diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll index a094631267e64..e9e4d5ebed41c 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll @@ -1,17 +1,35 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8 %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-TRUE16 %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-FAKE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GFX8,GFX8-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-TRUE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-FAKE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s + ; FIXME: promotion not handled without f16 insts define half @v_constained_fadd_f16_fpexcept_strict(half %x, half %y) #0 { -; GCN-LABEL: v_constained_fadd_f16_fpexcept_strict: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f16_e32 v0, v0, v1 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f16_fpexcept_strict: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-LABEL: v_constained_fadd_f16_fpexcept_strict: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: v_constained_fadd_f16_fpexcept_strict: ; GFX10: ; %bb.0: @@ -19,27 +37,55 @@ define half @v_constained_fadd_f16_fpexcept_strict(half %x, half %y) #0 { ; GFX10-NEXT: v_add_f16_e32 v0, v0, v1 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict: -; GFX11-TRUE16: ; %bb.0: -; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l -; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict: +; GFX11-SDAG-TRUE16: ; %bb.0: +; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l +; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_strict: -; GFX11-FAKE16: ; %bb.0: -; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1 -; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_strict: +; GFX11-SDAG-FAKE16: ; %bb.0: +; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l +; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_strict: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call half @llvm.experimental.constrained.fadd.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret half %val } define half @v_constained_fadd_f16_fpexcept_ignore(half %x, half %y) #0 { -; GCN-LABEL: v_constained_fadd_f16_fpexcept_ignore: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f16_e32 v0, v0, v1 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f16_fpexcept_ignore: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-LABEL: v_constained_fadd_f16_fpexcept_ignore: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: v_constained_fadd_f16_fpexcept_ignore: ; GFX10: ; %bb.0: @@ -47,12 +93,44 @@ define half @v_constained_fadd_f16_fpexcept_ignore(half %x, half %y) #0 { ; GFX10-NEXT: v_add_f16_e32 v0, v0, v1 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; +; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore: +; GFX11-SDAG-TRUE16: ; %bb.0: +; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l +; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_ignore: +; GFX11-SDAG-FAKE16: ; %bb.0: +; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l +; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_ignore: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f16_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_ignore: ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -63,11 +141,17 @@ define half @v_constained_fadd_f16_fpexcept_ignore(half %x, half %y) #0 { } define half @v_constained_fadd_f16_fpexcept_maytrap(half %x, half %y) #0 { -; GCN-LABEL: v_constained_fadd_f16_fpexcept_maytrap: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f16_e32 v0, v0, v1 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f16_fpexcept_maytrap: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-LABEL: v_constained_fadd_f16_fpexcept_maytrap: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: v_constained_fadd_f16_fpexcept_maytrap: ; GFX10: ; %bb.0: @@ -75,12 +159,44 @@ define half @v_constained_fadd_f16_fpexcept_maytrap(half %x, half %y) #0 { ; GFX10-NEXT: v_add_f16_e32 v0, v0, v1 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; +; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap: +; GFX11-SDAG-TRUE16: ; %bb.0: +; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l +; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap: +; GFX11-SDAG-FAKE16: ; %bb.0: +; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l +; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f16_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap: ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -97,19 +213,43 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_strict(<2 x half> %x, <2 x h ; GFX9-NEXT: v_pk_add_f16 v0, v0, v1 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX8-LABEL: v_constained_fadd_v2f16_fpexcept_strict: -; GFX8: ; %bb.0: -; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_e32 v0, v0, v1 -; GFX8-NEXT: v_or_b32_e32 v0, v0, v2 -; GFX8-NEXT: s_setpc_b64 s[30:31] +; GFX8-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_strict: +; GFX8-SDAG: ; %bb.0: +; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-SDAG-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v2 +; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict: +; GFX8-GISEL: ; %bb.0: +; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v0, v1 +; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-LABEL: v_constained_fadd_v2f16_fpexcept_strict: +; GFX10: ; %bb.0: +; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fadd_v2f16_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX11-NEXT: s_setpc_b64 s[30:31] ; -; GFX10PLUS-LABEL: v_constained_fadd_v2f16_fpexcept_strict: -; GFX10PLUS: ; %bb.0: -; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10PLUS-NEXT: v_pk_add_f16 v0, v0, v1 -; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; GFX12-LABEL: v_constained_fadd_v2f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x half> %val } @@ -121,19 +261,43 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h ; GFX9-NEXT: v_pk_add_f16 v0, v0, v1 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX8-LABEL: v_constained_fadd_v2f16_fpexcept_ignore: -; GFX8: ; %bb.0: -; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_e32 v0, v0, v1 -; GFX8-NEXT: v_or_b32_e32 v0, v0, v2 -; GFX8-NEXT: s_setpc_b64 s[30:31] +; GFX8-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_ignore: +; GFX8-SDAG: ; %bb.0: +; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-SDAG-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v2 +; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore: +; GFX8-GISEL: ; %bb.0: +; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v0, v1 +; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-LABEL: v_constained_fadd_v2f16_fpexcept_ignore: +; GFX10: ; %bb.0: +; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX10-NEXT: s_setpc_b64 s[30:31] ; -; GFX10PLUS-LABEL: v_constained_fadd_v2f16_fpexcept_ignore: -; GFX10PLUS: ; %bb.0: -; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10PLUS-NEXT: v_pk_add_f16 v0, v0, v1 -; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; GFX11-LABEL: v_constained_fadd_v2f16_fpexcept_ignore: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_v2f16_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x half> %val } @@ -145,54 +309,142 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v0, v1 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX8-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap: -; GFX8: ; %bb.0: -; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_e32 v0, v0, v1 -; GFX8-NEXT: v_or_b32_e32 v0, v0, v2 -; GFX8-NEXT: s_setpc_b64 s[30:31] +; GFX8-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap: +; GFX8-SDAG: ; %bb.0: +; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-SDAG-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v1 +; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v2 +; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap: +; GFX8-GISEL: ; %bb.0: +; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v0, v1 +; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31] ; -; GFX10PLUS-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap: -; GFX10PLUS: ; %bb.0: -; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10PLUS-NEXT: v_pk_add_f16 v0, v0, v1 -; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; GFX10-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap: +; GFX10: ; %bb.0: +; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_pk_add_f16 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x half> %val } define <3 x half> @v_constained_fadd_v3f16_fpexcept_strict(<3 x half> %x, <3 x half> %y) #0 { -; GFX9-LABEL: v_constained_fadd_v3f16_fpexcept_strict: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX9-NEXT: v_add_f16_e32 v1, v1, v3 -; GFX9-NEXT: s_setpc_b64 s[30:31] +; GFX9-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-SDAG-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX9-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31] ; -; GFX8-LABEL: v_constained_fadd_v3f16_fpexcept_strict: -; GFX8: ; %bb.0: -; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_e32 v0, v0, v2 -; GFX8-NEXT: v_or_b32_e32 v0, v0, v4 -; GFX8-NEXT: v_add_f16_e32 v1, v1, v3 -; GFX8-NEXT: s_setpc_b64 s[30:31] +; GFX9-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX9-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; -; GFX10-LABEL: v_constained_fadd_v3f16_fpexcept_strict: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX10-NEXT: v_add_f16_e32 v1, v1, v3 -; GFX10-NEXT: s_setpc_b64 s[30:31] +; GFX8-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX8-SDAG: ; %bb.0: +; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-SDAG-NEXT: v_add_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v2 +; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v4 +; GFX8-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX8-GISEL: ; %bb.0: +; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-GISEL-NEXT: v_add_f16_e32 v4, v0, v2 +; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-GISEL-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v4, v0 +; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX10-SDAG: ; %bb.0: +; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX10-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] ; +; GFX10-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX10-GISEL: ; %bb.0: +; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX10-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX11-SDAG-TRUE16: ; %bb.0: +; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l +; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX11-SDAG-FAKE16: ; %bb.0: +; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX12-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX12-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX11-TRUE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 ; GFX11-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-FAKE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict: ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -205,40 +457,127 @@ define <3 x half> @v_constained_fadd_v3f16_fpexcept_strict(<3 x half> %x, <3 x h ; FIXME: Scalarized define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x half> %y) #0 { -; GFX9-LABEL: v_constained_fadd_v4f16_fpexcept_strict: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX9-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX9-NEXT: v_add_f16_e32 v1, v1, v3 -; GFX9-NEXT: v_add_f16_e32 v0, v0, v2 -; GFX9-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-NEXT: v_perm_b32 v0, v5, v0, s4 -; GFX9-NEXT: v_perm_b32 v1, v4, v1, s4 -; GFX9-NEXT: s_setpc_b64 s[30:31] +; GFX9-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-SDAG-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX9-SDAG-NEXT: v_add_f16_e32 v0, v0, v2 +; GFX9-SDAG-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-SDAG-NEXT: v_perm_b32 v0, v5, v0, s4 +; GFX9-SDAG-NEXT: v_perm_b32 v1, v4, v1, s4 +; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31] ; -; GFX8-LABEL: v_constained_fadd_v4f16_fpexcept_strict: -; GFX8: ; %bb.0: -; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_e32 v1, v1, v3 -; GFX8-NEXT: v_add_f16_e32 v0, v0, v2 -; GFX8-NEXT: v_or_b32_e32 v0, v0, v5 -; GFX8-NEXT: v_or_b32_e32 v1, v1, v4 -; GFX8-NEXT: s_setpc_b64 s[30:31] +; GFX9-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX9-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; -; GFX10-LABEL: v_constained_fadd_v4f16_fpexcept_strict: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX10-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX10-NEXT: v_add_f16_e32 v0, v0, v2 -; GFX10-NEXT: v_add_f16_e32 v1, v1, v3 -; GFX10-NEXT: v_perm_b32 v0, v5, v0, 0x5040100 -; GFX10-NEXT: v_perm_b32 v1, v4, v1, 0x5040100 -; GFX10-NEXT: s_setpc_b64 s[30:31] +; GFX8-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX8-SDAG: ; %bb.0: +; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-SDAG-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-SDAG-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v2 +; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v5 +; GFX8-SDAG-NEXT: v_or_b32_e32 v1, v1, v4 +; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX8-GISEL: ; %bb.0: +; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-GISEL-NEXT: v_add_f16_e32 v4, v0, v2 +; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v1, v3 +; GFX8-GISEL-NEXT: v_add_f16_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v4, v0 +; GFX8-GISEL-NEXT: v_or_b32_e32 v1, v2, v1 +; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX10-SDAG: ; %bb.0: +; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-SDAG-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX10-SDAG-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v2 +; GFX10-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX10-SDAG-NEXT: v_perm_b32 v0, v5, v0, 0x5040100 +; GFX10-SDAG-NEXT: v_perm_b32 v1, v4, v1, 0x5040100 +; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] ; +; GFX10-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX10-GISEL: ; %bb.0: +; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX10-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX11-SDAG-TRUE16: ; %bb.0: +; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v1.h, v1.h, v3.h +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.h, v0.h, v2.h +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v2.l +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l +; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX11-SDAG-FAKE16: ; %bb.0: +; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v3 +; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v0 +; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v2 +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v2, v6, v5 +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v3, v7, v4 +; GFX11-SDAG-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX11-SDAG-FAKE16-NEXT: v_perm_b32 v1, v3, v1, 0x5040100 +; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v3 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v6, 16, v0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; GFX12-SDAG-NEXT: v_add_f16_e32 v1, v1, v3 +; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v2 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX12-SDAG-NEXT: v_add_f16_e32 v2, v6, v5 +; GFX12-SDAG-NEXT: v_add_f16_e32 v3, v7, v4 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX12-SDAG-NEXT: v_perm_b32 v1, v3, v1, 0x5040100 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX12-GISEL-NEXT: v_pk_add_f16 v1, v1, v3 +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX11-TRUE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -247,7 +586,6 @@ define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x h ; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v2.l ; GFX11-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-FAKE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict: ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -267,22 +605,53 @@ define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x h } define amdgpu_ps half @s_constained_fadd_f16_fpexcept_strict(half inreg %x, half inreg %y) #0 { -; GCN-LABEL: s_constained_fadd_f16_fpexcept_strict: -; GCN: ; %bb.0: -; GCN-NEXT: v_mov_b32_e32 v0, s3 -; GCN-NEXT: v_add_f16_e32 v0, s2, v0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: s_constained_fadd_f16_fpexcept_strict: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-NEXT: v_add_f16_e32 v0, s2, v0 +; GFX9-NEXT: ; return to shader part epilog +; +; GFX8-LABEL: s_constained_fadd_f16_fpexcept_strict: +; GFX8: ; %bb.0: +; GFX8-NEXT: v_mov_b32_e32 v0, s3 +; GFX8-NEXT: v_add_f16_e32 v0, s2, v0 +; GFX8-NEXT: ; return to shader part epilog ; ; GFX10-LABEL: s_constained_fadd_f16_fpexcept_strict: ; GFX10: ; %bb.0: ; GFX10-NEXT: v_add_f16_e64 v0, s2, s3 ; GFX10-NEXT: ; return to shader part epilog ; +; GFX11-SDAG-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict: +; GFX11-SDAG-TRUE16: ; %bb.0: +; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e64 v0.l, s2, s3 +; GFX11-SDAG-TRUE16-NEXT: ; return to shader part epilog +; +; GFX11-SDAG-FAKE16-LABEL: s_constained_fadd_f16_fpexcept_strict: +; GFX11-SDAG-FAKE16: ; %bb.0: +; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e64 v0, s2, s3 +; GFX11-SDAG-FAKE16-NEXT: ; return to shader part epilog +; +; GFX11-GISEL-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e64 v0.l, s2, s3 +; GFX11-GISEL-TRUE16-NEXT: ; return to shader part epilog +; +; GFX11-GISEL-FAKE16-LABEL: s_constained_fadd_f16_fpexcept_strict: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e64 v0, s2, s3 +; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: s_constained_fadd_f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_add_f16 s0, s2, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog ; GFX11-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: v_add_f16_e64 v0.l, s2, s3 ; GFX11-TRUE16-NEXT: ; return to shader part epilog -; ; GFX11-FAKE16-LABEL: s_constained_fadd_f16_fpexcept_strict: ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: v_add_f16_e64 v0, s2, s3 @@ -298,22 +667,44 @@ define amdgpu_ps <2 x half> @s_constained_fadd_v2f16_fpexcept_strict(<2 x half> ; GFX9-NEXT: v_pk_add_f16 v0, s2, v0 ; GFX9-NEXT: ; return to shader part epilog ; -; GFX8-LABEL: s_constained_fadd_v2f16_fpexcept_strict: -; GFX8: ; %bb.0: -; GFX8-NEXT: s_lshr_b32 s0, s3, 16 -; GFX8-NEXT: s_lshr_b32 s1, s2, 16 -; GFX8-NEXT: v_mov_b32_e32 v0, s0 -; GFX8-NEXT: v_mov_b32_e32 v1, s1 -; GFX8-NEXT: v_add_f16_sdwa v0, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_mov_b32_e32 v1, s3 -; GFX8-NEXT: v_add_f16_e32 v1, s2, v1 -; GFX8-NEXT: v_or_b32_e32 v0, v1, v0 -; GFX8-NEXT: ; return to shader part epilog +; GFX8-SDAG-LABEL: s_constained_fadd_v2f16_fpexcept_strict: +; GFX8-SDAG: ; %bb.0: +; GFX8-SDAG-NEXT: s_lshr_b32 s0, s3, 16 +; GFX8-SDAG-NEXT: s_lshr_b32 s1, s2, 16 +; GFX8-SDAG-NEXT: v_mov_b32_e32 v0, s0 +; GFX8-SDAG-NEXT: v_mov_b32_e32 v1, s1 +; GFX8-SDAG-NEXT: v_add_f16_sdwa v0, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-SDAG-NEXT: v_mov_b32_e32 v1, s3 +; GFX8-SDAG-NEXT: v_add_f16_e32 v1, s2, v1 +; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v1, v0 +; GFX8-SDAG-NEXT: ; return to shader part epilog +; +; GFX8-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict: +; GFX8-GISEL: ; %bb.0: +; GFX8-GISEL-NEXT: s_lshr_b32 s0, s2, 16 +; GFX8-GISEL-NEXT: s_lshr_b32 s1, s3, 16 +; GFX8-GISEL-NEXT: v_mov_b32_e32 v0, s3 +; GFX8-GISEL-NEXT: v_mov_b32_e32 v1, s1 +; GFX8-GISEL-NEXT: v_mov_b32_e32 v2, s0 +; GFX8-GISEL-NEXT: v_add_f16_e32 v0, s2, v0 +; GFX8-GISEL-NEXT: v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v0, v1 +; GFX8-GISEL-NEXT: ; return to shader part epilog +; +; GFX10-LABEL: s_constained_fadd_v2f16_fpexcept_strict: +; GFX10: ; %bb.0: +; GFX10-NEXT: v_pk_add_f16 v0, s2, s3 +; GFX10-NEXT: ; return to shader part epilog +; +; GFX11-LABEL: s_constained_fadd_v2f16_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: v_pk_add_f16 v0, s2, s3 +; GFX11-NEXT: ; return to shader part epilog ; -; GFX10PLUS-LABEL: s_constained_fadd_v2f16_fpexcept_strict: -; GFX10PLUS: ; %bb.0: -; GFX10PLUS-NEXT: v_pk_add_f16 v0, s2, s3 -; GFX10PLUS-NEXT: ; return to shader part epilog +; GFX12-LABEL: s_constained_fadd_v2f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_add_f16 v0, s2, s3 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x half> %val } @@ -325,5 +716,3 @@ declare <4 x half> @llvm.experimental.constrained.fadd.v4f16(<4 x half>, <4 x ha attributes #0 = { strictfp } attributes #1 = { inaccessiblememonly nounwind willreturn } -;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: -; GFX11: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll index 2aecf5fd8753c..a039c2629c395 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll @@ -1,137 +1,383 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10-GISEL %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11-GISEL %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s define float @v_constained_fadd_f32_fpexcept_strict(float %x, float %y) #0 { -; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e32 v0, v0, v1 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f32_fpexcept_strict: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v1 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } define float @v_constained_fadd_f32_fpexcept_ignore(float %x, float %y) #0 { -; GCN-LABEL: v_constained_fadd_f32_fpexcept_ignore: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e32 v0, v0, v1 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f32_fpexcept_ignore: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_ignore: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f32_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_ignore: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v1 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_ignore: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret float %val } define float @v_constained_fadd_f32_fpexcept_maytrap(float %x, float %y) #0 { -; GCN-LABEL: v_constained_fadd_f32_fpexcept_maytrap: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e32 v0, v0, v1 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f32_fpexcept_maytrap: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_maytrap: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f32_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_maytrap: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v1 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_maytrap: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret float %val } define <2 x float> @v_constained_fadd_v2f32_fpexcept_strict(<2 x float> %x, <2 x float> %y) #0 { -; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_strict: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e32 v0, v0, v2 -; GCN-NEXT: v_add_f32_e32 v1, v1, v3 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX9-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GFX10-SDAG: ; %bb.0: +; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GFX10-GISEL: ; %bb.0: +; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GFX11-SDAG: ; %bb.0: +; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_strict: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_f32_e32 v0, v0, v2 ; GFX10-NEXT: v_add_f32_e32 v1, v1, v3 ; GFX10-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_strict: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v2 +; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v3 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2 +; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x float> %val } define <2 x float> @v_constained_fadd_v2f32_fpexcept_ignore(<2 x float> %x, <2 x float> %y) #0 { -; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e32 v0, v0, v2 -; GCN-NEXT: v_add_f32_e32 v1, v1, v3 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX9-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GFX10-SDAG: ; %bb.0: +; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GFX10-GISEL: ; %bb.0: +; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GFX11-SDAG: ; %bb.0: +; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_f32_e32 v0, v0, v2 ; GFX10-NEXT: v_add_f32_e32 v1, v1, v3 ; GFX10-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v2 +; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v3 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2 +; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x float> %val } define <2 x float> @v_constained_fadd_v2f32_fpexcept_maytrap(<2 x float> %x, <2 x float> %y) #0 { -; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e32 v0, v0, v2 -; GCN-NEXT: v_add_f32_e32 v1, v1, v3 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX9-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GFX10-SDAG: ; %bb.0: +; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GFX10-GISEL: ; %bb.0: +; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v2 +; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v3 +; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GFX11-SDAG: ; %bb.0: +; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_f32_e32 v0, v0, v2 ; GFX10-NEXT: v_add_f32_e32 v1, v1, v3 ; GFX10-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v2 +; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v3 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2 +; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x float> %val } define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x float> %y) #0 { -; GCN-LABEL: v_constained_fadd_v3f32_fpexcept_strict: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e32 v0, v0, v3 -; GCN-NEXT: v_add_f32_e32 v1, v1, v4 -; GCN-NEXT: v_add_f32_e32 v2, v2, v5 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e32 v0, v0, v3 +; GFX9-NEXT: v_add_f32_e32 v1, v1, v4 +; GFX9-NEXT: v_add_f32_e32 v2, v2, v5 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GFX10-SDAG: ; %bb.0: +; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v3 +; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v4 +; GFX10-SDAG-NEXT: v_add_f32_e32 v2, v2, v5 +; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GFX10-GISEL: ; %bb.0: +; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v3 +; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v4 +; GFX10-GISEL-NEXT: v_add_f32_e32 v2, v2, v5 +; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GFX11-SDAG: ; %bb.0: +; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4 +; GFX11-SDAG-NEXT: v_add_f32_e32 v2, v2, v5 +; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4 +; GFX11-GISEL-NEXT: v_add_f32_e32 v2, v2, v5 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4 +; GFX12-NEXT: v_add_f32_e32 v2, v2, v5 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX10-LABEL: v_constained_fadd_v3f32_fpexcept_strict: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -139,78 +385,204 @@ define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x ; GFX10-NEXT: v_add_f32_e32 v1, v1, v4 ; GFX10-NEXT: v_add_f32_e32 v2, v2, v5 ; GFX10-NEXT: s_setpc_b64 s[30:31] -; ; GFX11-LABEL: v_constained_fadd_v3f32_fpexcept_strict: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4 ; GFX11-NEXT: v_add_f32_e32 v2, v2, v5 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v3 +; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v4 +; GCN-SDAG-NEXT: v_add_f32_e32 v2, v2, v5 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v3 +; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v4 +; GCN-GISEL-NEXT: v_add_f32_e32 v2, v2, v5 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %val = call <3 x float> @llvm.experimental.constrained.fadd.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <3 x float> %val } define amdgpu_ps float @s_constained_fadd_f32_fpexcept_strict(float inreg %x, float inreg %y) #0 { -; GCN-LABEL: s_constained_fadd_f32_fpexcept_strict: -; GCN: ; %bb.0: -; GCN-NEXT: v_mov_b32_e32 v0, s3 -; GCN-NEXT: v_add_f32_e32 v0, s2, v0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: s_constained_fadd_f32_fpexcept_strict: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-NEXT: v_add_f32_e32 v0, s2, v0 +; GFX9-NEXT: ; return to shader part epilog ; ; GFX10PLUS-LABEL: s_constained_fadd_f32_fpexcept_strict: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: v_add_f32_e64 v0, s2, s3 ; GFX10PLUS-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: s_constained_fadd_f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_add_f32 s0, s2, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog +; GCN-SDAG-LABEL: s_constained_fadd_f32_fpexcept_strict: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: v_mov_b32_e32 v0, s3 +; GCN-SDAG-NEXT: v_add_f32_e32 v0, s2, v0 +; GCN-SDAG-NEXT: ; return to shader part epilog +; GCN-GISEL-LABEL: s_constained_fadd_f32_fpexcept_strict: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: v_mov_b32_e32 v0, s3 +; GCN-GISEL-NEXT: v_add_f32_e32 v0, s2, v0 +; GCN-GISEL-NEXT: ; return to shader part epilog %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } define float @v_constained_fadd_f32_fpexcept_strict_fabs_lhs(float %x, float %y) #0 { -; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e64 v0, |v0|, v1 -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e64 v0, |v0|, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_add_f32_e64 v0, |v0|, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f32_e64 v0, |v0|, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e64 v0, |v0|, v1 +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e64 v0, |v0|, v1 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %fabs.x = call float @llvm.fabs.f32(float %x) #0 %val = call float @llvm.experimental.constrained.fadd.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } define float @v_constained_fadd_f32_fpexcept_strict_fabs_rhs(float %x, float %y) #0 { -; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_add_f32_e64 v0, v0, |v1| -; GCN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_add_f32_e64 v0, v0, |v1| +; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_add_f32_e64 v0, v0, |v1| ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_add_f32_e64 v0, v0, |v1| +; GFX12-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_add_f32_e64 v0, v0, |v1| +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e64 v0, v0, |v1| +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] %fabs.y = call float @llvm.fabs.f32(float %y) #0 %val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } define float @v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs(float %x, float %y) #0 { -; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: -; GCN: ; %bb.0: -; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_sub_f32_e64 v0, v1, |v0| -; GCN-NEXT: s_setpc_b64 s[30:31] +; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0| +; GCN-SDAG-NEXT: s_setpc_b64 s[30:31] ; -; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: -; GFX10PLUS: ; %bb.0: -; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10PLUS-NEXT: v_sub_f32_e64 v0, v1, |v0| -; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1 +; GCN-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX9-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0| +; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX9-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1 +; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX10-SDAG: ; %bb.0: +; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0| +; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX10-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX10-GISEL: ; %bb.0: +; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1 +; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX11-SDAG: ; %bb.0: +; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0| +; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0| +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1 +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] %fabs.x = call float @llvm.fabs.f32(float %x) #0 %neg.fabs.x = fneg float %fabs.x %val = call float @llvm.experimental.constrained.fadd.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll index faa0131c88c2d..5469fc8330971 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll @@ -1,7 +1,12 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s -; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GCN-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GCN-GISEL %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX10 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX10 %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX11 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX11 %s define double @v_constained_fadd_f64_fpexcept_strict(double %x, double %y) #0 { ; GCN-LABEL: v_constained_fadd_f64_fpexcept_strict: @@ -9,12 +14,6 @@ define double @v_constained_fadd_f64_fpexcept_strict(double %x, double %y) #0 { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3] ; GCN-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-LABEL: v_constained_fadd_f64_fpexcept_strict: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3] -; GFX10-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret double %val } @@ -25,12 +24,6 @@ define double @v_constained_fadd_f64_fpexcept_ignore(double %x, double %y) #0 { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3] ; GCN-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-LABEL: v_constained_fadd_f64_fpexcept_ignore: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3] -; GFX10-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret double %val } @@ -41,12 +34,6 @@ define double @v_constained_fadd_f64_fpexcept_maytrap(double %x, double %y) #0 { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3] ; GCN-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-LABEL: v_constained_fadd_f64_fpexcept_maytrap: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3] -; GFX10-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret double %val } @@ -58,13 +45,6 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_strict(<2 x double> %x, <2 ; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5] ; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7] ; GCN-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_strict: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5] -; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7] -; GFX10-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x double> %val } @@ -76,13 +56,6 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_ignore(<2 x double> %x, <2 ; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5] ; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7] ; GCN-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_ignore: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5] -; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7] -; GFX10-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x double> %val } @@ -94,13 +67,6 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_maytrap(<2 x double> %x, < ; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5] ; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7] ; GCN-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_maytrap: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5] -; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7] -; GFX10-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x double> %val } @@ -113,30 +79,29 @@ define <3 x double> @v_constained_fadd_v3f64_fpexcept_strict(<3 x double> %x, <3 ; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[8:9] ; GCN-NEXT: v_add_f64 v[4:5], v[4:5], v[10:11] ; GCN-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-LABEL: v_constained_fadd_v3f64_fpexcept_strict: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[6:7] -; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[8:9] -; GFX10-NEXT: v_add_f64 v[4:5], v[4:5], v[10:11] -; GFX10-NEXT: s_setpc_b64 s[30:31] %val = call <3 x double> @llvm.experimental.constrained.fadd.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <3 x double> %val } define amdgpu_ps <2 x float> @s_constained_fadd_f64_fpexcept_strict(double inreg %x, double inreg %y) #0 { -; GCN-LABEL: s_constained_fadd_f64_fpexcept_strict: -; GCN: ; %bb.0: -; GCN-NEXT: v_mov_b32_e32 v0, s4 -; GCN-NEXT: v_mov_b32_e32 v1, s5 -; GCN-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1] -; GCN-NEXT: ; return to shader part epilog +; GCN-SDAG-LABEL: s_constained_fadd_f64_fpexcept_strict: +; GCN-SDAG: ; %bb.0: +; GCN-SDAG-NEXT: v_mov_b32_e32 v0, s4 +; GCN-SDAG-NEXT: v_mov_b32_e32 v1, s5 +; GCN-SDAG-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1] +; GCN-SDAG-NEXT: ; return to shader part epilog +; +; GCN-GISEL-LABEL: s_constained_fadd_f64_fpexcept_strict: +; GCN-GISEL: ; %bb.0: +; GCN-GISEL-NEXT: v_mov_b32_e32 v0, s4 +; GCN-GISEL-NEXT: v_mov_b32_e32 v1, s5 +; GCN-GISEL-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1] +; GCN-GISEL-NEXT: ; return to shader part epilog ; -; GFX10-LABEL: s_constained_fadd_f64_fpexcept_strict: -; GFX10: ; %bb.0: -; GFX10-NEXT: v_add_f64 v[0:1], s[2:3], s[4:5] -; GFX10-NEXT: ; return to shader part epilog +; GFX10PLUS-LABEL: s_constained_fadd_f64_fpexcept_strict: +; GFX10PLUS: ; %bb.0: +; GFX10PLUS-NEXT: v_add_f64 v[0:1], s[2:3], s[4:5] +; GFX10PLUS-NEXT: ; return to shader part epilog %val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict") %cast = bitcast double %val to <2 x float> ret <2 x float> %cast @@ -148,3 +113,6 @@ declare <3 x double> @llvm.experimental.constrained.fadd.v3f64(<3 x double>, <3 attributes #0 = { strictfp } attributes #1 = { inaccessiblememonly nounwind willreturn } +;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: +; GFX10: {{.*}} +; GFX11: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll index eed5f016aa787..79154d0db16ec 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll @@ -10,9 +10,11 @@ ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX1-GISEL,GFX1-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX1-GISEL,GFX1-GISEL-FAKE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s ; FIXME: promotion not handled without f16 insts @@ -41,12 +43,32 @@ define half @v_constained_fmul_f16_fpexcept_strict(half %x, half %y) #0 { ; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1 ; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX11-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_strict: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l +; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_strict: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1 +; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX1-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_strict: ; GFX1-GISEL-TRUE16: ; %bb.0: ; GFX1-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l ; GFX1-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] -; ; GFX1-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_strict: ; GFX1-GISEL-FAKE16: ; %bb.0: ; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -81,12 +103,32 @@ define half @v_constained_fmul_f16_fpexcept_ignore(half %x, half %y) #0 { ; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1 ; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX11-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_ignore: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l +; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_ignore: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1 +; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f16_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX1-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_ignore: ; GFX1-GISEL-TRUE16: ; %bb.0: ; GFX1-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l ; GFX1-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] -; ; GFX1-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_ignore: ; GFX1-GISEL-FAKE16: ; %bb.0: ; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -121,12 +163,32 @@ define half @v_constained_fmul_f16_fpexcept_maytrap(half %x, half %y) #0 { ; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1 ; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX11-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l +; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1 +; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f16_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] ; GFX1-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap: ; GFX1-GISEL-TRUE16: ; %bb.0: ; GFX1-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l ; GFX1-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31] -; ; GFX1-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap: ; GFX1-GISEL-FAKE16: ; %bb.0: ; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -164,6 +226,16 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_strict(<2 x half> %x, <2 x h ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_v2f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_pk_mul_f16 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x half> %val } @@ -196,6 +268,16 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_v2f16_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_pk_mul_f16 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x half> %val } @@ -228,6 +310,16 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_v2f16_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_pk_mul_f16 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x half> %val } @@ -293,6 +385,34 @@ define <3 x half> @v_constained_fmul_v3f16_fpexcept_strict(<3 x half> %x, <3 x h ; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v1, v1, v3 ; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX11-GISEL-LABEL: v_constained_fmul_v3f16_fpexcept_strict: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2 +; GFX11-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fmul_v3f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v2 +; GFX12-SDAG-NEXT: v_mul_f16_e32 v1, v1, v3 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fmul_v3f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2 +; GFX12-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3 +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX1-GISEL-LABEL: v_constained_fmul_v3f16_fpexcept_strict: ; GFX1-GISEL: ; %bb.0: ; GFX1-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -388,6 +508,44 @@ define <4 x half> @v_constained_fmul_v4f16_fpexcept_strict(<4 x half> %x, <4 x h ; GFX11-SDAG-FAKE16-NEXT: v_perm_b32 v1, v3, v1, 0x5040100 ; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX11-GISEL-LABEL: v_constained_fmul_v4f16_fpexcept_strict: +; GFX11-GISEL: ; %bb.0: +; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2 +; GFX11-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3 +; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fmul_v4f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v3 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v6, 16, v0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; GFX12-SDAG-NEXT: v_mul_f16_e32 v1, v1, v3 +; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v2 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX12-SDAG-NEXT: v_mul_f16_e32 v2, v6, v5 +; GFX12-SDAG-NEXT: v_mul_f16_e32 v3, v7, v4 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX12-SDAG-NEXT: v_perm_b32 v1, v3, v1, 0x5040100 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fmul_v4f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2 +; GFX12-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3 +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX1-GISEL-LABEL: v_constained_fmul_v4f16_fpexcept_strict: ; GFX1-GISEL: ; %bb.0: ; GFX1-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -420,11 +578,26 @@ define amdgpu_ps half @s_constained_fmul_f16_fpexcept_strict(half inreg %x, half ; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e64 v0, s2, s3 ; GFX11-SDAG-FAKE16-NEXT: ; return to shader part epilog ; +; GFX11-GISEL-TRUE16-LABEL: s_constained_fmul_f16_fpexcept_strict: +; GFX11-GISEL-TRUE16: ; %bb.0: +; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e64 v0.l, s2, s3 +; GFX11-GISEL-TRUE16-NEXT: ; return to shader part epilog +; +; GFX11-GISEL-FAKE16-LABEL: s_constained_fmul_f16_fpexcept_strict: +; GFX11-GISEL-FAKE16: ; %bb.0: +; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e64 v0, s2, s3 +; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: s_constained_fmul_f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_mul_f16 s0, s2, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog ; GFX1-GISEL-TRUE16-LABEL: s_constained_fmul_f16_fpexcept_strict: ; GFX1-GISEL-TRUE16: ; %bb.0: ; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e64 v0.l, s2, s3 ; GFX1-GISEL-TRUE16-NEXT: ; return to shader part epilog -; ; GFX1-GISEL-FAKE16-LABEL: s_constained_fmul_f16_fpexcept_strict: ; GFX1-GISEL-FAKE16: ; %bb.0: ; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e64 v0, s2, s3 @@ -468,6 +641,11 @@ define amdgpu_ps <2 x half> @s_constained_fmul_v2f16_fpexcept_strict(<2 x half> ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: v_pk_mul_f16 v0, s2, s3 ; GFX10PLUS-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: s_constained_fmul_v2f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_mul_f16 v0, s2, s3 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x half> %val } diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll index 8df2834928395..4c1df046a6684 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll @@ -1,11 +1,16 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s + ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s + ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s + define float @v_constained_fmul_f32_fpexcept_strict(float %x, float %y) #0 { ; GCN-LABEL: v_constained_fmul_f32_fpexcept_strict: ; GCN: ; %bb.0: @@ -18,6 +23,16 @@ define float @v_constained_fmul_f32_fpexcept_strict(float %x, float %y) #0 { ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } @@ -34,6 +49,16 @@ define float @v_constained_fmul_f32_fpexcept_ignore(float %x, float %y) #0 { ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f32_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret float %val } @@ -50,6 +75,16 @@ define float @v_constained_fmul_f32_fpexcept_maytrap(float %x, float %y) #0 { ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f32_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret float %val } @@ -74,6 +109,16 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_strict(<2 x float> %x, <2 x ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_v2f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x float> %val } @@ -98,6 +143,16 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_ignore(<2 x float> %x, <2 x ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_v2f32_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x float> %val } @@ -122,6 +177,16 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_maytrap(<2 x float> %x, <2 ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_v2f32_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x float> %val } @@ -149,6 +214,17 @@ define <3 x float> @v_constained_fmul_v3f32_fpexcept_strict(<3 x float> %x, <3 x ; GFX11-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4 ; GFX11-NEXT: v_mul_f32_e32 v2, v2, v5 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_v3f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4 +; GFX12-NEXT: v_mul_f32_e32 v2, v2, v5 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <3 x float> @llvm.experimental.constrained.fmul.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <3 x float> %val } @@ -164,6 +240,13 @@ define amdgpu_ps float @s_constained_fmul_f32_fpexcept_strict(float inreg %x, fl ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: v_mul_f32_e64 v0, s2, s3 ; GFX10PLUS-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: s_constained_fmul_f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_mul_f32 s0, s2, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } @@ -180,6 +263,16 @@ define float @v_constained_fmul_f32_fpexcept_strict_fabs_lhs(float %x, float %y) ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_mul_f32_e64 v0, |v0|, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_lhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f32_e64 v0, |v0|, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %fabs.x = call float @llvm.fabs.f32(float %x) #0 %val = call float @llvm.experimental.constrained.fmul.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val @@ -197,6 +290,16 @@ define float @v_constained_fmul_f32_fpexcept_strict_fabs_rhs(float %x, float %y) ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_mul_f32_e64 v0, v0, |v1| ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_rhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f32_e64 v0, v0, |v1| +; GFX12-NEXT: s_setpc_b64 s[30:31] %fabs.y = call float @llvm.fabs.f32(float %y) #0 %val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val @@ -214,6 +317,16 @@ define float @v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs(float %x, floa ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_mul_f32_e64 v0, -|v0|, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_mul_f32_e64 v0, -|v0|, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %fabs.x = call float @llvm.fabs.f32(float %x) #0 %neg.fabs.x = fneg float %fabs.x %val = call float @llvm.experimental.constrained.fmul.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") @@ -226,3 +339,6 @@ declare <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float>, <2 x declare <3 x float> @llvm.experimental.constrained.fmul.v3f32(<3 x float>, <3 x float>, metadata, metadata) attributes #0 = { strictfp } +;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: +; GFX12-GISEL: {{.*}} +; GFX12-SDAG: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll index 8c98a662c59cc..4d2a93397e0c3 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll @@ -1,10 +1,12 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s + ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s -; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s define double @v_constained_fmul_f64_fpexcept_strict(double %x, double %y) #0 { ; GCN-LABEL: v_constained_fmul_f64_fpexcept_strict: @@ -18,6 +20,12 @@ define double @v_constained_fmul_f64_fpexcept_strict(double %x, double %y) #0 { ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fmul_f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret double %val } @@ -34,6 +42,12 @@ define double @v_constained_fmul_f64_fpexcept_ignore(double %x, double %y) #0 { ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fmul_f64_fpexcept_ignore: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret double %val } @@ -50,6 +64,12 @@ define double @v_constained_fmul_f64_fpexcept_maytrap(double %x, double %y) #0 { ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fmul_f64_fpexcept_maytrap: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret double %val } @@ -68,6 +88,13 @@ define <2 x double> @v_constained_fmul_v2f64_fpexcept_strict(<2 x double> %x, <2 ; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5] ; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fmul_v2f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5] +; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x double> %val } @@ -86,6 +113,13 @@ define <2 x double> @v_constained_fmul_v2f64_fpexcept_ignore(<2 x double> %x, <2 ; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5] ; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fmul_v2f64_fpexcept_ignore: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5] +; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x double> %val } @@ -104,6 +138,13 @@ define <2 x double> @v_constained_fmul_v2f64_fpexcept_maytrap(<2 x double> %x, < ; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5] ; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fmul_v2f64_fpexcept_maytrap: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5] +; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x double> %val } @@ -124,6 +165,14 @@ define <3 x double> @v_constained_fmul_v3f64_fpexcept_strict(<3 x double> %x, <3 ; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[8:9] ; GFX10-NEXT: v_mul_f64 v[4:5], v[4:5], v[10:11] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fmul_v3f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[6:7] +; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[8:9] +; GFX11-NEXT: v_mul_f64 v[4:5], v[4:5], v[10:11] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <3 x double> @llvm.experimental.constrained.fmul.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <3 x double> %val } @@ -140,6 +189,11 @@ define amdgpu_ps <2 x float> @s_constained_fmul_f64_fpexcept_strict(double inreg ; GFX10: ; %bb.0: ; GFX10-NEXT: v_mul_f64 v[0:1], s[2:3], s[4:5] ; GFX10-NEXT: ; return to shader part epilog +; +; GFX11-LABEL: s_constained_fmul_f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: v_mul_f64 v[0:1], s[2:3], s[4:5] +; GFX11-NEXT: ; return to shader part epilog %val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict") %cast = bitcast double %val to <2 x float> ret <2 x float> %cast diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll index 6daea572f58c6..45cc77486b509 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll @@ -5,13 +5,16 @@ ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-SDAG %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s -; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-GISEL %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s -; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-TRUE16 %s -; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-FAKE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-FAKE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-TRUE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-FAKE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s ; FIXME: promotion not handled without f16 insts @@ -22,17 +25,11 @@ define half @v_constained_fsub_f16_fpexcept_strict(half %x, half %y) #0 { ; GCN-NEXT: v_sub_f16_e32 v0, v0, v1 ; GCN-NEXT: s_setpc_b64 s[30:31] ; -; GFX10-SDAG-LABEL: v_constained_fsub_f16_fpexcept_strict: -; GFX10-SDAG: ; %bb.0: -; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1 -; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-GISEL-LABEL: v_constained_fsub_f16_fpexcept_strict: -; GFX10-GISEL: ; %bb.0: -; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1 -; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; GFX10-LABEL: v_constained_fsub_f16_fpexcept_strict: +; GFX10: ; %bb.0: +; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-SDAG-TRUE16-LABEL: v_constained_fsub_f16_fpexcept_strict: ; GFX11-SDAG-TRUE16: ; %bb.0: @@ -57,6 +54,16 @@ define half @v_constained_fsub_f16_fpexcept_strict(half %x, half %y) #0 { ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1 ; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret half %val } @@ -68,17 +75,11 @@ define half @v_constained_fsub_f16_fpexcept_ignore(half %x, half %y) #0 { ; GCN-NEXT: v_sub_f16_e32 v0, v0, v1 ; GCN-NEXT: s_setpc_b64 s[30:31] ; -; GFX10-SDAG-LABEL: v_constained_fsub_f16_fpexcept_ignore: -; GFX10-SDAG: ; %bb.0: -; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1 -; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-GISEL-LABEL: v_constained_fsub_f16_fpexcept_ignore: -; GFX10-GISEL: ; %bb.0: -; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1 -; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; GFX10-LABEL: v_constained_fsub_f16_fpexcept_ignore: +; GFX10: ; %bb.0: +; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-SDAG-TRUE16-LABEL: v_constained_fsub_f16_fpexcept_ignore: ; GFX11-SDAG-TRUE16: ; %bb.0: @@ -103,6 +104,16 @@ define half @v_constained_fsub_f16_fpexcept_ignore(half %x, half %y) #0 { ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1 ; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f16_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret half %val } @@ -114,17 +125,11 @@ define half @v_constained_fsub_f16_fpexcept_maytrap(half %x, half %y) #0 { ; GCN-NEXT: v_sub_f16_e32 v0, v0, v1 ; GCN-NEXT: s_setpc_b64 s[30:31] ; -; GFX10-SDAG-LABEL: v_constained_fsub_f16_fpexcept_maytrap: -; GFX10-SDAG: ; %bb.0: -; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1 -; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31] -; -; GFX10-GISEL-LABEL: v_constained_fsub_f16_fpexcept_maytrap: -; GFX10-GISEL: ; %bb.0: -; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1 -; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] +; GFX10-LABEL: v_constained_fsub_f16_fpexcept_maytrap: +; GFX10: ; %bb.0: +; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-SDAG-TRUE16-LABEL: v_constained_fsub_f16_fpexcept_maytrap: ; GFX11-SDAG-TRUE16: ; %bb.0: @@ -149,6 +154,16 @@ define half @v_constained_fsub_f16_fpexcept_maytrap(half %x, half %y) #0 { ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1 ; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f16_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret half %val } @@ -222,6 +237,31 @@ define <2 x half> @v_constained_fsub_v2f16_fpexcept_strict(<2 x half> %x, <2 x h ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v3, v2 +; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fsub_v2f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_strict: ; GFX10PLUS-SDAG: ; %bb.0: ; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -309,6 +349,31 @@ define <2 x half> @v_constained_fsub_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_ignore: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v3, v2 +; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fsub_v2f16_fpexcept_ignore: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_ignore: ; GFX10PLUS-SDAG: ; %bb.0: ; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -396,6 +461,31 @@ define <2 x half> @v_constained_fsub_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v3, v2 +; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap: ; GFX10PLUS-SDAG: ; %bb.0: ; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -509,6 +599,40 @@ define <3 x half> @v_constained_fsub_v3f16_fpexcept_strict(<3 x half> %x, <3 x h ; GFX11-GISEL-FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 ; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v0, v2, 16, v0 ; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fsub_v3f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v2 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v1, v1, v3 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v5, v4 +; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fsub_v3f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v4, 16, v0 +; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v2 +; GFX12-GISEL-NEXT: v_sub_f16_e32 v1, v1, v3 +; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX12-GISEL-NEXT: v_sub_f16_e32 v2, v4, v5 +; GFX12-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v3f16_fpexcept_strict: ; GFX10PLUS-SDAG: ; %bb.0: ; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -654,6 +778,51 @@ define <4 x half> @v_constained_fsub_v4f16_fpexcept_strict(<4 x half> %x, <4 x h ; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v0, v2, 16, v0 ; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v1, v3, 16, v1 ; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-SDAG-LABEL: v_constained_fsub_v4f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-SDAG-NEXT: s_wait_expcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0 +; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0 +; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v3 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v6, 16, v0 +; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v1, v1, v3 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v2 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v6, v5 +; GFX12-SDAG-NEXT: v_sub_f16_e32 v3, v7, v4 +; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX12-SDAG-NEXT: v_perm_b32 v1, v3, v1, 0x5040100 +; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-GISEL-LABEL: v_constained_fsub_v4f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-GISEL-NEXT: s_wait_expcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 +; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 +; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 +; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v4, 16, v0 +; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v1 +; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v7, 16, v3 +; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v2 +; GFX12-GISEL-NEXT: v_sub_f16_e32 v1, v1, v3 +; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX12-GISEL-NEXT: v_sub_f16_e32 v2, v4, v6 +; GFX12-GISEL-NEXT: v_sub_f16_e32 v3, v5, v7 +; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX12-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX12-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX12-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX12-GISEL-NEXT: v_lshl_or_b32 v1, v3, 16, v1 +; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] ; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v4f16_fpexcept_strict: ; GFX10PLUS-SDAG: ; %bb.0: ; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) @@ -695,15 +864,10 @@ define amdgpu_ps half @s_constained_fsub_f16_fpexcept_strict(half inreg %x, half ; GCN-NEXT: v_sub_f16_e32 v0, s2, v0 ; GCN-NEXT: ; return to shader part epilog ; -; GFX10-SDAG-LABEL: s_constained_fsub_f16_fpexcept_strict: -; GFX10-SDAG: ; %bb.0: -; GFX10-SDAG-NEXT: v_sub_f16_e64 v0, s2, s3 -; GFX10-SDAG-NEXT: ; return to shader part epilog -; -; GFX10-GISEL-LABEL: s_constained_fsub_f16_fpexcept_strict: -; GFX10-GISEL: ; %bb.0: -; GFX10-GISEL-NEXT: v_sub_f16_e64 v0, s2, s3 -; GFX10-GISEL-NEXT: ; return to shader part epilog +; GFX10-LABEL: s_constained_fsub_f16_fpexcept_strict: +; GFX10: ; %bb.0: +; GFX10-NEXT: v_sub_f16_e64 v0, s2, s3 +; GFX10-NEXT: ; return to shader part epilog ; ; GFX11-SDAG-TRUE16-LABEL: s_constained_fsub_f16_fpexcept_strict: ; GFX11-SDAG-TRUE16: ; %bb.0: @@ -724,6 +888,13 @@ define amdgpu_ps half @s_constained_fsub_f16_fpexcept_strict(half inreg %x, half ; GFX11-GISEL-FAKE16: ; %bb.0: ; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e64 v0, s2, s3 ; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: s_constained_fsub_f16_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_sub_f16 s0, s2, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret half %val } @@ -808,6 +979,23 @@ define amdgpu_ps <2 x half> @s_constained_fsub_v2f16_fpexcept_strict(<2 x half> ; GFX11-GISEL: ; %bb.0: ; GFX11-GISEL-NEXT: v_pk_add_f16 v0, s2, s3 neg_lo:[0,1] neg_hi:[0,1] ; GFX11-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-SDAG-LABEL: s_constained_fsub_v2f16_fpexcept_strict: +; GFX12-SDAG: ; %bb.0: +; GFX12-SDAG-NEXT: s_lshr_b32 s0, s3, 16 +; GFX12-SDAG-NEXT: s_lshr_b32 s1, s2, 16 +; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_3) +; GFX12-SDAG-NEXT: s_sub_f16 s0, s1, s0 +; GFX12-SDAG-NEXT: s_sub_f16 s1, s2, s3 +; GFX12-SDAG-NEXT: s_pack_ll_b32_b16 s0, s1, s0 +; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-SDAG-NEXT: ; return to shader part epilog +; +; GFX12-GISEL-LABEL: s_constained_fsub_v2f16_fpexcept_strict: +; GFX12-GISEL: ; %bb.0: +; GFX12-GISEL-NEXT: v_pk_add_f16 v0, s2, s3 neg_lo:[0,1] neg_hi:[0,1] +; GFX12-GISEL-NEXT: ; return to shader part epilog ; GFX10PLUS-SDAG-LABEL: s_constained_fsub_v2f16_fpexcept_strict: ; GFX10PLUS-SDAG: ; %bb.0: ; GFX10PLUS-SDAG-NEXT: v_sub_f16_e64 v0, s2, s3 @@ -833,5 +1021,6 @@ declare <4 x half> @llvm.experimental.constrained.fsub.v4f16(<4 x half>, <4 x ha attributes #0 = { strictfp } attributes #1 = { inaccessiblememonly nounwind willreturn } ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: +; GFX11: {{.*}} ; GFX8: {{.*}} ; GFX9: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll index 23dbe21379f7f..3e6db4d4ac6c8 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll @@ -8,6 +8,9 @@ ; RUN: llc -global-isel= -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12 %s + define float @v_constained_fsub_f32_fpexcept_strict(float %x, float %y) #0 { ; GCN-LABEL: v_constained_fsub_f32_fpexcept_strict: ; GCN: ; %bb.0: @@ -20,6 +23,16 @@ define float @v_constained_fsub_f32_fpexcept_strict(float %x, float %y) #0 { ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } @@ -36,6 +49,16 @@ define float @v_constained_fsub_f32_fpexcept_ignore(float %x, float %y) #0 { ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f32_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret float %val } @@ -52,6 +75,16 @@ define float @v_constained_fsub_f32_fpexcept_maytrap(float %x, float %y) #0 { ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f32_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret float %val } @@ -76,6 +109,16 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_strict(<2 x float> %x, <2 x ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_v2f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x float> %val } @@ -100,6 +143,16 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_ignore(<2 x float> %x, <2 x ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_v2f32_fpexcept_ignore: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x float> %val } @@ -124,6 +177,16 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_maytrap(<2 x float> %x, <2 ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_v2f32_fpexcept_maytrap: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x float> %val } @@ -151,6 +214,17 @@ define <3 x float> @v_constained_fsub_v3f32_fpexcept_strict(<3 x float> %x, <3 x ; GFX11-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4 ; GFX11-NEXT: v_sub_f32_e32 v2, v2, v5 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_v3f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4 +; GFX12-NEXT: v_sub_f32_e32 v2, v2, v5 +; GFX12-NEXT: s_setpc_b64 s[30:31] %val = call <3 x float> @llvm.experimental.constrained.fsub.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <3 x float> %val } @@ -166,6 +240,13 @@ define amdgpu_ps float @s_constained_fsub_f32_fpexcept_strict(float inreg %x, fl ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: v_sub_f32_e64 v0, s2, s3 ; GFX10PLUS-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: s_constained_fsub_f32_fpexcept_strict: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_sub_f32 s0, s2, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val } @@ -182,6 +263,16 @@ define float @v_constained_fsub_f32_fpexcept_strict_fabs_lhs(float %x, float %y) ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_sub_f32_e64 v0, |v0|, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_lhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f32_e64 v0, |v0|, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %fabs.x = call float @llvm.fabs.f32(float %x) #0 %val = call float @llvm.experimental.constrained.fsub.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val @@ -199,6 +290,16 @@ define float @v_constained_fsub_f32_fpexcept_strict_fabs_rhs(float %x, float %y) ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_sub_f32_e64 v0, v0, |v1| ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_rhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f32_e64 v0, v0, |v1| +; GFX12-NEXT: s_setpc_b64 s[30:31] %fabs.y = call float @llvm.fabs.f32(float %y) #0 %val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret float %val @@ -216,6 +317,16 @@ define float @v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs(float %x, floa ; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10PLUS-NEXT: v_sub_f32_e64 v0, -|v0|, v1 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_sub_f32_e64 v0, -|v0|, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] %fabs.x = call float @llvm.fabs.f32(float %x) #0 %neg.fabs.x = fneg float %fabs.x %val = call float @llvm.experimental.constrained.fsub.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") @@ -228,3 +339,6 @@ declare <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float>, <2 x declare <3 x float> @llvm.experimental.constrained.fsub.v3f32(<3 x float>, <3 x float>, metadata, metadata) attributes #0 = { strictfp } +;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: +; GFX12-GISEL: {{.*}} +; GFX12-SDAG: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll index e7d136c377079..2937065b7970a 100644 --- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll +++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll @@ -1,10 +1,12 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s + ; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s ; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s -; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s + +; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s define double @v_constained_fsub_f64_fpexcept_strict(double %x, double %y) #0 { ; GCN-LABEL: v_constained_fsub_f64_fpexcept_strict: @@ -18,6 +20,12 @@ define double @v_constained_fsub_f64_fpexcept_strict(double %x, double %y) #0 { ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fsub_f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret double %val } @@ -34,6 +42,12 @@ define double @v_constained_fsub_f64_fpexcept_ignore(double %x, double %y) #0 { ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fsub_f64_fpexcept_ignore: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret double %val } @@ -50,6 +64,12 @@ define double @v_constained_fsub_f64_fpexcept_maytrap(double %x, double %y) #0 { ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fsub_f64_fpexcept_maytrap: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret double %val } @@ -68,6 +88,13 @@ define <2 x double> @v_constained_fsub_v2f64_fpexcept_strict(<2 x double> %x, <2 ; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5] ; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fsub_v2f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5] +; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fsub.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <2 x double> %val } @@ -86,6 +113,13 @@ define <2 x double> @v_constained_fsub_v2f64_fpexcept_ignore(<2 x double> %x, <2 ; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5] ; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fsub_v2f64_fpexcept_ignore: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5] +; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fsub.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore") ret <2 x double> %val } @@ -104,6 +138,13 @@ define <2 x double> @v_constained_fsub_v2f64_fpexcept_maytrap(<2 x double> %x, < ; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5] ; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fsub_v2f64_fpexcept_maytrap: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5] +; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <2 x double> @llvm.experimental.constrained.fsub.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap") ret <2 x double> %val } @@ -124,6 +165,14 @@ define <3 x double> @v_constained_fsub_v3f64_fpexcept_strict(<3 x double> %x, <3 ; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[8:9] ; GFX10-NEXT: v_add_f64 v[4:5], v[4:5], -v[10:11] ; GFX10-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: v_constained_fsub_v3f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[6:7] +; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[8:9] +; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], -v[10:11] +; GFX11-NEXT: s_setpc_b64 s[30:31] %val = call <3 x double> @llvm.experimental.constrained.fsub.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict") ret <3 x double> %val } @@ -140,6 +189,11 @@ define amdgpu_ps <2 x float> @s_constained_fsub_f64_fpexcept_strict(double inreg ; GFX10: ; %bb.0: ; GFX10-NEXT: v_add_f64 v[0:1], s[2:3], -s[4:5] ; GFX10-NEXT: ; return to shader part epilog +; +; GFX11-LABEL: s_constained_fsub_f64_fpexcept_strict: +; GFX11: ; %bb.0: +; GFX11-NEXT: v_add_f64 v[0:1], s[2:3], -s[4:5] +; GFX11-NEXT: ; return to shader part epilog %val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict") %cast = bitcast double %val to <2 x float> ret <2 x float> %cast