223 lines · plain
1; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefix=GCN %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -enable-var-scope -check-prefix=GCN %s3 4; GCN-LABEL: {{^}}test_fmax_f32_ieee_mode_on:5; GCN: v_mul_f32_e64 [[QUIET0:v[0-9]+]], 1.0, s{{[0-9]+}}6; GCN: v_mul_f32_e64 [[QUIET1:v[0-9]+]], 1.0, s{{[0-9]+}}7; GCN: v_max_f32_e32 [[RESULT:v[0-9]+]], [[QUIET1]], [[QUIET0]]8; GCN-NOT: [[RESULT]]9; GCN: buffer_store_dword [[RESULT]]10define amdgpu_kernel void @test_fmax_f32_ieee_mode_on(ptr addrspace(1) %out, float %a, float %b) #0 {11 %val = call float @llvm.maxnum.f32(float %a, float %b) #112 store float %val, ptr addrspace(1) %out, align 413 ret void14}15 16; GCN-LABEL: {{^}}test_fmax_f32_ieee_mode_off:17; GCN: v_max_f32_e32 v0, v0, v118; GCN-NEXT: ; return19define amdgpu_ps float @test_fmax_f32_ieee_mode_off(float %a, float %b) #0 {20 %val = call float @llvm.maxnum.f32(float %a, float %b) #121 ret float %val22}23 24; GCN-LABEL: {{^}}test_fmax_v2f32:25; GCN: v_max_f32_e3226; GCN: v_max_f32_e3227define amdgpu_kernel void @test_fmax_v2f32(ptr addrspace(1) %out, <2 x float> %a, <2 x float> %b) #0 {28 %val = call <2 x float> @llvm.maxnum.v2f32(<2 x float> %a, <2 x float> %b)29 store <2 x float> %val, ptr addrspace(1) %out, align 830 ret void31}32 33; GCN-LABEL: {{^}}test_fmax_v3f32:34; GCN: v_max_f32_e3235; GCN: v_max_f32_e3236; GCN: v_max_f32_e3237; GCN-NOT: v_max_f3238define amdgpu_kernel void @test_fmax_v3f32(ptr addrspace(1) %out, <3 x float> %a, <3 x float> %b) nounwind {39 %val = call <3 x float> @llvm.maxnum.v3f32(<3 x float> %a, <3 x float> %b) #040 store <3 x float> %val, ptr addrspace(1) %out, align 1641 ret void42}43 44; GCN-LABEL: {{^}}test_fmax_v4f32:45; GCN: v_max_f32_e3246; GCN: v_max_f32_e3247; GCN: v_max_f32_e3248; GCN: v_max_f32_e3249define amdgpu_kernel void @test_fmax_v4f32(ptr addrspace(1) %out, <4 x float> %a, <4 x float> %b) #0 {50 %val = call <4 x float> @llvm.maxnum.v4f32(<4 x float> %a, <4 x float> %b)51 store <4 x float> %val, ptr addrspace(1) %out, align 1652 ret void53}54 55; GCN-LABEL: {{^}}test_fmax_v8f32:56; GCN: v_max_f32_e3257; GCN: v_max_f32_e3258; GCN: v_max_f32_e3259; GCN: v_max_f32_e3260; GCN: v_max_f32_e3261; GCN: v_max_f32_e3262; GCN: v_max_f32_e3263; GCN: v_max_f32_e3264define amdgpu_kernel void @test_fmax_v8f32(ptr addrspace(1) %out, <8 x float> %a, <8 x float> %b) #0 {65 %val = call <8 x float> @llvm.maxnum.v8f32(<8 x float> %a, <8 x float> %b)66 store <8 x float> %val, ptr addrspace(1) %out, align 3267 ret void68}69 70; GCN-LABEL: {{^}}test_fmax_v16f32:71; GCN: v_max_f32_e3272; GCN: v_max_f32_e3273; GCN: v_max_f32_e3274; GCN: v_max_f32_e3275; GCN: v_max_f32_e3276; GCN: v_max_f32_e3277; GCN: v_max_f32_e3278; GCN: v_max_f32_e3279; GCN: v_max_f32_e3280; GCN: v_max_f32_e3281; GCN: v_max_f32_e3282; GCN: v_max_f32_e3283; GCN: v_max_f32_e3284; GCN: v_max_f32_e3285; GCN: v_max_f32_e3286; GCN: v_max_f32_e3287define amdgpu_kernel void @test_fmax_v16f32(ptr addrspace(1) %out, <16 x float> %a, <16 x float> %b) #0 {88 %val = call <16 x float> @llvm.maxnum.v16f32(<16 x float> %a, <16 x float> %b)89 store <16 x float> %val, ptr addrspace(1) %out, align 6490 ret void91}92 93; GCN-LABEL: {{^}}constant_fold_fmax_f32:94; GCN-NOT: v_max_f32_e3295; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 2.096; GCN: buffer_store_dword [[REG]]97define amdgpu_kernel void @constant_fold_fmax_f32(ptr addrspace(1) %out) #0 {98 %val = call float @llvm.maxnum.f32(float 1.0, float 2.0)99 store float %val, ptr addrspace(1) %out, align 4100 ret void101}102 103; GCN-LABEL: {{^}}constant_fold_fmax_f32_nan_nan:104; GCN-NOT: v_max_f32_e32105; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7fc00000106; GCN: buffer_store_dword [[REG]]107define amdgpu_kernel void @constant_fold_fmax_f32_nan_nan(ptr addrspace(1) %out) #0 {108 %val = call float @llvm.maxnum.f32(float 0x7FF8000000000000, float 0x7FF8000000000000)109 store float %val, ptr addrspace(1) %out, align 4110 ret void111}112 113; GCN-LABEL: {{^}}constant_fold_fmax_f32_val_nan:114; GCN-NOT: v_max_f32_e32115; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0116; GCN: buffer_store_dword [[REG]]117define amdgpu_kernel void @constant_fold_fmax_f32_val_nan(ptr addrspace(1) %out) #0 {118 %val = call float @llvm.maxnum.f32(float 1.0, float 0x7FF8000000000000)119 store float %val, ptr addrspace(1) %out, align 4120 ret void121}122 123; GCN-LABEL: {{^}}constant_fold_fmax_f32_nan_val:124; GCN-NOT: v_max_f32_e32125; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 1.0126; GCN: buffer_store_dword [[REG]]127define amdgpu_kernel void @constant_fold_fmax_f32_nan_val(ptr addrspace(1) %out) #0 {128 %val = call float @llvm.maxnum.f32(float 0x7FF8000000000000, float 1.0)129 store float %val, ptr addrspace(1) %out, align 4130 ret void131}132 133; GCN-LABEL: {{^}}constant_fold_fmax_f32_p0_p0:134; GCN-NOT: v_max_f32_e32135; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0136; GCN: buffer_store_dword [[REG]]137define amdgpu_kernel void @constant_fold_fmax_f32_p0_p0(ptr addrspace(1) %out) #0 {138 %val = call float @llvm.maxnum.f32(float 0.0, float 0.0)139 store float %val, ptr addrspace(1) %out, align 4140 ret void141}142 143; GCN-LABEL: {{^}}constant_fold_fmax_f32_p0_n0:144; GCN-NOT: v_max_f32_e32145; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0146; GCN: buffer_store_dword [[REG]]147define amdgpu_kernel void @constant_fold_fmax_f32_p0_n0(ptr addrspace(1) %out) #0 {148 %val = call float @llvm.maxnum.f32(float 0.0, float -0.0)149 store float %val, ptr addrspace(1) %out, align 4150 ret void151}152 153; GCN-LABEL: {{^}}constant_fold_fmax_f32_n0_p0:154; GCN-NOT: v_max_f32_e32155; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0{{$}}156; GCN: buffer_store_dword [[REG]]157define amdgpu_kernel void @constant_fold_fmax_f32_n0_p0(ptr addrspace(1) %out) #0 {158 %val = call float @llvm.maxnum.f32(float -0.0, float 0.0)159 store float %val, ptr addrspace(1) %out, align 4160 ret void161}162 163; GCN-LABEL: {{^}}constant_fold_fmax_f32_n0_n0:164; GCN-NOT: v_max_f32_e32165; GCN: v_bfrev_b32_e32 [[REG:v[0-9]+]], 1{{$}}166; GCN: buffer_store_dword [[REG]]167define amdgpu_kernel void @constant_fold_fmax_f32_n0_n0(ptr addrspace(1) %out) #0 {168 %val = call float @llvm.maxnum.f32(float -0.0, float -0.0)169 store float %val, ptr addrspace(1) %out, align 4170 ret void171}172 173; GCN-LABEL: {{^}}fmax_var_immediate_f32_no_ieee:174; GCN: v_max_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, 2.0175define amdgpu_ps float @fmax_var_immediate_f32_no_ieee(float inreg %a) #0 {176 %val = call float @llvm.maxnum.f32(float %a, float 2.0) #0177 ret float %val178}179 180; GCN-LABEL: {{^}}fmax_immediate_var_f32_no_ieee:181; GCN: v_max_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, 2.0182define amdgpu_ps float @fmax_immediate_var_f32_no_ieee(float inreg %a) #0 {183 %val = call float @llvm.maxnum.f32(float 2.0, float %a) #0184 ret float %val185}186 187; GCN-LABEL: {{^}}fmax_var_literal_f32_no_ieee:188; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000189; GCN: v_max_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]190define amdgpu_ps float @fmax_var_literal_f32_no_ieee(float inreg %a) #0 {191 %val = call float @llvm.maxnum.f32(float %a, float 99.0) #0192 ret float %val193}194 195; GCN-LABEL: {{^}}fmax_literal_var_f32_no_ieee:196; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x42c60000197; GCN: v_max_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[REG]]198define amdgpu_ps float @fmax_literal_var_f32_no_ieee(float inreg %a) #0 {199 %val = call float @llvm.maxnum.f32(float 99.0, float %a) #0200 ret float %val201}202 203; GCN-LABEL: {{^}}test_func_fmax_v3f32:204; GCN: v_max_f32_e32205; GCN: v_max_f32_e32206; GCN: v_max_f32_e32207; GCN-NOT: v_max_f32208define <3 x float> @test_func_fmax_v3f32(<3 x float> %a, <3 x float> %b) #0 {209 %val = call <3 x float> @llvm.maxnum.v3f32(<3 x float> %a, <3 x float> %b) #0210 ret <3 x float> %val211}212 213declare float @llvm.maxnum.f32(float, float) #1214declare <2 x float> @llvm.maxnum.v2f32(<2 x float>, <2 x float>) #1215declare <3 x float> @llvm.maxnum.v3f32(<3 x float>, <3 x float>) #1216declare <4 x float> @llvm.maxnum.v4f32(<4 x float>, <4 x float>) #1217declare <8 x float> @llvm.maxnum.v8f32(<8 x float>, <8 x float>) #1218declare <16 x float> @llvm.maxnum.v16f32(<16 x float>, <16 x float>) #1219declare double @llvm.maxnum.f64(double, double)220 221attributes #0 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }222attributes #1 = { nounwind readnone }223