128xi32> to tensor<1x128xi64>
+      %66 = tt.broadcast %65 : (tensor<1x128xi64>) -> tensor<2x128xi64>
+      %67 = arith.addi %66, %45 : tensor<2x128xi64>
+      %68 = tt.addptr %46, %67 : tensor<2x128x!tt.ptr<f32, 1>>, tensor<2x128xi64>
+      %69 = tt.load %68, %57, %cst_8 {cache = 1 : i32, evict = 2 : i32, isVolatile = false} : tensor<2x128xf32>
+      %70 = arith.addf %69, %58 : tensor<2x128xf32>
+      %71 = arith.addf %70, %62 : tensor<2x128xf32>
+      %72 = arith.subf %71, %47 : tensor<2x128xf32>
+      %73 = tt.extern_elementwise %49 {libname = "libdevice", libpath = "/usr/local/lib/python3.10/dist-packages/triton/language/../third_party/cuda/lib/libdevice.10.bc", pure = true, symbol = "__nv_rsqrtf"} : (tensor<2x1xf32>) -> tensor<2x1xf32>
+      %74 = tt.broadcast %73 : (tensor<2x1xf32>) -> tensor<2x128xf32>
+      %75 = arith.mulf %72, %74 : tensor<2x128xf32>
+      %76 = tt.broadcast %64 : (tensor<1x128xf32>) -> tensor<2x128xf32>
+      %77 = arith.mulf %75, %76 : tensor<2x128xf32>
+      %78 = tt.addptr %50, %59 : tensor<2x128x!tt.ptr<bf16, 1>>, tensor<2x128xi32>
+      %79 = arith.truncf %77 : tensor<2x128xf32> to tensor<2x128xbf16>
+      tt.store %78, %79, %57 {cache = 1 : i32, evict = 1 : i32} : tensor<2x128xbf16>
+    }
+    tt.return
+  }
+}

.triton/dump/791dcf81763c6dee467e1d2c436fd6cf/triton_.llir

@@ -0,0 +1,745 @@
+; ModuleID = 'LLVMDialectModule'
+source_filename = "LLVMDialectModule"
+
+@global_smem = external addrspace(3) global [0 x i8]
+@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1
+
+define void @triton__0d1d2d3d4de5(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, i64 %4, i64 %5) local_unnamed_addr !dbg !7 {
+  %7 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !10
+  %8 = lshr i32 %7, 5, !dbg !10
+  %urem = and i32 %7, 255, !dbg !10
+  %9 = or i32 %urem, 256, !dbg !10
+  %10 = or i32 %urem, 512, !dbg !10
+  %11 = or i32 %urem, 768, !dbg !10
+  %12 = or i32 %urem, 1024, !dbg !10
+  %13 = or i32 %urem, 1280, !dbg !10
+  %14 = or i32 %urem, 1536, !dbg !10
+  %15 = or i32 %urem, 1792, !dbg !10
+  %16 = tail call i32 asm "mov.u32 $0, %ctaid.x;", "=r"() #5, !dbg !11
+  %17 = sext i32 %16 to i64, !dbg !12
+  %18 = insertelement <8 x i32> poison, i32 %urem, i64 0
+  %19 = insertelement <8 x i32> %18, i32 %9, i64 1
+  %20 = insertelement <8 x i32> %19, i32 %10, i64 2
+  %21 = insertelement <8 x i32> %20, i32 %11, i64 3
+  %22 = insertelement <8 x i32> %21, i32 %12, i64 4
+  %23 = insertelement <8 x i32> %22, i32 %13, i64 5
+  %24 = insertelement <8 x i32> %23, i32 %14, i64 6
+  %25 = insertelement <8 x i32> %24, i32 %15, i64 7
+  %26 = zext <8 x i32> %25 to <8 x i64>
+  %27 = mul nsw i64 %17, 50257, !dbg !13
+  %invariant.gep = getelementptr i16, ptr addrspace(1) %0, i64 %27, !dbg !14
+  br label %28, !dbg !14
+
+28:                                               ; preds = %6, %28
+  %29 = phi i32 [ 0, %6 ], [ %81, %28 ]
+  %30 = phi <8 x float> [ <float 0xFFF0000000000000, float 0xFFF0000000000000, float 0xFFF0000000000000, float 0xFFF0000000000000, float 0xFFF0000000000000, float 0xFFF0000000000000, float 0xFFF0000000000000, float 0xFFF0000000000000>, %6 ], [ %80, %28 ]
+  %31 = zext nneg i32 %29 to i64, !dbg !15
+  %32 = fcmp ord <8 x float> %30, zeroinitializer, !dbg !16
+  %33 = insertelement <8 x i64> poison, i64 %31, i64 0, !dbg !15
+  %34 = shufflevector <8 x i64> %33, <8 x i64> poison, <8 x i32> zeroinitializer, !dbg !15
+  %35 = or <8 x i64> %34, %26, !dbg !15
+  %36 = icmp ult <8 x i64> %35, <i64 50257, i64 50257, i64 50257, i64 50257, i64 50257, i64 50257, i64 50257, i64 50257>, !dbg !20
+  %37 = extractelement <8 x i64> %35, i64 0, !dbg !21
+  %gep = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %37, !dbg !21
+  %38 = extractelement <8 x i64> %35, i64 1, !dbg !21
+  %gep21 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %38, !dbg !21
+  %39 = extractelement <8 x i64> %35, i64 2, !dbg !21
+  %gep23 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %39, !dbg !21
+  %40 = extractelement <8 x i64> %35, i64 3, !dbg !21
+  %gep25 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %40, !dbg !21
+  %41 = extractelement <8 x i64> %35, i64 4, !dbg !21
+  %gep27 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %41, !dbg !21
+  %42 = extractelement <8 x i64> %35, i64 5, !dbg !21
+  %gep29 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %42, !dbg !21
+  %43 = extractelement <8 x i64> %35, i64 6, !dbg !21
+  %gep31 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %43, !dbg !21
+  %44 = extractelement <8 x i64> %35, i64 7, !dbg !21
+  %gep33 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %44, !dbg !21
+  %45 = extractelement <8 x i1> %36, i64 0, !dbg !22
+  %46 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep, i1 %45, i16 0, i1 %45) #5, !dbg !22
+  %47 = extractelement <8 x i1> %36, i64 1, !dbg !22
+  %48 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep21, i1 %47, i16 0, i1 %47) #5, !dbg !22
+  %49 = extractelement <8 x i1> %36, i64 2, !dbg !22
+  %50 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep23, i1 %49, i16 0, i1 %49) #5, !dbg !22
+  %51 = extractelement <8 x i1> %36, i64 3, !dbg !22
+  %52 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep25, i1 %51, i16 0, i1 %51) #5, !dbg !22
+  %53 = extractelement <8 x i1> %36, i64 4, !dbg !22
+  %54 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep27, i1 %53, i16 0, i1 %53) #5, !dbg !22
+  %55 = extractelement <8 x i1> %36, i64 5, !dbg !22
+  %56 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep29, i1 %55, i16 0, i1 %55) #5, !dbg !22
+  %57 = extractelement <8 x i1> %36, i64 6, !dbg !22
+  %58 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep31, i1 %57, i16 0, i1 %57) #5, !dbg !22
+  %59 = extractelement <8 x i1> %36, i64 7, !dbg !22
+  %60 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep33, i1 %59, i16 0, i1 %59) #5, !dbg !22
+  %61 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %46) #5, !dbg !23
+  %62 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %48) #5, !dbg !23
+  %63 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %50) #5, !dbg !23
+  %64 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %52) #5, !dbg !23
+  %65 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %54) #5, !dbg !23
+  %66 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %56) #5, !dbg !23
+  %67 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %58) #5, !dbg !23
+  %68 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %60) #5, !dbg !23
+  %69 = insertelement <8 x float> poison, float %61, i64 0, !dbg !24
+  %70 = insertelement <8 x float> %69, float %62, i64 1, !dbg !24
+  %71 = insertelement <8 x float> %70, float %63, i64 2, !dbg !24
+  %72 = insertelement <8 x float> %71, float %64, i64 3, !dbg !24
+  %73 = insertelement <8 x float> %72, float %65, i64 4, !dbg !24
+  %74 = insertelement <8 x float> %73, float %66, i64 5, !dbg !24
+  %75 = insertelement <8 x float> %74, float %67, i64 6, !dbg !24
+  %76 = insertelement <8 x float> %75, float %68, i64 7, !dbg !24
+  %77 = fcmp ule <8 x float> %30, %76, !dbg !24
+  %78 = and <8 x i1> %32, %77, !dbg !25
+  %79 = and <8 x i1> %36, %78, !dbg !26
+  %80 = select <8 x i1> %79, <8 x float> %76, <8 x float> %30, !dbg !26
+  %81 = add nuw nsw i32 %29, 2048, !dbg !14
+  %82 = icmp ult i32 %29, 48209, !dbg !14
+  br i1 %82, label %28, label %83, !dbg !14
+
+83:                                               ; preds = %28
+  %84 = and i32 %7, 31, !dbg !10
+  %85 = and i32 %8, 7, !dbg !10
+  %86 = extractelement <8 x float> %80, i64 0, !dbg !27
+  %87 = extractelement <8 x float> %80, i64 1, !dbg !27
+  %88 = fcmp ogt float %86, %87, !dbg !27
+  %89 = fcmp uno float %86, 0.000000e+00, !dbg !31
+  %90 = or i1 %88, %89, !dbg !32
+  %91 = select i1 %90, float %86, float %87, !dbg !33
+  %92 = extractelement <8 x float> %80, i64 2, !dbg !27
+  %93 = fcmp ogt float %91, %92, !dbg !27
+  %94 = fcmp uno float %91, 0.000000e+00, !dbg !31
+  %95 = or i1 %93, %94, !dbg !32
+  %96 = select i1 %95, float %91, float %92, !dbg !33
+  %97 = extractelement <8 x float> %80, i64 3, !dbg !27
+  %98 = fcmp ogt float %96, %97, !dbg !27
+  %99 = fcmp uno float %96, 0.000000e+00, !dbg !31
+  %100 = or i1 %98, %99, !dbg !32
+  %101 = select i1 %100, float %96, float %97, !dbg !33
+  %102 = extractelement <8 x float> %80, i64 4, !dbg !27
+  %103 = fcmp ogt float %101, %102, !dbg !27
+  %104 = fcmp uno float %101, 0.000000e+00, !dbg !31
+  %105 = or i1 %103, %104, !dbg !32
+  %106 = select i1 %105, float %101, float %102, !dbg !33
+  %107 = extractelement <8 x float> %80, i64 5, !dbg !27
+  %108 = fcmp ogt float %106, %107, !dbg !27
+  %109 = fcmp uno float %106, 0.000000e+00, !dbg !31
+  %110 = or i1 %108, %109, !dbg !32
+  %111 = select i1 %110, float %106, float %107, !dbg !33
+  %112 = extractelement <8 x float> %80, i64 6, !dbg !27
+  %113 = fcmp ogt float %111, %112, !dbg !27
+  %114 = fcmp uno float %111, 0.000000e+00, !dbg !31
+  %115 = or i1 %113, %114, !dbg !32
+  %116 = select i1 %115, float %111, float %112, !dbg !33
+  %117 = extractelement <8 x float> %80, i64 7, !dbg !27
+  %118 = fcmp ogt float %116, %117, !dbg !27
+  %119 = fcmp uno float %116, 0.000000e+00, !dbg !31
+  %120 = or i1 %118, %119, !dbg !32
+  %121 = select i1 %120, float %116, float %117, !dbg !33
+  %122 = bitcast float %121 to i32, !dbg !34
+  %123 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %122, i32 16, i32 31), !dbg !34
+  %124 = bitcast i32 %123 to float, !dbg !34
+  %125 = fcmp ogt float %121, %124, !dbg !27
+  %126 = fcmp uno float %121, 0.000000e+00, !dbg !31
+  %127 = or i1 %126, %125, !dbg !32
+  %128 = select i1 %127, float %121, float %124, !dbg !33
+  %129 = bitcast float %128 to i32, !dbg !34
+  %130 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %129, i32 8, i32 31), !dbg !34
+  %131 = bitcast i32 %130 to float, !dbg !34
+  %132 = fcmp ogt float %128, %131, !dbg !27
+  %133 = fcmp uno float %128, 0.000000e+00, !dbg !31
+  %134 = or i1 %132, %133, !dbg !32
+  %135 = select i1 %134, float %128, float %131, !dbg !33
+  %136 = bitcast float %135 to i32, !dbg !34
+  %137 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %136, i32 4, i32 31), !dbg !34
+  %138 = bitcast i32 %137 to float, !dbg !34
+  %139 = fcmp ogt float %135, %138, !dbg !27
+  %140 = fcmp uno float %135, 0.000000e+00, !dbg !31
+  %141 = or i1 %139, %140, !dbg !32
+  %142 = select i1 %141, float %135, float %138, !dbg !33
+  %143 = bitcast float %142 to i32, !dbg !34
+  %144 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %143, i32 2, i32 31), !dbg !34
+  %145 = bitcast i32 %144 to float, !dbg !34
+  %146 = fcmp ogt float %142, %145, !dbg !27
+  %147 = fcmp uno float %142, 0.000000e+00, !dbg !31
+  %148 = or i1 %146, %147, !dbg !32
+  %149 = select i1 %148, float %142, float %145, !dbg !33
+  %150 = bitcast float %149 to i32, !dbg !34
+  %151 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %150, i32 1, i32 31), !dbg !34
+  %152 = bitcast i32 %151 to float, !dbg !34
+  %153 = fcmp ogt float %149, %152, !dbg !27
+  %154 = fcmp uno float %149, 0.000000e+00, !dbg !31
+  %155 = or i1 %153, %154, !dbg !32
+  %156 = select i1 %155, float %149, float %152, !dbg !33
+  %157 = icmp eq i32 %84, 0, !dbg !34
+  %158 = zext nneg i32 %85 to i64, !dbg !34
+  %159 = getelementptr float, ptr addrspace(3) @global_smem, i64 %158, !dbg !34
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %159, float %156, i1 %157) #5, !dbg !34
+  tail call void @llvm.nvvm.barrier0(), !dbg !34
+  %160 = icmp slt i32 %7, 8, !dbg !34
+  %161 = sext i32 %7 to i64, !dbg !34
+  %162 = getelementptr float, ptr addrspace(3) @global_smem, i64 %161, !dbg !34
+  %163 = tail call float asm sideeffect "@$2 ld.shared.b32 $0, [ $1 + 0 ];", "=r,r,b"(ptr addrspace(3) %162, i1 %160) #5, !dbg !34
+  %164 = bitcast float %163 to i32, !dbg !34
+  %165 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %164, i32 4, i32 31), !dbg !34
+  %166 = bitcast i32 %165 to float, !dbg !34
+  %167 = fcmp ogt float %163, %166, !dbg !27
+  %168 = fcmp uno float %163, 0.000000e+00, !dbg !31
+  %169 = or i1 %168, %167, !dbg !32
+  %170 = select i1 %169, float %163, float %166, !dbg !33
+  %171 = bitcast float %170 to i32, !dbg !34
+  %172 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %171, i32 2, i32 31), !dbg !34
+  %173 = bitcast i32 %172 to float, !dbg !34
+  %174 = fcmp ogt float %170, %173, !dbg !27
+  %175 = fcmp uno float %170, 0.000000e+00, !dbg !31
+  %176 = or i1 %174, %175, !dbg !32
+  %177 = select i1 %176, float %170, float %173, !dbg !33
+  %178 = bitcast float %177 to i32, !dbg !34
+  %179 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %178, i32 1, i32 31), !dbg !34
+  %180 = bitcast i32 %179 to float, !dbg !34
+  %181 = fcmp ogt float %177, %180, !dbg !27
+  %182 = fcmp uno float %177, 0.000000e+00, !dbg !31
+  %183 = or i1 %181, %182, !dbg !32
+  %184 = select i1 %183, float %177, float %180, !dbg !33
+  %185 = and i32 %7, 7, !dbg !34
+  %186 = icmp eq i32 %185, 0, !dbg !34
+  %187 = and i1 %160, %186, !dbg !34
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %162, float %184, i1 %187) #5, !dbg !34
+  tail call void @llvm.nvvm.barrier0(), !dbg !34
+  %188 = load float, ptr addrspace(3) @global_smem, align 4, !dbg !34
+  tail call void @llvm.nvvm.barrier0(), !dbg !36
+  %189 = insertelement <1 x float> undef, float %188, i64 0, !dbg !36
+  store <1 x float> %189, ptr addrspace(3) @global_smem, align 4, !dbg !36
+  tail call void @llvm.nvvm.barrier0(), !dbg !36
+  %190 = load i32, ptr addrspace(3) @global_smem, align 4, !dbg !36
+  %191 = getelementptr float, ptr addrspace(1) %1, i64 %17, !dbg !37
+  %192 = icmp eq i32 %urem, 0, !dbg !38
+  tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %190, ptr addrspace(1) %191, i1 %192) #5, !dbg !38
+  br label %193, !dbg !39
+
+193:                                              ; preds = %83, %193
+  %194 = phi i32 [ 0, %83 ], [ %267, %193 ]
+  %195 = phi <8 x float> [ zeroinitializer, %83 ], [ %266, %193 ]
+  %196 = zext nneg i32 %194 to i64, !dbg !40
+  %197 = insertelement <8 x i64> poison, i64 %196, i64 0, !dbg !40
+  %198 = shufflevector <8 x i64> %197, <8 x i64> poison, <8 x i32> zeroinitializer, !dbg !40
+  %199 = or <8 x i64> %198, %26, !dbg !40
+  %200 = icmp ult <8 x i64> %199, <i64 50257, i64 50257, i64 50257, i64 50257, i64 50257, i64 50257, i64 50257, i64 50257>, !dbg !41
+  %201 = extractelement <8 x i64> %199, i64 0, !dbg !42
+  %gep35 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %201, !dbg !42
+  %202 = extractelement <8 x i64> %199, i64 1, !dbg !42
+  %gep37 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %202, !dbg !42
+  %203 = extractelement <8 x i64> %199, i64 2, !dbg !42
+  %gep39 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %203, !dbg !42
+  %204 = extractelement <8 x i64> %199, i64 3, !dbg !42
+  %gep41 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %204, !dbg !42
+  %205 = extractelement <8 x i64> %199, i64 4, !dbg !42
+  %gep43 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %205, !dbg !42
+  %206 = extractelement <8 x i64> %199, i64 5, !dbg !42
+  %gep45 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %206, !dbg !42
+  %207 = extractelement <8 x i64> %199, i64 6, !dbg !42
+  %gep47 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %207, !dbg !42
+  %208 = extractelement <8 x i64> %199, i64 7, !dbg !42
+  %gep49 = getelementptr i16, ptr addrspace(1) %invariant.gep, i64 %208, !dbg !42
+  %209 = extractelement <8 x i1> %200, i64 0, !dbg !43
+  %210 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep35, i1 %209, i16 0, i1 %209) #5, !dbg !43
+  %211 = extractelement <8 x i1> %200, i64 1, !dbg !43
+  %212 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep37, i1 %211, i16 0, i1 %211) #5, !dbg !43
+  %213 = extractelement <8 x i1> %200, i64 2, !dbg !43
+  %214 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep39, i1 %213, i16 0, i1 %213) #5, !dbg !43
+  %215 = extractelement <8 x i1> %200, i64 3, !dbg !43
+  %216 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep41, i1 %215, i16 0, i1 %215) #5, !dbg !43
+  %217 = extractelement <8 x i1> %200, i64 4, !dbg !43
+  %218 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep43, i1 %217, i16 0, i1 %217) #5, !dbg !43
+  %219 = extractelement <8 x i1> %200, i64 5, !dbg !43
+  %220 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep45, i1 %219, i16 0, i1 %219) #5, !dbg !43
+  %221 = extractelement <8 x i1> %200, i64 6, !dbg !43
+  %222 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep47, i1 %221, i16 0, i1 %221) #5, !dbg !43
+  %223 = extractelement <8 x i1> %200, i64 7, !dbg !43
+  %224 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_last.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %gep49, i1 %223, i16 0, i1 %223) #5, !dbg !43
+  %225 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %210) #5, !dbg !44
+  %226 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %212) #5, !dbg !44
+  %227 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %214) #5, !dbg !44
+  %228 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %216) #5, !dbg !44
+  %229 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %218) #5, !dbg !44
+  %230 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %220) #5, !dbg !44
+  %231 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %222) #5, !dbg !44
+  %232 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %224) #5, !dbg !44
+  %233 = fsub float %225, %188, !dbg !45
+  %234 = fsub float %226, %188, !dbg !45
+  %235 = fsub float %227, %188, !dbg !45
+  %236 = fsub float %228, %188, !dbg !45
+  %237 = fsub float %229, %188, !dbg !45
+  %238 = fsub float %230, %188, !dbg !45
+  %239 = fsub float %231, %188, !dbg !45
+  %240 = fsub float %232, %188, !dbg !45
+  %241 = fmul float %233, 0x3FF7154760000000, !dbg !46
+  %242 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %241) #5, !dbg !46
+  %243 = fmul float %234, 0x3FF7154760000000, !dbg !46
+  %244 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %243) #5, !dbg !46
+  %245 = fmul float %235, 0x3FF7154760000000, !dbg !46
+  %246 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %245) #5, !dbg !46
+  %247 = fmul float %236, 0x3FF7154760000000, !dbg !46
+  %248 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %247) #5, !dbg !46
+  %249 = fmul float %237, 0x3FF7154760000000, !dbg !46
+  %250 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %249) #5, !dbg !46
+  %251 = fmul float %238, 0x3FF7154760000000, !dbg !46
+  %252 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %251) #5, !dbg !46
+  %253 = fmul float %239, 0x3FF7154760000000, !dbg !46
+  %254 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %253) #5, !dbg !46
+  %255 = fmul float %240, 0x3FF7154760000000, !dbg !46
+  %256 = tail call float asm "ex2.approx.f32 $0, $1;", "=f,f"(float %255) #5, !dbg !46
+  %257 = insertelement <8 x float> poison, float %242, i64 0, !dbg !47
+  %258 = insertelement <8 x float> %257, float %244, i64 1, !dbg !47
+  %259 = insertelement <8 x float> %258, float %246, i64 2, !dbg !47
+  %260 = insertelement <8 x float> %259, float %248, i64 3, !dbg !47
+  %261 = insertelement <8 x float> %260, float %250, i64 4, !dbg !47
+  %262 = insertelement <8 x float> %261, float %252, i64 5, !dbg !47
+  %263 = insertelement <8 x float> %262, float %254, i64 6, !dbg !47
+  %264 = insertelement <8 x float> %263, float %256, i64 7, !dbg !47
+  %265 = select <8 x i1> %200, <8 x float> %264, <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, !dbg !47
+  %266 = fadd <8 x float> %195, %265, !dbg !47
+  %267 = add nuw nsw i32 %194, 2048, !dbg !39
+  %268 = icmp ult i32 %194, 48209, !dbg !39
+  br i1 %268, label %193, label %269, !dbg !39
+
+269:                                              ; preds = %193
+  tail call void @llvm.nvvm.barrier0(), !dbg !48
+  %shift = shufflevector <8 x float> %266, <8 x float> poison, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !52
+  %270 = fadd <8 x float> %266, %shift, !dbg !52
+  %shift95 = shufflevector <8 x float> %266, <8 x float> poison, <8 x i32> <i32 2, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !52
+  %271 = fadd <8 x float> %shift95, %270, !dbg !52
+  %shift96 = shufflevector <8 x float> %266, <8 x float> poison, <8 x i32> <i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !52
+  %272 = fadd <8 x float> %shift96, %271, !dbg !52
+  %shift97 = shufflevector <8 x float> %266, <8 x float> poison, <8 x i32> <i32 4, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !52
+  %273 = fadd <8 x float> %shift97, %272, !dbg !52
+  %shift98 = shufflevector <8 x float> %266, <8 x float> poison, <8 x i32> <i32 5, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !52
+  %274 = fadd <8 x float> %shift98, %273, !dbg !52
+  %shift99 = shufflevector <8 x float> %266, <8 x float> poison, <8 x i32> <i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !52
+  %275 = fadd <8 x float> %shift99, %274, !dbg !52
+  %shift100 = shufflevector <8 x float> %266, <8 x float> poison, <8 x i32> <i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !52
+  %276 = fadd <8 x float> %shift100, %275, !dbg !52
+  %277 = extractelement <8 x float> %276, i64 0, !dbg !52
+  %278 = bitcast float %277 to i32, !dbg !48
+  %279 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %278, i32 16, i32 31), !dbg !48
+  %280 = bitcast i32 %279 to float, !dbg !48
+  %281 = fadd float %277, %280, !dbg !52
+  %282 = bitcast float %281 to i32, !dbg !48
+  %283 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %282, i32 8, i32 31), !dbg !48
+  %284 = bitcast i32 %283 to float, !dbg !48
+  %285 = fadd float %281, %284, !dbg !52
+  %286 = bitcast float %285 to i32, !dbg !48
+  %287 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %286, i32 4, i32 31), !dbg !48
+  %288 = bitcast i32 %287 to float, !dbg !48
+  %289 = fadd float %285, %288, !dbg !52
+  %290 = bitcast float %289 to i32, !dbg !48
+  %291 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %290, i32 2, i32 31), !dbg !48
+  %292 = bitcast i32 %291 to float, !dbg !48
+  %293 = fadd float %289, %292, !dbg !52
+  %294 = bitcast float %293 to i32, !dbg !48
+  %295 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %294, i32 1, i32 31), !dbg !48
+  %296 = bitcast i32 %295 to float, !dbg !48
+  %297 = fadd float %293, %296, !dbg !52
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %159, float %297, i1 %157) #5, !dbg !48
+  tail call void @llvm.nvvm.barrier0(), !dbg !48
+  %298 = tail call float asm sideeffect "@$2 ld.shared.b32 $0, [ $1 + 0 ];", "=r,r,b"(ptr addrspace(3) %162, i1 %160) #5, !dbg !48
+  %299 = bitcast float %298 to i32, !dbg !48
+  %300 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %299, i32 4, i32 31), !dbg !48
+  %301 = bitcast i32 %300 to float, !dbg !48
+  %302 = fadd float %298, %301, !dbg !52
+  %303 = bitcast float %302 to i32, !dbg !48
+  %304 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %303, i32 2, i32 31), !dbg !48
+  %305 = bitcast i32 %304 to float, !dbg !48
+  %306 = fadd float %302, %305, !dbg !52
+  %307 = bitcast float %306 to i32, !dbg !48
+  %308 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %307, i32 1, i32 31), !dbg !48
+  %309 = bitcast i32 %308 to float, !dbg !48
+  %310 = fadd float %306, %309, !dbg !52
+  tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %162, float %310, i1 %187) #5, !dbg !48
+  tail call void @llvm.nvvm.barrier0(), !dbg !48
+  %311 = load float, ptr addrspace(3) @global_smem, align 4, !dbg !48
+  tail call void @llvm.nvvm.barrier0(), !dbg !56
+  %312 = insertelement <1 x float> undef, float %311, i64 0, !dbg !56
+  store <1 x float> %312, ptr addrspace(3) @global_smem, align 4, !dbg !56
+  tail call void @llvm.nvvm.barrier0(), !dbg !56
+  %313 = load i32, ptr addrspace(3) @global_smem, align 4, !dbg !56
+  %314 = getelementptr float, ptr addrspace(1) %2, i64 %17, !dbg !57
+  tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %313, ptr addrspace(1) %314, i1 %192) #5, !dbg !58
+  %315 = fcmp olt float %311, 0x3810000000000000, !dbg !59
+  %316 = fmul float %311, 0x4160000000000000, !dbg !59
+  %.02.i = select i1 %315, float %316, float %311, !dbg !59
+  %i.i.0.i = select i1 %315, float -2.300000e+01, float 0.000000e+00, !dbg !59
+  %317 = bitcast float %.02.i to i32, !dbg !59
+  %318 = add i32 %317, -1059760811, !dbg !59
+  %319 = and i32 %318, -8388608, !dbg !59
+  %320 = sub i32 %317, %319, !dbg !59
+  %321 = bitcast i32 %320 to float, !dbg !59
+  %322 = sitofp i32 %319 to float, !dbg !59
+  %323 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not.i = icmp eq i32 %323, 0, !dbg !59
+  %324 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %322, float 0x3E80000000000000, float %i.i.0.i) #5, !dbg !59
+  %325 = tail call float @llvm.nvvm.fma.rn.f(float %322, float 0x3E80000000000000, float %i.i.0.i) #5, !dbg !59
+  %.08.i = select i1 %.not.i, float %325, float %324, !dbg !59
+  %326 = fadd float %321, -1.000000e+00, !dbg !59
+  %327 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not1.i = icmp eq i32 %327, 0, !dbg !59
+  %328 = tail call float @llvm.nvvm.fma.rn.ftz.f(float 0xBFC0AA04E0000000, float %326, float 0x3FC2073EC0000000) #5, !dbg !59
+  %329 = tail call float @llvm.nvvm.fma.rn.f(float 0xBFC0AA04E0000000, float %326, float 0x3FC2073EC0000000) #5, !dbg !59
+  %.010.i = select i1 %.not1.i, float %329, float %328, !dbg !59
+  %330 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not2.i = icmp eq i32 %330, 0, !dbg !59
+  %331 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.010.i, float %326, float 0xBFBF19B980000000) #5, !dbg !59
+  %332 = tail call float @llvm.nvvm.fma.rn.f(float %.010.i, float %326, float 0xBFBF19B980000000) #5, !dbg !59
+  %.011.i = select i1 %.not2.i, float %332, float %331, !dbg !59
+  %333 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not3.i = icmp eq i32 %333, 0, !dbg !59
+  %334 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.011.i, float %326, float 0x3FC1E52AA0000000) #5, !dbg !59
+  %335 = tail call float @llvm.nvvm.fma.rn.f(float %.011.i, float %326, float 0x3FC1E52AA0000000) #5, !dbg !59
+  %.012.i = select i1 %.not3.i, float %335, float %334, !dbg !59
+  %336 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not4.i = icmp eq i32 %336, 0, !dbg !59
+  %337 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.012.i, float %326, float 0xBFC55B1720000000) #5, !dbg !59
+  %338 = tail call float @llvm.nvvm.fma.rn.f(float %.012.i, float %326, float 0xBFC55B1720000000) #5, !dbg !59
+  %.09.i = select i1 %.not4.i, float %338, float %337, !dbg !59
+  %339 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not5.i = icmp eq i32 %339, 0, !dbg !59
+  %340 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.09.i, float %326, float 0x3FC99DA160000000) #5, !dbg !59
+  %341 = tail call float @llvm.nvvm.fma.rn.f(float %.09.i, float %326, float 0x3FC99DA160000000) #5, !dbg !59
+  %.05.i = select i1 %.not5.i, float %341, float %340, !dbg !59
+  %342 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not6.i = icmp eq i32 %342, 0, !dbg !59
+  %343 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.05.i, float %326, float 0xBFCFFFE440000000) #5, !dbg !59
+  %344 = tail call float @llvm.nvvm.fma.rn.f(float %.05.i, float %326, float 0xBFCFFFE440000000) #5, !dbg !59
+  %.01.i = select i1 %.not6.i, float %344, float %343, !dbg !59
+  %345 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not7.i = icmp eq i32 %345, 0, !dbg !59
+  %346 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.01.i, float %326, float 0x3FD5554F00000000) #5, !dbg !59
+  %347 = tail call float @llvm.nvvm.fma.rn.f(float %.01.i, float %326, float 0x3FD5554F00000000) #5, !dbg !59
+  %.0.i = select i1 %.not7.i, float %347, float %346, !dbg !59
+  %348 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not8.i = icmp eq i32 %348, 0, !dbg !59
+  %349 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.0.i, float %326, float -5.000000e-01) #5, !dbg !59
+  %350 = tail call float @llvm.nvvm.fma.rn.f(float %.0.i, float %326, float -5.000000e-01) #5, !dbg !59
+  %.07.i = select i1 %.not8.i, float %350, float %349, !dbg !59
+  %351 = fmul float %326, %.07.i, !dbg !59
+  %352 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not9.i = icmp eq i32 %352, 0, !dbg !59
+  %353 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %351, float %326, float %326) #5, !dbg !59
+  %354 = tail call float @llvm.nvvm.fma.rn.f(float %351, float %326, float %326) #5, !dbg !59
+  %.06.i = select i1 %.not9.i, float %354, float %353, !dbg !59
+  %355 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not10.i = icmp eq i32 %355, 0, !dbg !59
+  %356 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.08.i, float 0x3FE62E4300000000, float %.06.i) #5, !dbg !59
+  %357 = tail call float @llvm.nvvm.fma.rn.f(float %.08.i, float 0x3FE62E4300000000, float %.06.i) #5, !dbg !59
+  %.04.i = select i1 %.not10.i, float %357, float %356, !dbg !59
+  %358 = icmp ugt i32 %317, 2139095039, !dbg !59
+  br i1 %358, label %__nv_fmaf_rn.exit.i.i, label %__nv_logf.exit, !dbg !59
+
+__nv_fmaf_rn.exit.i.i:                            ; preds = %269
+  %359 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5, !dbg !59
+  %.not11.i = icmp eq i32 %359, 0, !dbg !59
+  %360 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.02.i, float 0x7FF0000000000000, float 0x7FF0000000000000) #5, !dbg !59
+  %361 = tail call float @llvm.nvvm.fma.rn.f(float %.02.i, float 0x7FF0000000000000, float 0x7FF0000000000000) #5, !dbg !59
+  %.03.i = select i1 %.not11.i, float %361, float %360, !dbg !59
+  br label %__nv_logf.exit, !dbg !59
+
+__nv_logf.exit:                                   ; preds = %269, %__nv_fmaf_rn.exit.i.i
+  %r.i.0.i = phi float [ %.03.i, %__nv_fmaf_rn.exit.i.i ], [ %.04.i, %269 ], !dbg !59
+  %362 = fcmp oeq float %.02.i, 0.000000e+00, !dbg !59
+  %r.i.1.i = select i1 %362, float 0xFFF0000000000000, float %r.i.0.i, !dbg !59
+  %363 = extractelement <8 x i64> %26, i64 0, !dbg !60
+  %364 = extractelement <8 x i64> %26, i64 1, !dbg !60
+  %365 = extractelement <8 x i64> %26, i64 2, !dbg !60
+  %366 = extractelement <8 x i64> %26, i64 3, !dbg !60
+  %367 = extractelement <8 x i64> %26, i64 4, !dbg !60
+  %368 = extractelement <8 x i64> %26, i64 5, !dbg !60
+  %369 = extractelement <8 x i64> %26, i64 6, !dbg !60
+  %370 = extractelement <8 x i64> %26, i64 7, !dbg !60
+  br label %371, !dbg !61
+
+371:                                              ; preds = %__nv_logf.exit, %371
+  %372 = phi i32 [ 0, %__nv_logf.exit ], [ %454, %371 ]
+  %373 = zext nneg i32 %372 to i64, !dbg !60
+  %374 = or i64 %363, %373, !dbg !60
+  %375 = or i64 %364, %373, !dbg !60
+  %376 = or i64 %365, %373, !dbg !60
+  %377 = or i64 %366, %373, !dbg !60
+  %378 = or i64 %367, %373, !dbg !60
+  %379 = or i64 %368, %373, !dbg !60
+  %380 = or i64 %369, %373, !dbg !60
+  %381 = or i64 %370, %373, !dbg !60
+  %382 = icmp ult i64 %374, 50257, !dbg !62
+  %383 = icmp ult i64 %375, 50257, !dbg !62
+  %384 = icmp ult i64 %376, 50257, !dbg !62
+  %385 = icmp ult i64 %377, 50257, !dbg !62
+  %386 = icmp ult i64 %378, 50257, !dbg !62
+  %387 = icmp ult i64 %379, 50257, !dbg !62
+  %388 = icmp ult i64 %380, 50257, !dbg !62
+  %389 = icmp ult i64 %381, 50257, !dbg !62
+  %390 = add nsw i64 %374, %27, !dbg !63
+  %391 = add nsw i64 %375, %27, !dbg !63
+  %392 = add nsw i64 %376, %27, !dbg !63
+  %393 = add nsw i64 %377, %27, !dbg !63
+  %394 = add nsw i64 %378, %27, !dbg !63
+  %395 = add nsw i64 %379, %27, !dbg !63
+  %396 = add nsw i64 %380, %27, !dbg !63
+  %397 = add nsw i64 %381, %27, !dbg !63
+  %398 = getelementptr i16, ptr addrspace(1) %0, i64 %390, !dbg !64
+  %399 = getelementptr i16, ptr addrspace(1) %0, i64 %391, !dbg !64
+  %400 = getelementptr i16, ptr addrspace(1) %0, i64 %392, !dbg !64
+  %401 = getelementptr i16, ptr addrspace(1) %0, i64 %393, !dbg !64
+  %402 = getelementptr i16, ptr addrspace(1) %0, i64 %394, !dbg !64
+  %403 = getelementptr i16, ptr addrspace(1) %0, i64 %395, !dbg !64
+  %404 = getelementptr i16, ptr addrspace(1) %0, i64 %396, !dbg !64
+  %405 = getelementptr i16, ptr addrspace(1) %0, i64 %397, !dbg !64
+  %406 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %398, i1 %382, i16 0, i1 %382) #5, !dbg !65
+  %407 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %399, i1 %383, i16 0, i1 %383) #5, !dbg !65
+  %408 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %400, i1 %384, i16 0, i1 %384) #5, !dbg !65
+  %409 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %401, i1 %385, i16 0, i1 %385) #5, !dbg !65
+  %410 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %402, i1 %386, i16 0, i1 %386) #5, !dbg !65
+  %411 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %403, i1 %387, i16 0, i1 %387) #5, !dbg !65
+  %412 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %404, i1 %388, i16 0, i1 %388) #5, !dbg !65
+  %413 = tail call i16 asm sideeffect "mov.u16 $0, 0x0;\0A\09@$2 ld.global.L1::evict_first.b16 { $0 }, [ $1 + 0 ];\0A\09@!$4 mov.u16 $0, $3;", "=c,l,b,c,b"(ptr addrspace(1) %405, i1 %389, i16 0, i1 %389) #5, !dbg !65
+  %414 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %406) #5, !dbg !66
+  %415 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %407) #5, !dbg !66
+  %416 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %408) #5, !dbg !66
+  %417 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %409) #5, !dbg !66
+  %418 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %410) #5, !dbg !66
+  %419 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %411) #5, !dbg !66
+  %420 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %412) #5, !dbg !66
+  %421 = tail call float asm "cvt.f32.bf16 $0, $1;", "=r,h"(i16 %413) #5, !dbg !66
+  %422 = fsub float %414, %188, !dbg !67
+  %423 = fsub float %415, %188, !dbg !67
+  %424 = fsub float %416, %188, !dbg !67
+  %425 = fsub float %417, %188, !dbg !67
+  %426 = fsub float %418, %188, !dbg !67
+  %427 = fsub float %419, %188, !dbg !67
+  %428 = fsub float %420, %188, !dbg !67
+  %429 = fsub float %421, %188, !dbg !67
+  %430 = fsub float %422, %r.i.1.i, !dbg !68
+  %431 = fsub float %423, %r.i.1.i, !dbg !68
+  %432 = fsub float %424, %r.i.1.i, !dbg !68
+  %433 = fsub float %425, %r.i.1.i, !dbg !68
+  %434 = fsub float %426, %r.i.1.i, !dbg !68
+  %435 = fsub float %427, %r.i.1.i, !dbg !68
+  %436 = fsub float %428, %r.i.1.i, !dbg !68
+  %437 = fsub float %429, %r.i.1.i, !dbg !68
+  %438 = getelementptr i16, ptr addrspace(1) %3, i64 %390, !dbg !69
+  %439 = getelementptr i16, ptr addrspace(1) %3, i64 %391, !dbg !69
+  %440 = getelementptr i16, ptr addrspace(1) %3, i64 %392, !dbg !69
+  %441 = getelementptr i16, ptr addrspace(1) %3, i64 %393, !dbg !69
+  %442 = getelementptr i16, ptr addrspace(1) %3, i64 %394, !dbg !69
+  %443 = getelementptr i16, ptr addrspace(1) %3, i64 %395, !dbg !69
+  %444 = getelementptr i16, ptr addrspace(1) %3, i64 %396, !dbg !69
+  %445 = getelementptr i16, ptr addrspace(1) %3, i64 %397, !dbg !69
+  %446 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %430) #5, !dbg !70
+  %447 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %431) #5, !dbg !70
+  %448 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %432) #5, !dbg !70
+  %449 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %433) #5, !dbg !70
+  %450 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %434) #5, !dbg !70
+  %451 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %435) #5, !dbg !70
+  %452 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %436) #5, !dbg !70
+  %453 = tail call i16 asm "cvt.rn.bf16.f32 $0, $1;", "=h,r"(float %437) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %446, ptr addrspace(1) %438, i1 %382) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %447, ptr addrspace(1) %439, i1 %383) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %448, ptr addrspace(1) %440, i1 %384) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %449, ptr addrspace(1) %441, i1 %385) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %450, ptr addrspace(1) %442, i1 %386) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %451, ptr addrspace(1) %443, i1 %387) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %452, ptr addrspace(1) %444, i1 %388) #5, !dbg !70
+  tail call void asm sideeffect "@$2 st.global.b16 [ $1 + 0 ], { $0 };", "c,l,b"(i16 %453, ptr addrspace(1) %445, i1 %389) #5, !dbg !70
+  %454 = add nuw nsw i32 %372, 2048, !dbg !61
+  %455 = icmp ult i32 %372, 48209, !dbg !61
+  br i1 %455, label %371, label %456, !dbg !61
+
+456:                                              ; preds = %371
+  ret void, !dbg !71
+}
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare noundef i32 @llvm.nvvm.read.ptx.sreg.tid.x() #0
+
+; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite)
+declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #1
+
+; Function Attrs: convergent nocallback nounwind
+declare void @llvm.nvvm.barrier0() #2
+
+; Function Attrs: alwaysinline nounwind
+define float @__nv_logf(float %a) local_unnamed_addr #3 {
+__nv_fmaf_rn.exit10.i:
+  %0 = fcmp olt float %a, 0x3810000000000000
+  %1 = fmul float %a, 0x4160000000000000
+  %.02 = select i1 %0, float %1, float %a
+  %i.i.0 = select i1 %0, float -2.300000e+01, float 0.000000e+00
+  %2 = bitcast float %.02 to i32
+  %3 = add i32 %2, -1059760811
+  %4 = and i32 %3, -8388608
+  %5 = sub i32 %2, %4
+  %6 = bitcast i32 %5 to float
+  %7 = sitofp i32 %4 to float
+  %8 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not = icmp eq i32 %8, 0
+  %9 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %7, float 0x3E80000000000000, float %i.i.0) #5
+  %10 = tail call float @llvm.nvvm.fma.rn.f(float %7, float 0x3E80000000000000, float %i.i.0) #5
+  %.08 = select i1 %.not, float %10, float %9
+  %11 = fadd float %6, -1.000000e+00
+  %12 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not1 = icmp eq i32 %12, 0
+  %13 = tail call float @llvm.nvvm.fma.rn.ftz.f(float 0xBFC0AA04E0000000, float %11, float 0x3FC2073EC0000000) #5
+  %14 = tail call float @llvm.nvvm.fma.rn.f(float 0xBFC0AA04E0000000, float %11, float 0x3FC2073EC0000000) #5
+  %.010 = select i1 %.not1, float %14, float %13
+  %15 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not2 = icmp eq i32 %15, 0
+  %16 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.010, float %11, float 0xBFBF19B980000000) #5
+  %17 = tail call float @llvm.nvvm.fma.rn.f(float %.010, float %11, float 0xBFBF19B980000000) #5
+  %.011 = select i1 %.not2, float %17, float %16
+  %18 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not3 = icmp eq i32 %18, 0
+  %19 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.011, float %11, float 0x3FC1E52AA0000000) #5
+  %20 = tail call float @llvm.nvvm.fma.rn.f(float %.011, float %11, float 0x3FC1E52AA0000000) #5
+  %.012 = select i1 %.not3, float %20, float %19
+  %21 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not4 = icmp eq i32 %21, 0
+  %22 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.012, float %11, float 0xBFC55B1720000000) #5
+  %23 = tail call float @llvm.nvvm.fma.rn.f(float %.012, float %11, float 0xBFC55B1720000000) #5
+  %.09 = select i1 %.not4, float %23, float %22
+  %24 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not5 = icmp eq i32 %24, 0
+  %25 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.09, float %11, float 0x3FC99DA160000000) #5
+  %26 = tail call float @llvm.nvvm.fma.rn.f(float %.09, float %11, float 0x3FC99DA160000000) #5
+  %.05 = select i1 %.not5, float %26, float %25
+  %27 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not6 = icmp eq i32 %27, 0
+  %28 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.05, float %11, float 0xBFCFFFE440000000) #5
+  %29 = tail call float @llvm.nvvm.fma.rn.f(float %.05, float %11, float 0xBFCFFFE440000000) #5
+  %.01 = select i1 %.not6, float %29, float %28
+  %30 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not7 = icmp eq i32 %30, 0
+  %31 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.01, float %11, float 0x3FD5554F00000000) #5
+  %32 = tail call float @llvm.nvvm.fma.rn.f(float %.01, float %11, float 0x3FD5554F00000000) #5
+  %.0 = select i1 %.not7, float %32, float %31
+  %33 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not8 = icmp eq i32 %33, 0
+  %34 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.0, float %11, float -5.000000e-01) #5
+  %35 = tail call float @llvm.nvvm.fma.rn.f(float %.0, float %11, float -5.000000e-01) #5
+  %.07 = select i1 %.not8, float %35, float %34
+  %36 = fmul float %11, %.07
+  %37 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not9 = icmp eq i32 %37, 0
+  %38 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %36, float %11, float %11) #5
+  %39 = tail call float @llvm.nvvm.fma.rn.f(float %36, float %11, float %11) #5
+  %.06 = select i1 %.not9, float %39, float %38
+  %40 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not10 = icmp eq i32 %40, 0
+  %41 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.08, float 0x3FE62E4300000000, float %.06) #5
+  %42 = tail call float @llvm.nvvm.fma.rn.f(float %.08, float 0x3FE62E4300000000, float %.06) #5
+  %.04 = select i1 %.not10, float %42, float %41
+  %43 = icmp ugt i32 %2, 2139095039
+  br i1 %43, label %__nv_fmaf_rn.exit.i, label %__internal_accurate_logf.exit
+
+__nv_fmaf_rn.exit.i:                              ; preds = %__nv_fmaf_rn.exit10.i
+  %44 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #5
+  %.not11 = icmp eq i32 %44, 0
+  %45 = tail call float @llvm.nvvm.fma.rn.ftz.f(float %.02, float 0x7FF0000000000000, float 0x7FF0000000000000) #5
+  %46 = tail call float @llvm.nvvm.fma.rn.f(float %.02, float 0x7FF0000000000000, float 0x7FF0000000000000) #5
+  %.03 = select i1 %.not11, float %46, float %45
+  br label %__internal_accurate_logf.exit
+
+__internal_accurate_logf.exit:                    ; preds = %__nv_fmaf_rn.exit.i, %__nv_fmaf_rn.exit10.i
+  %r.i.0 = phi float [ %.03, %__nv_fmaf_rn.exit.i ], [ %.04, %__nv_fmaf_rn.exit10.i ]
+  %47 = fcmp oeq float %.02, 0.000000e+00
+  %r.i.1 = select i1 %47, float 0xFFF0000000000000, float %r.i.0
+  ret float %r.i.1
+}
+
+declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #4
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.nvvm.fma.rn.ftz.f(float, float, float) #0
+
+; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.nvvm.fma.rn.f(float, float, float) #0
+
+attributes #0 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #1 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) }
+attributes #2 = { convergent nocallback nounwind }
+attributes #3 = { alwaysinline nounwind "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #4 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #5 = { nounwind }
+
+!llvm.module.flags = !{!0, !1}
+!llvm.dbg.cu = !{!2}
+!nvvm.annotations = !{!4, !5, !5, !4}
+!llvm.ident = !{!6}
+
+!0 = !{i32 2, !"Debug Info Version", i32 3}
+!1 = !{i32 4, !"nvvm-reflect-ftz", i32 1}
+!2 = distinct !DICompileUnit(language: DW_LANG_C, file: !3, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug)
+!3 = !DIFile(filename: "cgu6pijtlte2d3yicmpedfh2c7wgxsaexd6ichnxwbwh4deqe6ck.py", directory: "/tmp/torchinductor_root/gu")
+!4 = !{ptr @triton__0d1d2d3d4de5, !"kernel", i32 1}
+!5 = !{ptr @triton__0d1d2d3d4de5, !"maxntidx", i32 256}
+!6 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"}
+!7 = distinct !DISubprogram(name: "triton__0d1d2d3d4de5", linkageName: "triton__0d1d2d3d4de5", scope: !3, file: !3, line: 18, type: !8, scopeLine: 18, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
+!8 = !DISubroutineType(cc: DW_CC_normal, types: !9)
+!9 = !{}
+!10 = !DILocation(line: 24, column: 33, scope: !7)
+!11 = !DILocation(line: 21, column: 28, scope: !7)
+!12 = !DILocation(line: 21, column: 34, scope: !7)
+!13 = !DILocation(line: 31, column: 46, scope: !7)
+!14 = !DILocation(line: 27, column: 36, scope: !7)
+!15 = !DILocation(line: 28, column: 27, scope: !7)
+!16 = !DILocation(line: 38, column: 21, scope: !17, inlinedAt: !19)
+!17 = distinct !DILexicalBlockFile(scope: !7, file: !18, discriminator: 0)
+!18 = !DIFile(filename: "triton_helpers.py", directory: "/usr/local/lib/python3.10/dist-packages/torch/_inductor")
+!19 = !DILocation(line: 34, column: 45, scope: !17)
+!20 = !DILocation(line: 29, column: 25, scope: !7)
+!21 = !DILocation(line: 31, column: 34, scope: !7)
+!22 = !DILocation(line: 31, column: 52, scope: !7)
+!23 = !DILocation(line: 31, column: 103, scope: !7)
+!24 = !DILocation(line: 36, column: 15, scope: !17, inlinedAt: !19)
+!25 = !DILocation(line: 38, column: 16, scope: !17, inlinedAt: !19)
+!26 = !DILocation(line: 0, scope: !7)
+!27 = !DILocation(line: 36, column: 15, scope: !28, inlinedAt: !29)
+!28 = distinct !DILexicalBlockFile(scope: !17, file: !18, discriminator: 0)
+!29 = !DILocation(line: 49, column: 29, scope: !28, inlinedAt: !30)
+!30 = !DILocation(line: 36, column: 38, scope: !28)
+!31 = !DILocation(line: 38, column: 21, scope: !28, inlinedAt: !29)
+!32 = !DILocation(line: 38, column: 16, scope: !28, inlinedAt: !29)
+!33 = !DILocation(line: 39, column: 29, scope: !28, inlinedAt: !29)
+!34 = !DILocation(line: 49, column: 29, scope: !17, inlinedAt: !35)
+!35 = !DILocation(line: 36, column: 38, scope: !17)
+!36 = !DILocation(line: 36, column: 41, scope: !7)
+!37 = !DILocation(line: 37, column: 25, scope: !7)
+!38 = !DILocation(line: 37, column: 36, scope: !7)
+!39 = !DILocation(line: 39, column: 36, scope: !7)
+!40 = !DILocation(line: 40, column: 27, scope: !7)
+!41 = !DILocation(line: 41, column: 25, scope: !7)
+!42 = !DILocation(line: 43, column: 34, scope: !7)
+!43 = !DILocation(line: 43, column: 52, scope: !7)
+!44 = !DILocation(line: 43, column: 103, scope: !7)
+!45 = !DILocation(line: 45, column: 22, scope: !7)
+!46 = !DILocation(line: 46, column: 22, scope: !7)
+!47 = !DILocation(line: 49, column: 40, scope: !7)
+!48 = !DILocation(line: 243, column: 36, scope: !49, inlinedAt: !51)
+!49 = distinct !DILexicalBlockFile(scope: !7, file: !50, discriminator: 0)
+!50 = !DIFile(filename: "standard.py", directory: "/usr/local/lib/python3.10/dist-packages/triton/language")
+!51 = !DILocation(line: 50, column: 27, scope: !49)
+!52 = !DILocation(line: 233, column: 15, scope: !53, inlinedAt: !54)
+!53 = distinct !DILexicalBlockFile(scope: !49, file: !50, discriminator: 0)
+!54 = !DILocation(line: 243, column: 36, scope: !53, inlinedAt: !55)
+!55 = !DILocation(line: 50, column: 27, scope: !53)
+!56 = !DILocation(line: 50, column: 30, scope: !7)
+!57 = !DILocation(line: 51, column: 25, scope: !7)
+!58 = !DILocation(line: 51, column: 37, scope: !7)
+!59 = !DILocation(line: 59, column: 23, scope: !7)
+!60 = !DILocation(line: 53, column: 27, scope: !7)
+!61 = !DILocation(line: 52, column: 36, scope: !7)
+!62 = !DILocation(line: 54, column: 25, scope: !7)
+!63 = !DILocation(line: 56, column: 41, scope: !7)
+!64 = !DILocation(line: 56, column: 35, scope: !7)
+!65 = !DILocation(line: 56, column: 53, scope: !7)
+!66 = !DILocation(line: 56, column: 105, scope: !7)
+!67 = !DILocation(line: 58, column: 24, scope: !7)
+!68 = !DILocation(line: 60, column: 24, scope: !7)
+!69 = !DILocation(line: 62, column: 29, scope: !7)
+!70 = !DILocation(line: 62, column: 54, scope: !7)
+!71 = !DILocation(line: 52, column: 4, scope: !7)

.triton/dump/791dcf81763c6dee467e1d2c436fd6cf/triton_.ttir

@@ -0,0 +1,101 @@
+module {
+  tt.func public @triton__0d1d2d3d4de5(%arg0: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg3: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg4: i64 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}, %arg5: i64) attributes {noinline = false} {
+    %cst = arith.constant dense<0.000000e+00> : tensor<1x2048xbf16>
+    %c50257_i64 = arith.constant 50257 : i64
+    %cst_0 = arith.constant dense<true> : tensor<1x2048xi1>
+    %c50257_i32 = arith.constant 50257 : i32
+    %c2048_i32 = arith.constant 2048 : i32
+    %c0_i32 = arith.constant 0 : i32
+    %cst_1 = arith.constant dense<50257> : tensor<1x2048xi64>
+    %cst_2 = arith.constant dense<0.000000e+00> : tensor<1x2048xf32>
+    %cst_3 = arith.constant dense<0xFF800000> : tensor<1x2048xf32>
+    %0 = tt.get_program_id x : i32
+    %1 = arith.extsi %0 : i32 to i64
+    %2 = tt.make_range {end = 2048 : i32, start = 0 : i32} : tensor<2048xi32>
+    %3 = tt.expand_dims %2 {axis = 0 : i32} : (tensor<2048xi32>) -> tensor<1x2048xi32>
+    %4 = arith.extsi %3 : tensor<1x2048xi32> to tensor<1x2048xi64>
+    %5 = arith.muli %1, %c50257_i64 : i64
+    %6 = tt.splat %5 : (i64) -> tensor<1x2048xi64>
+    %7 = tt.splat %arg0 : (!tt.ptr<bf16, 1>) -> tensor<1x2048x!tt.ptr<bf16, 1>>
+    %8 = scf.for %arg6 = %c0_i32 to %c50257_i32 step %c2048_i32 iter_args(%arg7 = %cst_3) -> (tensor<1x2048xf32>)  : i32 {
+      %29 = arith.extsi %arg6 : i32 to i64
+      %30 = tt.splat %29 : (i64) -> tensor<1x2048xi64>
+      %31 = arith.addi %30, %4 : tensor<1x2048xi64>
+      %32 = arith.cmpi slt, %31, %cst_1 : tensor<1x2048xi64>
+      %33 = arith.addi %31, %6 : tensor<1x2048xi64>
+      %34 = tt.addptr %7, %33 : tensor<1x2048x!tt.ptr<bf16, 1>>, tensor<1x2048xi64>
+      %35 = tt.load %34, %32, %cst {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<1x2048xbf16>
+      %36 = arith.extf %35 : tensor<1x2048xbf16> to tensor<1x2048xf32>
+      %37 = arith.cmpf ogt, %arg7, %36 : tensor<1x2048xf32>
+      %38 = arith.cmpf une, %arg7, %arg7 : tensor<1x2048xf32>
+      %39 = arith.ori %37, %38 : tensor<1x2048xi1>
+      %40 = arith.xori %39, %cst_0 : tensor<1x2048xi1>
+      %41 = arith.andi %32, %40 : tensor<1x2048xi1>
+      %42 = arith.select %41, %36, %arg7 : tensor<1x2048xi1>, tensor<1x2048xf32>
+      scf.yield %42 : tensor<1x2048xf32>
+    }
+    %9 = "tt.reduce"(%8) <{axis = 1 : i32}> ({
+    ^bb0(%arg6: f32, %arg7: f32):
+      %29 = arith.cmpf ogt, %arg6, %arg7 : f32
+      %30 = arith.cmpf une, %arg6, %arg6 : f32
+      %31 = arith.ori %29, %30 : i1
+      %32 = arith.select %31, %arg6, %arg7 : f32
+      tt.reduce.return %32 : f32
+    }) : (tensor<1x2048xf32>) -> tensor<1xf32>
+    %10 = tt.expand_dims %9 {axis = 1 : i32} : (tensor<1xf32>) -> tensor<1x1xf32>
+    %11 = tt.addptr %arg1, %1 : !tt.ptr<f32, 1>, i64
+    %12 = tt.splat %11 : (!tt.ptr<f32, 1>) -> tensor<1x1x!tt.ptr<f32, 1>>
+    tt.store %12, %10 {cache = 1 : i32, evict = 1 : i32} : tensor<1x1xf32>
+    %13 = arith.muli %1, %c50257_i64 : i64
+    %14 = tt.splat %13 : (i64) -> tensor<1x2048xi64>
+    %15 = tt.splat %arg0 : (!tt.ptr<bf16, 1>) -> tensor<1x2048x!tt.ptr<bf16, 1>>
+    %16 = tt.broadcast %10 : (tensor<1x1xf32>) -> tensor<1x2048xf32>
+    %17 = scf.for %arg6 = %c0_i32 to %c50257_i32 step %c2048_i32 iter_args(%arg7 = %cst_2) -> (tensor<1x2048xf32>)  : i32 {
+      %29 = arith.extsi %arg6 : i32 to i64
+      %30 = tt.splat %29 : (i64) -> tensor<1x2048xi64>
+      %31 = arith.addi %30, %4 : tensor<1x2048xi64>
+      %32 = arith.cmpi slt, %31, %cst_1 : tensor<1x2048xi64>
+      %33 = arith.addi %31, %14 : tensor<1x2048xi64>
+      %34 = tt.addptr %15, %33 : tensor<1x2048x!tt.ptr<bf16, 1>>, tensor<1x2048xi64>
+      %35 = tt.load %34, %32, %cst {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<1x2048xbf16>
+      %36 = arith.extf %35 : tensor<1x2048xbf16> to tensor<1x2048xf32>
+      %37 = arith.subf %36, %16 : tensor<1x2048xf32>
+      %38 = math.exp %37 : tensor<1x2048xf32>
+      %39 = arith.addf %arg7, %38 : tensor<1x2048xf32>
+      %40 = arith.select %32, %39, %arg7 : tensor<1x2048xi1>, tensor<1x2048xf32>
+      scf.yield %40 : tensor<1x2048xf32>
+    }
+    %18 = "tt.reduce"(%17) <{axis = 1 : i32}> ({
+    ^bb0(%arg6: f32, %arg7: f32):
+      %29 = arith.addf %arg6, %arg7 : f32
+      tt.reduce.return %29 : f32
+    }) : (tensor<1x2048xf32>) -> tensor<1xf32>
+    %19 = tt.expand_dims %18 {axis = 1 : i32} : (tensor<1xf32>) -> tensor<1x1xf32>
+    %20 = tt.addptr %arg2, %1 : !tt.ptr<f32, 1>, i64
+    %21 = tt.splat %20 : (!tt.ptr<f32, 1>) -> tensor<1x1x!tt.ptr<f32, 1>>
+    tt.store %21, %19 {cache = 1 : i32, evict = 1 : i32} : tensor<1x1xf32>
+    %22 = arith.muli %1, %c50257_i64 : i64
+    %23 = tt.splat %22 : (i64) -> tensor<1x2048xi64>
+    %24 = tt.splat %arg0 : (!tt.ptr<bf16, 1>) -> tensor<1x2048x!tt.ptr<bf16, 1>>
+    %25 = tt.broadcast %10 : (tensor<1x1xf32>) -> tensor<1x2048xf32>
+    %26 = math.log %19 : tensor<1x1xf32>
+    %27 = tt.broadcast %26 : (tensor<1x1xf32>) -> tensor<1x2048xf32>
+    %28 = tt.splat %arg3 : (!tt.ptr<bf16, 1>) -> tensor<1x2048x!tt.ptr<bf16, 1>>
+    scf.for %arg6 = %c0_i32 to %c50257_i32 step %c2048_i32  : i32 {
+      %29 = arith.extsi %arg6 : i32 to i64
+      %30 = tt.splat %29 : (i64) -> tensor<1x2048xi64>
+      %31 = arith.addi %30, %4 : tensor<1x2048xi64>
+      %32 = arith.cmpi slt, %31, %cst_1 : tensor<1x2048xi64>
+      %33 = arith.addi %31, %23 : tensor<1x2048xi64>
+      %34 = tt.addptr %24, %33 : tensor<1x2048x!tt.ptr<bf16, 1>>, tensor<1x2048xi64>
+      %35 = tt.load %34, %32, %cst {cache = 1 : i32, evict = 2 : i32, isVolatile = false} : tensor<1x2048xbf16>
+      %36 = arith.extf %35 : tensor<1x2048xbf16> to tensor<1x2048xf32>
+      %37 = arith.subf %36, %25 : tensor<1x2048xf32>
+      %38 = arith.subf %37, %27 : tensor<1x2048xf32>
+      %39 = tt.addptr %28, %33 : tensor<1x2048x!tt.ptr<bf16, 1>>, tensor<1x2048xi64>
+      %40 = arith.truncf %38 : tensor<1x2048xf32> to tensor<1x2048xbf16>
+      tt.store %39, %40, %32 {cache = 1 : i32, evict = 1 : i32} : tensor<1x2048xbf16>
+    }
+    tt.return
+  }
+}

.triton/dump/7dc5bb3e5c2bb99527fff34c6fba7810/triton_.ptx

@@ -0,0 +1,277 @@
+//
+// Generated by LLVM NVPTX Back-End
+//
+
+.version 8.2
+.target sm_89
+.address_size 64
+
+	// .globl	triton__0d1de
+
+.visible .entry triton__0d1de(
+	.param .u64 triton__0d1de_param_0,
+	.param .u32 triton__0d1de_param_1
+)
+.maxntid 128, 1, 1
+{
+	.reg .pred 	%p<2>;
+	.reg .b32 	%r<6>;
+	.reg .b64 	%rd<5>;
+	.loc	1 18 0
+$L__func_begin0:
+	.loc	1 18 0
+
+	ld.param.u64 	%rd3, [triton__0d1de_param_0];
+$L__tmp0:
+	.loc	1 21 36
+	mov.u32 	%r2, %tid.x;
+	and.b32  	%r3, %r2, 127;
+	.loc	1 20 28
+	mov.u32 %r1, %ctaid.x;
+	.loc	1 20 33
+	shl.b32 	%r4, %r1, 7;
+	.loc	1 21 23
+	or.b32  	%r5, %r4, %r3;
+	.loc	1 22 21
+	setp.lt.s32 	%p1, %r5, 512;
+	.loc	1 25 25
+	cvt.s64.s32 	%rd1, %r5;
+	mul.wide.s32 	%rd4, %r5, 8;
+	add.s64 	%rd2, %rd3, %rd4;
+	.loc	1 25 36
+	@%p1 st.global.b64 [ %rd2 + 0 ], { %rd1 };
+	.loc	1 25 4
+	ret;
+$L__tmp1:
+$L__func_end0:
+
+}
+	.file	1 "/tmp/torchinductor_root/wx/cwxxgxdevnyc453z7hh4nxzgmvlhh6suwokktps3dw62btskgxt4.py"
+	.section	.debug_abbrev
+	{
+.b8 1
+.b8 17
+.b8 1
+.b8 37
+.b8 8
+.b8 19
+.b8 5
+.b8 3
+.b8 8
+.b8 16
+.b8 6
+.b8 27
+.b8 8
+.b8 180
+.b8 66
+.b8 12
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+.b8 0
+.b8 2
+.b8 46
+.b8 0
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 64
+.b8 10
+.b8 135
+.b8 64
+.b8 8
+.b8 3
+.b8 8
+.b8 58
+.b8 11
+.b8 59
+.b8 11
+.b8 63
+.b8 12
+.b8 0
+.b8 0
+.b8 0
+	}
+	.section	.debug_info
+	{
+.b32 172
+.b8 2
+.b8 0
+.b32 .debug_abbrev
+.b8 8
+.b8 1
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 0
+.b8 2
+.b8 0
+.b8 99
+.b8 119
+.b8 120
+.b8 120
+.b8 103
+.b8 120
+.b8 100
+.b8 101
+.b8 118
+.b8 110
+.b8 121
+.b8 99
+.b8 52
+.b8 53
+.b8 51
+.b8 122
+.b8 55
+.b8 104
+.b8 104
+.b8 52
+.b8 110
+.b8 120
+.b8 122
+.b8 103
+.b8 109
+.b8 118
+.b8 108
+.b8 104
+.b8 104
+.b8 54
+.b8 115
+.b8 117
+.b8 119
+.b8 111
+.b8 107
+.b8 107
+.b8 116
+.b8 112
+.b8 115
+.b8 51
+.b8 100
+.b8 119
+.b8 54
+.b8 50
+.b8 98
+.b8 116
+.b8 115
+.b8 107
+.b8 103
+.b8 120
+.b8 116
+.b8 52
+.b8 46
+.b8 112
+.b8 121
+.b8 0
+.b32 .debug_line
+.b8 47
+.b8 116
+.b8 109
+.b8 112
+.b8 47
+.b8 116
+.b8 111
+.b8 114
+.b8 99
+.b8 104
+.b8 105
+.b8 110
+.b8 100
+.b8 117
+.b8 99
+.b8 116
+.b8 111
+.b8 114
+.b8 95
+.b8 114
+.b8 111
+.b8 111
+.b8 116
+.b8 47
+.b8 119
+.b8 120
+.b8 0
+.b8 1
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 2
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 1
+.b8 156
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 101
+.b8 0
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 101
+.b8 0
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+	}
+	.section	.debug_pubnames
+	{
+.b32 $L__pubNames_end0-$L__pubNames_start0
+$L__pubNames_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 176
+.b32 125
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 101
+.b8 0
+.b32 0
+$L__pubNames_end0:
+	}
+	.section	.debug_pubtypes
+	{
+.b32 $L__pubTypes_end0-$L__pubTypes_start0
+$L__pubTypes_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 176
+.b32 0
+$L__pubTypes_end0:
+	}
+	.section	.debug_loc	{	}

.triton/dump/884b5df35d2a25fd91308249e7657806/triton_.ttir

@@ -0,0 +1,17 @@
+module {
+  tt.func public @triton__0d1de(%arg0: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg1: i64 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %c1024_i64 = arith.constant 1024 : i64
+    %cst = arith.constant dense<0.000000e+00> : tensor<1024xf32>
+    %0 = tt.get_program_id x : i32
+    %1 = arith.extsi %0 : i32 to i64
+    %2 = arith.muli %1, %c1024_i64 : i64
+    %3 = tt.make_range {end = 1024 : i32, start = 0 : i32} : tensor<1024xi32>
+    %4 = arith.extsi %3 : tensor<1024xi32> to tensor<1024xi64>
+    %5 = tt.splat %2 : (i64) -> tensor<1024xi64>
+    %6 = arith.addi %5, %4 : tensor<1024xi64>
+    %7 = tt.splat %arg0 : (!tt.ptr<f32, 1>) -> tensor<1024x!tt.ptr<f32, 1>>
+    %8 = tt.addptr %7, %6 : tensor<1024x!tt.ptr<f32, 1>>, tensor<1024xi64>
+    tt.store %8, %cst {cache = 1 : i32, evict = 1 : i32} : tensor<1024xf32>
+    tt.return
+  }
+}

.triton/dump/9f68cc707cb8f8bff3232abf59cbd9ec/triton_.ptx

@@ -0,0 +1,886 @@
+//
+// Generated by LLVM NVPTX Back-End
+//
+
+.version 8.2
+.target sm_89
+.address_size 64
+
+	// .globl	triton__0d1d2d3d4d5de6de
+.extern .func __assertfail
+(
+	.param .b64 __assertfail_param_0,
+	.param .b64 __assertfail_param_1,
+	.param .b32 __assertfail_param_2,
+	.param .b64 __assertfail_param_3,
+	.param .b64 __assertfail_param_4
+)
+;
+.global .align 1 .b8 assertFunc_1[25] = {95, 99, 97, 108, 108, 95, 119, 105, 116, 104, 95, 102, 114, 97, 109, 101, 115, 95, 114, 101, 109, 111, 118, 101, 100};
+.global .align 1 .b8 assertFile_1[38] = {60, 102, 114, 111, 122, 101, 110, 32, 105, 109, 112, 111, 114, 116, 108, 105, 98, 46, 95, 98, 111, 111, 116, 115, 116, 114, 97, 112, 95, 101, 120, 116, 101, 114, 110, 97, 108, 62};
+.global .align 1 .b8 assertMessage_1[39] = {105, 110, 100, 101, 120, 32, 111, 117, 116, 32, 111, 102, 32, 98, 111, 117, 110, 100, 115, 58, 32, 48, 32, 60, 61, 32, 116, 109, 112, 49, 51, 32, 60, 32, 53, 48, 50, 53, 55};
+.global .align 1 .b8 assertFunc_0[25] = {95, 99, 97, 108, 108, 95, 119, 105, 116, 104, 95, 102, 114, 97, 109, 101, 115, 95, 114, 101, 109, 111, 118, 101, 100};
+.global .align 1 .b8 assertFile_0[38] = {60, 102, 114, 111, 122, 101, 110, 32, 105, 109, 112, 111, 114, 116, 108, 105, 98, 46, 95, 98, 111, 111, 116, 115, 116, 114, 97, 112, 95, 101, 120, 116, 101, 114, 110, 97, 108, 62};
+.global .align 1 .b8 assertMessage_0[38] = {105, 110, 100, 101, 120, 32, 111, 117, 116, 32, 111, 102, 32, 98, 111, 117, 110, 100, 115, 58, 32, 48, 32, 60, 61, 32, 116, 109, 112, 51, 32, 60, 32, 53, 48, 50, 53, 55};
+.extern .shared .align 1 .b8 global_smem[];
+.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90, 0};
+
+.visible .entry triton__0d1d2d3d4d5de6de(
+	.param .u64 triton__0d1d2d3d4d5de6de_param_0,
+	.param .u64 triton__0d1d2d3d4d5de6de_param_1,
+	.param .u64 triton__0d1d2d3d4d5de6de_param_2,
+	.param .u64 triton__0d1d2d3d4d5de6de_param_3,
+	.param .u64 triton__0d1d2d3d4d5de6de_param_4,
+	.param .u32 triton__0d1d2d3d4d5de6de_param_5,
+	.param .u32 triton__0d1d2d3d4d5de6de_param_6
+)
+.maxntid 128, 1, 1
+{
+	.reg .pred 	%p<42>;
+	.reg .b16 	%rs<5>;
+	.reg .b32 	%r<140>;
+	.reg .f32 	%f<148>;
+	.reg .b64 	%rd<67>;
+	.loc	1 18 0
+$L__func_begin0:
+	.loc	1 18 0
+
+	ld.param.u64 	%rd17, [triton__0d1d2d3d4d5de6de_param_4];
+	ld.param.u64 	%rd16, [triton__0d1d2d3d4d5de6de_param_3];
+	ld.param.u64 	%rd28, [triton__0d1d2d3d4d5de6de_param_0];
+	ld.param.u64 	%rd29, [triton__0d1d2d3d4d5de6de_param_1];
+$L__tmp0:
+	.loc	1 22 44
+	mov.u32 	%r16, %tid.x;
+	and.b32  	%r1, %r16, 31;
+	ld.param.u64 	%rd30, [triton__0d1d2d3d4d5de6de_param_2];
+	bfe.u32 	%r2, %r16, 5, 2;
+	bfe.u32 	%r3, %r16, 1, 4;
+	shl.b32 	%r17, %r2, 4;
+	or.b32  	%r4, %r17, %r3;
+	and.b32  	%r18, %r16, 63;
+	.loc	1 24 33
+	shl.b32 	%r19, %r16, 2;
+	and.b32  	%r5, %r19, 4;
+	and.b32  	%r6, %r16, 7;
+	.loc	1 31 36
+	shl.b32 	%r7, %r2, 2;
+	.loc	1 21 28
+	mov.u32 %r14, %ctaid.x;
+	.loc	1 21 33
+	shl.b32 	%r20, %r14, 6;
+	.loc	1 22 23
+	or.b32  	%r21, %r20, %r4;
+	or.b32  	%r22, %r20, %r18;
+	.loc	1 26 30
+	mul.wide.s32 	%rd31, %r21, 8;
+	add.s64 	%rd19, %rd28, %rd31;
+	mul.wide.s32 	%rd32, %r22, 8;
+	add.s64 	%rd27, %rd28, %rd32;
+	mov.pred 	%p1, -1;
+	.loc	1 26 35
+	mov.u64 %rd18, 0x0;
+	@%p1 ld.global.L1::evict_last.b64 { %rd18 }, [ %rd19 + 0 ];
+	mov.u64 %rd20, 0x0;
+	@%p1 ld.global.L1::evict_last.b64 { %rd20 }, [ %rd19 + 0 ];
+	mov.u64 %rd22, 0x0;
+	@%p1 ld.global.L1::evict_last.b64 { %rd22 }, [ %rd19 + 0 ];
+	mov.u64 %rd24, 0x0;
+	@%p1 ld.global.L1::evict_last.b64 { %rd24 }, [ %rd19 + 0 ];
+	mov.u64 %rd26, 0x0;
+	@%p1 ld.global.L1::evict_last.b64 { %rd26 }, [ %rd27 + 0 ];
+	.loc	1 27 18
+	bfe.s32 	%r23, %r14, 25, 1;
+	shr.u32 	%r24, %r23, 23;
+	add.s32 	%r25, %r21, %r24;
+	and.b32  	%r26, %r25, 16776704;
+	sub.s32 	%r27, %r21, %r26;
+	.loc	1 35 44
+	shl.b32 	%r28, %r27, 8;
+	.loc	1 36 22
+	add.s64 	%rd33, %rd26, 50257;
+	.loc	1 37 22
+	setp.lt.s64 	%p6, %rd18, 0;
+	setp.lt.s64 	%p7, %rd26, 0;
+	.loc	1 38 36
+	selp.b64 	%rd1, %rd33, %rd26, %p7;
+	.loc	1 40 44
+	shl.b64 	%rd34, %rd18, 8;
+	add.s64 	%rd35, %rd34, 12865792;
+	selp.b64 	%rd36, %rd35, %rd34, %p6;
+	.loc	1 31 36
+	and.b32  	%r29, %r16, 1;
+	mul.wide.u32 	%rd2, %r29, 16;
+	shl.b64 	%rd37, %rd36, 2;
+	or.b64  	%rd38, %rd2, %rd37;
+	add.s64 	%rd66, %rd29, %rd38;
+	or.b32  	%r30, %r28, %r5;
+	mul.wide.s32 	%rd39, %r30, 4;
+	add.s64 	%rd64, %rd30, %rd39;
+	mov.f32 	%f132, 0f00000000;
+	mov.b32 	%r138, -8;
+	mov.u64 	%rd62, %rd64;
+	mov.u64 	%rd63, %rd66;
+	mov.f32 	%f133, %f132;
+	mov.f32 	%f134, %f132;
+	mov.f32 	%f135, %f132;
+	mov.f32 	%f136, %f132;
+	mov.f32 	%f137, %f132;
+	mov.f32 	%f138, %f132;
+	mov.f32 	%f139, %f132;
+	mov.f32 	%f140, %f132;
+	mov.f32 	%f141, %f132;
+	mov.f32 	%f142, %f132;
+	mov.f32 	%f143, %f132;
+	mov.f32 	%f144, %f132;
+	mov.f32 	%f145, %f132;
+	mov.f32 	%f146, %f132;
+	mov.f32 	%f147, %f132;
+	bra.uni 	$L__BB0_1;
+$L__BB0_3:
+	.loc	1 0 0
+	mov.b32 	%f17, %r31;
+	mov.b32 	%f18, %r32;
+	mov.b32 	%f19, %r33;
+	mov.b32 	%f20, %r34;
+	.loc	1 40 52
+	mov.u32 %r40, 0x0;
+	mov.u32 %r41, 0x0;
+	mov.u32 %r42, 0x0;
+	mov.u32 %r43, 0x0;
+	@%p1 ld.global.L1::evict_last.v4.b32 { %r40, %r41, %r42, %r43 }, [ %rd63 + 0 ];
+	@!%p1 mov.u32 %r40, %r124;
+	@!%p1 mov.u32 %r41, %r124;
+	@!%p1 mov.u32 %r42, %r124;
+	@!%p1 mov.u32 %r43, %r124;
+	mov.b32 	%f48, %r40;
+	mov.b32 	%f49, %r41;
+	mov.b32 	%f50, %r42;
+	mov.b32 	%f51, %r43;
+	.loc	1 41 22
+	add.f32 	%f52, %f17, %f48;
+	add.f32 	%f53, %f18, %f49;
+	add.f32 	%f54, %f19, %f50;
+	add.f32 	%f55, %f20, %f51;
+$L__tmp1:
+	.loc	2 96 20
+	sub.f32 	%f56, %f52, %f144;
+	sub.f32 	%f57, %f53, %f145;
+	sub.f32 	%f58, %f54, %f146;
+	sub.f32 	%f59, %f55, %f147;
+	.loc	2 97 26
+	add.f32 	%f132, %f132, 0f3F800000;
+	add.f32 	%f133, %f133, 0f3F800000;
+	add.f32 	%f134, %f134, 0f3F800000;
+	add.f32 	%f135, %f135, 0f3F800000;
+	add.f32 	%f136, %f136, 0f3F800000;
+	add.f32 	%f137, %f137, 0f3F800000;
+	add.f32 	%f138, %f138, 0f3F800000;
+	add.f32 	%f139, %f139, 0f3F800000;
+	.loc	2 98 30
+	mov.b32 	%r49, %f56;
+	mov.b32 	%r50, %f132;
+	div.full.f32 %r48, %r49, %r50;
+	mov.b32 	%f60, %r48;
+	mov.b32 	%r52, %f57;
+	mov.b32 	%r53, %f133;
+	div.full.f32 %r51, %r52, %r53;
+	mov.b32 	%f61, %r51;
+	mov.b32 	%r55, %f58;
+	mov.b32 	%r56, %f134;
+	div.full.f32 %r54, %r55, %r56;
+	mov.b32 	%f62, %r54;
+	mov.b32 	%r58, %f59;
+	mov.b32 	%r59, %f135;
+	div.full.f32 %r57, %r58, %r59;
+	mov.b32 	%f63, %r57;
+	.loc	2 98 22
+	add.f32 	%f144, %f144, %f60;
+	add.f32 	%f145, %f145, %f61;
+	add.f32 	%f146, %f146, %f62;
+	add.f32 	%f147, %f147, %f63;
+	.loc	2 101 30
+	sub.f32 	%f64, %f52, %f144;
+	sub.f32 	%f65, %f53, %f145;
+	sub.f32 	%f66, %f54, %f146;
+	sub.f32 	%f67, %f55, %f147;
+$L__tmp2:
+	.loc	1 47 48
+	fma.rn.f32 	%f140, %f56, %f64, %f140;
+	fma.rn.f32 	%f141, %f57, %f65, %f141;
+	fma.rn.f32 	%f142, %f58, %f66, %f142;
+	fma.rn.f32 	%f143, %f59, %f67, %f143;
+	.loc	1 31 36
+	add.s32 	%r138, %r138, 8;
+	add.s64 	%rd63, %rd63, 32;
+	add.s64 	%rd62, %rd62, 32;
+	setp.lt.u32 	%p19, %r138, 248;
+	@%p19 bra 	$L__BB0_1;
+	bra.uni 	$L__BB0_4;
+$L__BB0_1:
+	.loc	1 39 40
+	setp.lt.u64 	%p13, %rd1, 50257;
+	mov.b32 	%r124, 0;
+	.loc	1 35 50
+	mov.u32 %r31, 0x0;
+	mov.u32 %r32, 0x0;
+	mov.u32 %r33, 0x0;
+	mov.u32 %r34, 0x0;
+	@%p1 ld.global.L1::evict_last.v4.b32 { %r31, %r32, %r33, %r34 }, [ %rd62 + 0 ];
+	@!%p1 mov.u32 %r31, %r124;
+	@!%p1 mov.u32 %r32, %r124;
+	@!%p1 mov.u32 %r33, %r124;
+	@!%p1 mov.u32 %r34, %r124;
+	mov.b32 	%r137, 883;
+	mov.u64 	%rd61, 1;
+	.loc	1 39 55
+	@%p13 bra 	$L__BB0_3;
+	mov.u64 	%rd41, assertMessage_0;
+	cvta.global.u64 	%rd42, %rd41;
+	mov.u64 	%rd43, assertFile_0;
+	cvta.global.u64 	%rd44, %rd43;
+	mov.u64 	%rd45, assertFunc_0;
+	cvta.global.u64 	%rd46, %rd45;
+	{ // callseq 2, 0
+	.reg .b32 temp_param_reg;
+	.param .b64 param0;
+	st.param.b64 	[param0+0], %rd42;
+	.param .b64 param1;
+	st.param.b64 	[param1+0], %rd44;
+	.param .b32 param2;
+	st.param.b32 	[param2+0], %r137;
+	.param .b64 param3;
+	st.param.b64 	[param3+0], %rd46;
+	.param .b64 param4;
+	st.param.b64 	[param4+0], %rd61;
+	call.uni 
+	__assertfail, 
+	(
+	param0, 
+	param1, 
+	param2, 
+	param3, 
+	param4
+	);
+	} // callseq 2
+	bra.uni 	$L__BB0_3;
+$L__BB0_4:
+	.loc	1 31 36
+	shr.u32 	%r85, %r1, 3;
+	or.b32  	%r86, %r7, %r85;
+	mad.lo.s32 	%r87, %r86, 12, %r6;
+	shl.b32 	%r88, %r87, 2;
+	mov.u32 	%r89, global_smem;
+	add.s32 	%r90, %r89, %r88;
+	st.shared.f32 	[%r90], %f136;
+	st.shared.f32 	[%r90+768], %f137;
+	st.shared.f32 	[%r90+1536], %f138;
+	st.shared.f32 	[%r90+2304], %f139;
+	bar.sync 	0;
+	mad.lo.s32 	%r91, %r4, 12, %r5;
+	shl.b32 	%r92, %r91, 2;
+	add.s32 	%r93, %r89, %r92;
+	ld.shared.v4.f32 	{%f68, %f69, %f70, %f71}, [%r93];
+$L__tmp3:
+	.loc	2 108 21
+	sub.f32 	%f72, %f145, %f144;
+	.loc	2 109 28
+	add.f32 	%f73, %f68, %f69;
+	.loc	2 110 39
+	setp.eq.f32 	%p20, %f73, 0f00000000;
+	.loc	2 110 60
+	mov.b32 	%r61, %f69;
+	mov.b32 	%r62, %f73;
+	div.full.f32 %r60, %r61, %r62;
+	mov.b32 	%f74, %r60;
+	.loc	2 110 49
+	selp.f32 	%f75, 0f00000000, %f74, %p20;
+	.loc	2 112 17
+	fma.rn.f32 	%f76, %f72, %f75, %f144;
+	.loc	2 113 15
+	add.f32 	%f77, %f140, %f141;
+	.loc	2 113 30
+	mul.f32 	%f78, %f72, %f72;
+	.loc	2 113 38
+	mul.f32 	%f79, %f78, %f68;
+	.loc	2 113 22
+	fma.rn.f32 	%f80, %f79, %f75, %f77;
+	.loc	2 108 21
+	sub.f32 	%f81, %f146, %f76;
+	.loc	2 109 28
+	add.f32 	%f82, %f70, %f73;
+	.loc	2 110 39
+	setp.eq.f32 	%p21, %f82, 0f00000000;
+	.loc	2 110 60
+	mov.b32 	%r65, %f82;
+	mov.b32 	%r64, %f70;
+	div.full.f32 %r63, %r64, %r65;
+	mov.b32 	%f83, %r63;
+	.loc	2 110 49
+	selp.f32 	%f84, 0f00000000, %f83, %p21;
+	.loc	2 112 17
+	fma.rn.f32 	%f85, %f84, %f81, %f76;
+	.loc	2 113 15
+	add.f32 	%f86, %f142, %f80;
+	.loc	2 113 30
+	mul.f32 	%f87, %f81, %f81;
+	.loc	2 113 38
+	mul.f32 	%f88, %f73, %f87;
+	.loc	2 113 22
+	fma.rn.f32 	%f89, %f84, %f88, %f86;
+	.loc	2 108 21
+	sub.f32 	%f90, %f147, %f85;
+	.loc	2 109 28
+	add.f32 	%f91, %f71, %f82;
+	.loc	2 110 39
+	setp.eq.f32 	%p22, %f91, 0f00000000;
+	.loc	2 110 60
+	mov.b32 	%r68, %f91;
+	mov.b32 	%r67, %f71;
+	div.full.f32 %r66, %r67, %r68;
+	mov.b32 	%f92, %r66;
+	.loc	2 110 49
+	selp.f32 	%f93, 0f00000000, %f92, %p22;
+	.loc	2 112 17
+	fma.rn.f32 	%f94, %f93, %f90, %f85;
+	.loc	2 113 15
+	add.f32 	%f95, %f143, %f89;
+	.loc	2 113 30
+	mul.f32 	%f96, %f90, %f90;
+	.loc	2 113 38
+	mul.f32 	%f97, %f82, %f96;
+	.loc	2 113 22
+	fma.rn.f32 	%f98, %f93, %f97, %f95;
+$L__tmp4:
+	.loc	2 120 46
+	mov.b32 	%r94, %f94;
+	shfl.sync.bfly.b32	%r95, %r94, 1, 31, -1;
+	mov.b32 	%f99, %r95;
+	mov.b32 	%r96, %f98;
+	shfl.sync.bfly.b32	%r97, %r96, 1, 31, -1;
+	mov.b32 	%f100, %r97;
+	shfl.sync.bfly.b32	%r70, %r68, 1, 31, -1;
+	mov.b32 	%f101, %r70;
+$L__tmp5:
+	.loc	2 108 21
+	sub.f32 	%f102, %f99, %f94;
+	.loc	2 109 28
+	add.f32 	%f103, %f91, %f101;
+	.loc	2 110 39
+	setp.eq.f32 	%p23, %f103, 0f00000000;
+	.loc	2 110 60
+	mov.b32 	%r71, %f103;
+	div.full.f32 %r69, %r70, %r71;
+	mov.b32 	%f104, %r69;
+	.loc	2 110 49
+	selp.f32 	%f105, 0f00000000, %f104, %p23;
+	.loc	2 112 17
+	fma.rn.f32 	%f37, %f105, %f102, %f94;
+	.loc	2 113 15
+	add.f32 	%f106, %f98, %f100;
+	.loc	2 113 30
+	mul.f32 	%f107, %f102, %f102;
+	.loc	2 113 38
+	mul.f32 	%f108, %f91, %f107;
+	.loc	2 113 22
+	fma.rn.f32 	%f109, %f105, %f108, %f106;
+$L__tmp6:
+	.loc	1 69 23
+	mov.b32 	%r73, %f109;
+	mov.b32 	%r74, 1132462080;
+	div.full.f32 %r72, %r73, %r74;
+	mov.b32 	%f110, %r72;
+	.loc	1 71 24
+	add.f32 	%f38, %f110, 0f3727C5AC;
+	.loc	1 55 36
+	shl.b32 	%r98, %r14, 14;
+	shl.b32 	%r99, %r2, 12;
+	or.b32  	%r100, %r98, %r99;
+	shl.b32 	%r101, %r3, 8;
+	or.b32  	%r102, %r100, %r101;
+	or.b32  	%r11, %r102, %r5;
+	add.s64 	%rd65, %rd16, %rd2;
+	mov.b32 	%r139, -8;
+	rsqrt.approx.ftz.f32 	%f123, %f38;
+	bra.uni 	$L__BB0_5;
+$L__BB0_7:
+	.loc	1 65 54
+	mov.u32 %r120, 0x0;
+	mov.u32 %r121, 0x0;
+	mov.u32 %r122, 0x0;
+	mov.u32 %r123, 0x0;
+	@%p1 ld.global.L1::evict_first.v4.b32 { %r120, %r121, %r122, %r123 }, [ %rd66 + 0 ];
+	@!%p1 mov.u32 %r120, %r124;
+	@!%p1 mov.u32 %r121, %r124;
+	@!%p1 mov.u32 %r122, %r124;
+	@!%p1 mov.u32 %r123, %r124;
+	mov.b32 	%f111, %r120;
+	mov.b32 	%f112, %r121;
+	mov.b32 	%f113, %r122;
+	mov.b32 	%f114, %r123;
+	.loc	1 66 24
+	add.f32 	%f115, %f39, %f111;
+	add.f32 	%f116, %f40, %f112;
+	add.f32 	%f117, %f41, %f113;
+	add.f32 	%f118, %f42, %f114;
+	.loc	1 67 24
+	sub.f32 	%f119, %f115, %f37;
+	sub.f32 	%f120, %f116, %f37;
+	sub.f32 	%f121, %f117, %f37;
+	sub.f32 	%f122, %f118, %f37;
+	.loc	1 73 24
+	mul.f32 	%f124, %f119, %f123;
+	mul.f32 	%f125, %f120, %f123;
+	mul.f32 	%f126, %f121, %f123;
+	mul.f32 	%f127, %f122, %f123;
+	.loc	1 74 24
+	mul.f32 	%f128, %f124, %f43;
+	mul.f32 	%f129, %f125, %f44;
+	mul.f32 	%f130, %f126, %f45;
+	mul.f32 	%f131, %f127, %f46;
+	.loc	1 55 36
+	add.s32 	%r139, %r139, 8;
+	.loc	1 76 29
+	add.s32 	%r134, %r139, %r11;
+	mul.wide.s32 	%rd60, %r134, 2;
+	add.s64 	%rd59, %rd17, %rd60;
+	.loc	1 76 52
+	mov.b32 	%r128, %f128;
+	cvt.rn.bf16.f32 %rs1, %r128;
+	mov.b32 	%r129, %f129;
+	cvt.rn.bf16.f32 %rs2, %r129;
+	mov.b32 	%r130, %f130;
+	cvt.rn.bf16.f32 %rs3, %r130;
+	mov.b32 	%r131, %f131;
+	cvt.rn.bf16.f32 %rs4, %r131;
+	mov.b32 	%r135, {%rs1, %rs2};
+	mov.b32 	%r136, {%rs3, %rs4};
+	@%p1 st.global.v2.b32 [ %rd59 + 0 ], { %r135, %r136 };
+	.loc	1 55 36
+	add.s64 	%rd66, %rd66, 32;
+	add.s64 	%rd65, %rd65, 32;
+	add.s64 	%rd64, %rd64, 32;
+	setp.lt.u32 	%p41, %r139, 248;
+	@%p41 bra 	$L__BB0_5;
+	bra.uni 	$L__BB0_8;
+$L__BB0_5:
+	.loc	1 59 51
+	mov.u32 %r103, 0x0;
+	mov.u32 %r104, 0x0;
+	mov.u32 %r105, 0x0;
+	mov.u32 %r106, 0x0;
+	@%p1 ld.global.L1::evict_last.v4.b32 { %r103, %r104, %r105, %r106 }, [ %rd64 + 0 ];
+	@!%p1 mov.u32 %r103, %r124;
+	@!%p1 mov.u32 %r104, %r124;
+	@!%p1 mov.u32 %r105, %r124;
+	@!%p1 mov.u32 %r106, %r124;
+	mov.b32 	%f39, %r103;
+	mov.b32 	%f40, %r104;
+	mov.b32 	%f41, %r105;
+	mov.b32 	%f42, %r106;
+	.loc	1 60 40
+	mov.u32 %r111, 0x0;
+	mov.u32 %r112, 0x0;
+	mov.u32 %r113, 0x0;
+	mov.u32 %r114, 0x0;
+	@%p1 ld.global.L1::evict_last.v4.b32 { %r111, %r112, %r113, %r114 }, [ %rd65 + 0 ];
+	@!%p1 mov.u32 %r111, %r124;
+	@!%p1 mov.u32 %r112, %r124;
+	@!%p1 mov.u32 %r113, %r124;
+	@!%p1 mov.u32 %r114, %r124;
+	mov.b32 	%f43, %r111;
+	mov.b32 	%f44, %r112;
+	mov.b32 	%f45, %r113;
+	mov.b32 	%f46, %r114;
+	.loc	1 64 57
+	@%p13 bra 	$L__BB0_7;
+	mov.u64 	%rd51, assertMessage_1;
+	cvta.global.u64 	%rd52, %rd51;
+	mov.u64 	%rd53, assertFile_1;
+	cvta.global.u64 	%rd54, %rd53;
+	mov.u64 	%rd55, assertFunc_1;
+	cvta.global.u64 	%rd56, %rd55;
+	{ // callseq 3, 0
+	.reg .b32 temp_param_reg;
+	.param .b64 param0;
+	st.param.b64 	[param0+0], %rd52;
+	.param .b64 param1;
+	st.param.b64 	[param1+0], %rd54;
+	.param .b32 param2;
+	st.param.b32 	[param2+0], %r137;
+	.param .b64 param3;
+	st.param.b64 	[param3+0], %rd56;
+	.param .b64 param4;
+	st.param.b64 	[param4+0], %rd61;
+	call.uni 
+	__assertfail, 
+	(
+	param0, 
+	param1, 
+	param2, 
+	param3, 
+	param4
+	);
+	} // callseq 3
+	bra.uni 	$L__BB0_7;
+$L__BB0_8:
+	.loc	1 55 4
+	ret;
+$L__tmp7:
+$L__func_end0:
+
+}
+	// .globl	__nv_rsqrtf
+.visible .func  (.param .b32 func_retval0) __nv_rsqrtf(
+	.param .b32 __nv_rsqrtf_param_0
+)
+{
+	.reg .f32 	%f<3>;
+$L__func_begin1:
+
+	ld.param.f32 	%f1, [__nv_rsqrtf_param_0];
+	rsqrt.approx.ftz.f32 	%f2, %f1;
+	st.param.f32 	[func_retval0+0], %f2;
+	ret;
+$L__func_end1:
+
+}
+	.file	1 "/tmp/torchinductor_root/lh/clhe4a3stvufxafmq3kk5hodazz2efctffte646znjdnv3lqi5oa.py"
+	.file	2 "/usr/local/lib/python3.10/dist-packages/torch/_inductor/triton_helpers.py"
+	.section	.debug_abbrev
+	{
+.b8 1
+.b8 17
+.b8 1
+.b8 37
+.b8 8
+.b8 19
+.b8 5
+.b8 3
+.b8 8
+.b8 16
+.b8 6
+.b8 27
+.b8 8
+.b8 180
+.b8 66
+.b8 12
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 0
+.b8 0
+.b8 2
+.b8 46
+.b8 0
+.b8 135
+.b8 64
+.b8 8
+.b8 3
+.b8 8
+.b8 58
+.b8 11
+.b8 59
+.b8 11
+.b8 63
+.b8 12
+.b8 32
+.b8 11
+.b8 0
+.b8 0
+.b8 3
+.b8 46
+.b8 1
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 64
+.b8 10
+.b8 49
+.b8 19
+.b8 0
+.b8 0
+.b8 4
+.b8 29
+.b8 0
+.b8 49
+.b8 19
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 88
+.b8 11
+.b8 89
+.b8 11
+.b8 87
+.b8 11
+.b8 0
+.b8 0
+.b8 5
+.b8 29
+.b8 1
+.b8 49
+.b8 19
+.b8 17
+.b8 1
+.b8 18
+.b8 1
+.b8 88
+.b8 11
+.b8 89
+.b8 11
+.b8 87
+.b8 11
+.b8 0
+.b8 0
+.b8 0
+	}
+	.section	.debug_info
+	{
+.b32 298
+.b8 2
+.b8 0
+.b32 .debug_abbrev
+.b8 8
+.b8 1
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 0
+.b8 2
+.b8 0
+.b8 99
+.b8 108
+.b8 104
+.b8 101
+.b8 52
+.b8 97
+.b8 51
+.b8 115
+.b8 116
+.b8 118
+.b8 117
+.b8 102
+.b8 120
+.b8 97
+.b8 102
+.b8 109
+.b8 113
+.b8 51
+.b8 107
+.b8 107
+.b8 53
+.b8 104
+.b8 111
+.b8 100
+.b8 97
+.b8 122
+.b8 122
+.b8 50
+.b8 101
+.b8 102
+.b8 99
+.b8 116
+.b8 102
+.b8 102
+.b8 116
+.b8 101
+.b8 54
+.b8 52
+.b8 54
+.b8 122
+.b8 110
+.b8 106
+.b8 100
+.b8 110
+.b8 118
+.b8 51
+.b8 108
+.b8 113
+.b8 105
+.b8 53
+.b8 111
+.b8 97
+.b8 46
+.b8 112
+.b8 121
+.b8 0
+.b32 .debug_line
+.b8 47
+.b8 116
+.b8 109
+.b8 112
+.b8 47
+.b8 116
+.b8 111
+.b8 114
+.b8 99
+.b8 104
+.b8 105
+.b8 110
+.b8 100
+.b8 117
+.b8 99
+.b8 116
+.b8 111
+.b8 114
+.b8 95
+.b8 114
+.b8 111
+.b8 111
+.b8 116
+.b8 47
+.b8 108
+.b8 104
+.b8 0
+.b8 1
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 2
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 51
+.b8 100
+.b8 52
+.b8 100
+.b8 53
+.b8 100
+.b8 101
+.b8 54
+.b8 100
+.b8 101
+.b8 0
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 51
+.b8 100
+.b8 52
+.b8 100
+.b8 53
+.b8 100
+.b8 101
+.b8 54
+.b8 100
+.b8 101
+.b8 0
+.b8 1
+.b8 18
+.b8 1
+.b8 1
+.b8 3
+.b64 $L__func_begin0
+.b64 $L__func_end0
+.b8 1
+.b8 156
+.b32 125
+.b8 4
+.b32 125
+.b64 $L__tmp1
+.b64 $L__tmp2
+.b8 2
+.b8 44
+.b8 38
+.b8 5
+.b32 125
+.b64 $L__tmp3
+.b64 $L__tmp6
+.b8 2
+.b8 50
+.b8 41
+.b8 4
+.b32 125
+.b64 $L__tmp3
+.b64 $L__tmp6
+.b8 2
+.b8 120
+.b8 46
+.b8 0
+.b8 4
+.b32 125
+.b64 $L__tmp4
+.b64 $L__tmp5
+.b8 2
+.b8 50
+.b8 41
+.b8 0
+.b8 0
+	}
+	.section	.debug_pubnames
+	{
+.b32 $L__pubNames_end0-$L__pubNames_start0
+$L__pubNames_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 302
+.b32 125
+.b8 116
+.b8 114
+.b8 105
+.b8 116
+.b8 111
+.b8 110
+.b8 95
+.b8 95
+.b8 48
+.b8 100
+.b8 49
+.b8 100
+.b8 50
+.b8 100
+.b8 51
+.b8 100
+.b8 52
+.b8 100
+.b8 53
+.b8 100
+.b8 101
+.b8 54
+.b8 100
+.b8 101
+.b8 0
+.b32 0
+$L__pubNames_end0:
+	}
+	.section	.debug_pubtypes
+	{
+.b32 $L__pubTypes_end0-$L__pubTypes_start0
+$L__pubTypes_start0:
+.b8 2
+.b8 0
+.b32 .debug_info
+.b32 302
+.b32 0
+$L__pubTypes_end0:
+	}
+	.section	.debug_loc	{	}