326 lines · plain
1; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_80 -mattr=+ptx70 | FileCheck %s2; RUN: %if ptxas-sm_80 && ptxas-isa-7.0 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_80 -mattr=+ptx70 | %ptxas-verify -arch=sm_80 %}3 4declare bfloat @llvm.nvvm.abs.bf16(bfloat)5declare <2 x bfloat> @llvm.nvvm.abs.bf16x2(<2 x bfloat>)6 7; CHECK-LABEL: abs_bf168define bfloat @abs_bf16(bfloat %0) {9 ; CHECK-NOT: call10 ; CHECK: abs.bf1611 %res = call bfloat @llvm.nvvm.abs.bf16(bfloat %0);12 ret bfloat %res13}14 15; CHECK-LABEL: abs_bf16x216define <2 x bfloat> @abs_bf16x2(<2 x bfloat> %0) {17 ; CHECK-NOT: call18 ; CHECK: abs.bf16x219 %res = call <2 x bfloat> @llvm.nvvm.abs.bf16x2(<2 x bfloat> %0);20 ret <2 x bfloat> %res21}22 23; CHECK-LABEL: neg_bf1624define bfloat @neg_bf16(bfloat %0) {25 ; CHECK-NOT: call26 ; CHECK: neg.bf1627 %res = call bfloat @llvm.nvvm.neg.bf16(bfloat %0);28 ret bfloat %res29}30 31; CHECK-LABEL: neg_bf16x232define <2 x bfloat> @neg_bf16x2(<2 x bfloat> %0) {33 ; CHECK-NOT: call34 ; CHECK: neg.bf16x235 %res = call <2 x bfloat> @llvm.nvvm.neg.bf16x2(<2 x bfloat> %0);36 ret <2 x bfloat> %res37}38 39; CHECK-LABEL: fmin_nan_f40define float @fmin_nan_f(float %0, float %1) {41 ; CHECK-NOT: call42 ; CHECK: min.NaN.f3243 %res = call float @llvm.nvvm.fmin.nan.f(float %0, float %1);44 ret float %res45}46 47; CHECK-LABEL: fmin_ftz_nan_f48define float @fmin_ftz_nan_f(float %0, float %1) {49 ; CHECK-NOT: call50 ; CHECK: min.ftz.NaN.f3251 %res = call float @llvm.nvvm.fmin.ftz.nan.f(float %0, float %1);52 ret float %res53}54 55; CHECK-LABEL: fmin_f1656define half @fmin_f16(half %0, half %1) {57 ; CHECK-NOT: call58 ; CHECK: min.f1659 %res = call half @llvm.nvvm.fmin.f16(half %0, half %1)60 ret half %res61}62 63; CHECK-LABEL: fmin_ftz_f1664define half @fmin_ftz_f16(half %0, half %1) {65 ; CHECK-NOT: call66 ; CHECK: min.ftz.f1667 %res = call half @llvm.nvvm.fmin.ftz.f16(half %0, half %1)68 ret half %res69}70 71; CHECK-LABEL: fmin_nan_f1672define half @fmin_nan_f16(half %0, half %1) {73 ; CHECK-NOT: call74 ; CHECK: min.NaN.f1675 %res = call half @llvm.nvvm.fmin.nan.f16(half %0, half %1)76 ret half %res77}78 79; CHECK-LABEL: fmin_ftz_nan_f1680define half @fmin_ftz_nan_f16(half %0, half %1) {81 ; CHECK-NOT: call82 ; CHECK: min.ftz.NaN.f1683 %res = call half @llvm.nvvm.fmin.ftz.nan.f16(half %0, half %1)84 ret half %res85}86 87; CHECK-LABEL: fmin_f16x288define <2 x half> @fmin_f16x2(<2 x half> %0, <2 x half> %1) {89 ; CHECK-NOT: call90 ; CHECK: min.f16x291 %res = call <2 x half> @llvm.nvvm.fmin.f16x2(<2 x half> %0, <2 x half> %1)92 ret <2 x half> %res93}94 95; CHECK-LABEL: fmin_ftz_f16x296define <2 x half> @fmin_ftz_f16x2(<2 x half> %0, <2 x half> %1) {97 ; CHECK-NOT: call98 ; CHECK: min.ftz.f16x299 %res = call <2 x half> @llvm.nvvm.fmin.ftz.f16x2(<2 x half> %0, <2 x half> %1)100 ret <2 x half> %res101}102 103; CHECK-LABEL: fmin_nan_f16x2104define <2 x half> @fmin_nan_f16x2(<2 x half> %0, <2 x half> %1) {105 ; CHECK-NOT: call106 ; CHECK: min.NaN.f16x2107 %res = call <2 x half> @llvm.nvvm.fmin.nan.f16x2(<2 x half> %0, <2 x half> %1)108 ret <2 x half> %res109}110 111; CHECK-LABEL: fmin_ftz_nan_f16x2112define <2 x half> @fmin_ftz_nan_f16x2(<2 x half> %0, <2 x half> %1) {113 ; CHECK-NOT: call114 ; CHECK: min.ftz.NaN.f16x2115 %res = call <2 x half> @llvm.nvvm.fmin.ftz.nan.f16x2(<2 x half> %0, <2 x half> %1)116 ret <2 x half> %res117}118 119; CHECK-LABEL: fmin_bf16120define bfloat @fmin_bf16(bfloat %0, bfloat %1) {121 ; CHECK-NOT: call122 ; CHECK: min.bf16123 %res = call bfloat @llvm.nvvm.fmin.bf16(bfloat %0, bfloat %1)124 ret bfloat %res125}126 127; CHECK-LABEL: fmin_nan_bf16128define bfloat @fmin_nan_bf16(bfloat %0, bfloat %1) {129 ; CHECK-NOT: call130 ; CHECK: min.NaN.bf16131 %res = call bfloat @llvm.nvvm.fmin.nan.bf16(bfloat %0, bfloat %1)132 ret bfloat %res133}134 135; CHECK-LABEL: fmin_bf16x2136define <2 x bfloat> @fmin_bf16x2(<2 x bfloat> %0, <2 x bfloat> %1) {137 ; CHECK-NOT: call138 ; CHECK: min.bf16x2139 %res = call <2 x bfloat> @llvm.nvvm.fmin.bf16x2(<2 x bfloat> %0, <2 x bfloat> %1)140 ret <2 x bfloat> %res141}142 143; CHECK-LABEL: fmin_nan_bf16x2144define <2 x bfloat> @fmin_nan_bf16x2(<2 x bfloat> %0, <2 x bfloat> %1) {145 ; CHECK-NOT: call146 ; CHECK: min.NaN.bf16x2147 %res = call <2 x bfloat> @llvm.nvvm.fmin.nan.bf16x2(<2 x bfloat> %0, <2 x bfloat> %1)148 ret <2 x bfloat> %res149}150 151; CHECK-LABEL: fmax_nan_f152define float @fmax_nan_f(float %0, float %1) {153 ; CHECK-NOT: call154 ; CHECK: max.NaN.f32155 %res = call float @llvm.nvvm.fmax.nan.f(float %0, float %1);156 ret float %res157}158 159; CHECK-LABEL: fmax_ftz_nan_f160define float @fmax_ftz_nan_f(float %0, float %1) {161 ; CHECK-NOT: call162 ; CHECK: max.ftz.NaN.f32163 %res = call float @llvm.nvvm.fmax.ftz.nan.f(float %0, float %1);164 ret float %res165}166 167; CHECK-LABEL: fmax_f16168define half @fmax_f16(half %0, half %1) {169 ; CHECK-NOT: call170 ; CHECK: max.f16171 %res = call half @llvm.nvvm.fmax.f16(half %0, half %1)172 ret half %res173}174 175; CHECK-LABEL: fmax_ftz_f16176define half @fmax_ftz_f16(half %0, half %1) {177 ; CHECK-NOT: call178 ; CHECK: max.ftz.f16179 %res = call half @llvm.nvvm.fmax.ftz.f16(half %0, half %1)180 ret half %res181}182 183; CHECK-LABEL: fmax_nan_f16184define half @fmax_nan_f16(half %0, half %1) {185 ; CHECK-NOT: call186 ; CHECK: max.NaN.f16187 %res = call half @llvm.nvvm.fmax.nan.f16(half %0, half %1)188 ret half %res189}190 191; CHECK-LABEL: fmax_ftz_nan_f16192define half @fmax_ftz_nan_f16(half %0, half %1) {193 ; CHECK-NOT: call194 ; CHECK: max.ftz.NaN.f16195 %res = call half @llvm.nvvm.fmax.ftz.nan.f16(half %0, half %1)196 ret half %res197}198 199; CHECK-LABEL: fmax_f16x2200define <2 x half> @fmax_f16x2(<2 x half> %0, <2 x half> %1) {201 ; CHECK-NOT: call202 ; CHECK: max.f16x2203 %res = call <2 x half> @llvm.nvvm.fmax.f16x2(<2 x half> %0, <2 x half> %1)204 ret <2 x half> %res205}206 207; CHECK-LABEL: fmax_ftz_f16x2208define <2 x half> @fmax_ftz_f16x2(<2 x half> %0, <2 x half> %1) {209 ; CHECK-NOT: call210 ; CHECK: max.ftz.f16x2211 %res = call <2 x half> @llvm.nvvm.fmax.ftz.f16x2(<2 x half> %0, <2 x half> %1)212 ret <2 x half> %res213}214 215; CHECK-LABEL: fmax_nan_f16x2216define <2 x half> @fmax_nan_f16x2(<2 x half> %0, <2 x half> %1) {217 ; CHECK-NOT: call218 ; CHECK: max.NaN.f16x2219 %res = call <2 x half> @llvm.nvvm.fmax.nan.f16x2(<2 x half> %0, <2 x half> %1)220 ret <2 x half> %res221}222 223; CHECK-LABEL: fmax_ftz_nan_f16x2224define <2 x half> @fmax_ftz_nan_f16x2(<2 x half> %0, <2 x half> %1) {225 ; CHECK-NOT: call226 ; CHECK: max.ftz.NaN.f16x2227 %res = call <2 x half> @llvm.nvvm.fmax.ftz.nan.f16x2(<2 x half> %0, <2 x half> %1)228 ret <2 x half> %res229}230 231; CHECK-LABEL: fmax_bf16232define bfloat @fmax_bf16(bfloat %0, bfloat %1) {233 ; CHECK-NOT: call234 ; CHECK: max.bf16235 %res = call bfloat @llvm.nvvm.fmax.bf16(bfloat %0, bfloat %1)236 ret bfloat %res237}238 239; CHECK-LABEL: fmax_nan_bf16240define bfloat @fmax_nan_bf16(bfloat %0, bfloat %1) {241 ; CHECK-NOT: call242 ; CHECK: max.NaN.bf16243 %res = call bfloat @llvm.nvvm.fmax.nan.bf16(bfloat %0, bfloat %1)244 ret bfloat %res245}246 247; CHECK-LABEL: fmax_bf16x2248define <2 x bfloat> @fmax_bf16x2(<2 x bfloat> %0, <2 x bfloat> %1) {249 ; CHECK-NOT: call250 ; CHECK: max.bf16x2251 %res = call <2 x bfloat> @llvm.nvvm.fmax.bf16x2(<2 x bfloat> %0, <2 x bfloat> %1)252 ret <2 x bfloat> %res253}254 255; CHECK-LABEL: fmax_nan_bf16x2256define <2 x bfloat> @fmax_nan_bf16x2(<2 x bfloat> %0, <2 x bfloat> %1) {257 ; CHECK-NOT: call258 ; CHECK: max.NaN.bf16x2259 %res = call <2 x bfloat> @llvm.nvvm.fmax.nan.bf16x2(<2 x bfloat> %0, <2 x bfloat> %1)260 ret <2 x bfloat> %res261}262 263; CHECK-LABEL: fma_rn_relu_f16264define half @fma_rn_relu_f16(half %0, half %1, half %2) {265 ; CHECK-NOT: call266 ; CHECK: fma.rn.relu.f16267 %res = call half @llvm.nvvm.fma.rn.relu.f16(half %0, half %1, half %2)268 ret half %res269}270 271; CHECK-LABEL: fma_rn_ftz_relu_f16272define half @fma_rn_ftz_relu_f16(half %0, half %1, half %2) {273 ; CHECK-NOT: call274 ; CHECK: fma.rn.ftz.relu.f16275 %res = call half @llvm.nvvm.fma.rn.ftz.relu.f16(half %0, half %1, half %2)276 ret half %res277}278 279; CHECK-LABEL: fma_rn_relu_f16x2280define <2 x half> @fma_rn_relu_f16x2(<2 x half> %0, <2 x half> %1, <2 x half> %2) {281 ; CHECK-NOT: call282 ; CHECK: fma.rn.relu.f16x2283 %res = call <2 x half> @llvm.nvvm.fma.rn.relu.f16x2(<2 x half> %0, <2 x half> %1, <2 x half> %2)284 ret <2 x half> %res285}286 287; CHECK-LABEL: fma_rn_ftz_relu_f16x2288define <2 x half> @fma_rn_ftz_relu_f16x2(<2 x half> %0, <2 x half> %1, <2 x half> %2) {289 ; CHECK-NOT: call290 ; CHECK: fma.rn.ftz.relu.f16x2291 %res = call <2 x half> @llvm.nvvm.fma.rn.ftz.relu.f16x2(<2 x half> %0, <2 x half> %1, <2 x half> %2)292 ret <2 x half> %res293}294 295; CHECK-LABEL: fma_rn_bf16296define bfloat @fma_rn_bf16(bfloat %0, bfloat %1, bfloat %2) {297 ; CHECK-NOT: call298 ; CHECK: fma.rn.bf16299 %res = call bfloat @llvm.nvvm.fma.rn.bf16(bfloat %0, bfloat %1, bfloat %2)300 ret bfloat %res301}302 303; CHECK-LABEL: fma_rn_relu_bf16304define bfloat @fma_rn_relu_bf16(bfloat %0, bfloat %1, bfloat %2) {305 ; CHECK-NOT: call306 ; CHECK: fma.rn.relu.bf16307 %res = call bfloat @llvm.nvvm.fma.rn.relu.bf16(bfloat %0, bfloat %1, bfloat %2)308 ret bfloat %res309}310 311; CHECK-LABEL: fma_rn_bf16x2312define <2 x bfloat> @fma_rn_bf16x2(<2 x bfloat> %0, <2 x bfloat> %1, <2 x bfloat> %2) {313 ; CHECK-NOT: call314 ; CHECK: fma.rn.bf16x2315 %res = call <2 x bfloat> @llvm.nvvm.fma.rn.bf16x2(<2 x bfloat> %0, <2 x bfloat> %1, <2 x bfloat> %2)316 ret <2 x bfloat> %res317}318 319; CHECK-LABEL: fma_rn_relu_bf16x2320define <2 x bfloat> @fma_rn_relu_bf16x2(<2 x bfloat> %0, <2 x bfloat> %1, <2 x bfloat> %2) {321 ; CHECK-NOT: call322 ; CHECK: fma.rn.relu.bf16x2323 %res = call <2 x bfloat> @llvm.nvvm.fma.rn.relu.bf16x2(<2 x bfloat> %0, <2 x bfloat> %1, <2 x bfloat> %2)324 ret <2 x bfloat> %res325}326