109 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1030 < %s | FileCheck -check-prefix=GFX10 %s3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1030 < %s | FileCheck -check-prefix=GFX10 %s4; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX11 %s5; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX11 %s6 7define float @v_fma(float %a, float %b, float %c) {8; GFX10-LABEL: v_fma:9; GFX10: ; %bb.0:10; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)11; GFX10-NEXT: v_fma_legacy_f32 v0, v0, v1, v212; GFX10-NEXT: s_setpc_b64 s[30:31]13;14; GFX11-LABEL: v_fma:15; GFX11: ; %bb.0:16; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)17; GFX11-NEXT: v_fma_dx9_zero_f32 v0, v0, v1, v218; GFX11-NEXT: s_setpc_b64 s[30:31]19 %fma = call float @llvm.amdgcn.fma.legacy(float %a, float %b, float %c)20 ret float %fma21}22 23define float @v_fma_imm(float %a, float %c) {24; GFX10-LABEL: v_fma_imm:25; GFX10: ; %bb.0:26; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)27; GFX10-NEXT: v_fma_legacy_f32 v0, 0x41200000, v0, v128; GFX10-NEXT: s_setpc_b64 s[30:31]29;30; GFX11-LABEL: v_fma_imm:31; GFX11: ; %bb.0:32; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)33; GFX11-NEXT: v_fma_dx9_zero_f32 v0, 0x41200000, v0, v134; GFX11-NEXT: s_setpc_b64 s[30:31]35 %fma = call float @llvm.amdgcn.fma.legacy(float %a, float 10.0, float %c)36 ret float %fma37}38 39define float @v_fabs_fma(float %a, float %b, float %c) {40; GFX10-LABEL: v_fabs_fma:41; GFX10: ; %bb.0:42; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)43; GFX10-NEXT: v_fma_legacy_f32 v0, |v0|, v1, v244; GFX10-NEXT: s_setpc_b64 s[30:31]45;46; GFX11-LABEL: v_fabs_fma:47; GFX11: ; %bb.0:48; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)49; GFX11-NEXT: v_fma_dx9_zero_f32 v0, |v0|, v1, v250; GFX11-NEXT: s_setpc_b64 s[30:31]51 %fabs.a = call float @llvm.fabs.f32(float %a)52 %fma = call float @llvm.amdgcn.fma.legacy(float %fabs.a, float %b, float %c)53 ret float %fma54}55 56define float @v_fneg_fabs_fma(float %a, float %b, float %c) {57; GFX10-LABEL: v_fneg_fabs_fma:58; GFX10: ; %bb.0:59; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)60; GFX10-NEXT: v_fma_legacy_f32 v0, v0, -|v1|, v261; GFX10-NEXT: s_setpc_b64 s[30:31]62;63; GFX11-LABEL: v_fneg_fabs_fma:64; GFX11: ; %bb.0:65; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)66; GFX11-NEXT: v_fma_dx9_zero_f32 v0, v0, -|v1|, v267; GFX11-NEXT: s_setpc_b64 s[30:31]68 %fabs.b = call float @llvm.fabs.f32(float %b)69 %neg.fabs.b = fneg float %fabs.b70 %fma = call float @llvm.amdgcn.fma.legacy(float %a, float %neg.fabs.b, float %c)71 ret float %fma72}73 74define float @v_fneg_fma(float %a, float %b, float %c) {75; GFX10-LABEL: v_fneg_fma:76; GFX10: ; %bb.0:77; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)78; GFX10-NEXT: v_fma_legacy_f32 v0, v0, v1, -v279; GFX10-NEXT: s_setpc_b64 s[30:31]80;81; GFX11-LABEL: v_fneg_fma:82; GFX11: ; %bb.0:83; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)84; GFX11-NEXT: v_fma_dx9_zero_f32 v0, v0, v1, -v285; GFX11-NEXT: s_setpc_b64 s[30:31]86 %neg.c = fneg float %c87 %fma = call float @llvm.amdgcn.fma.legacy(float %a, float %b, float %neg.c)88 ret float %fma89}90 91define float @v_fma_const_const(float %a) {92; GFX10-LABEL: v_fma_const_const:93; GFX10: ; %bb.0:94; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)95; GFX10-NEXT: v_fma_legacy_f32 v0, v0, 2.0, -1.096; GFX10-NEXT: s_setpc_b64 s[30:31]97;98; GFX11-LABEL: v_fma_const_const:99; GFX11: ; %bb.0:100; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)101; GFX11-NEXT: v_fma_dx9_zero_f32 v0, v0, 2.0, -1.0102; GFX11-NEXT: s_setpc_b64 s[30:31]103 %fma = call float @llvm.amdgcn.fma.legacy(float %a, float 2.0, float -1.0)104 ret float %fma105}106 107declare float @llvm.amdgcn.fma.legacy(float, float, float)108declare float @llvm.fabs.f32(float)109