[PATCH] D89168: [AMDGPU] Corrected declaration of VOPC instructions with SDWA addressing mode.
Konstantin Pyzhov via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Oct 9 15:56:55 PDT 2020
kpyzhov created this revision.
kpyzhov added a reviewer: kzhuravl.
kpyzhov added a project: AMDGPU.
Herald added subscribers: llvm-commits, kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, arsenm.
Herald added a project: LLVM.
kpyzhov requested review of this revision.
Herald added a subscriber: wdng.
Removed "implicit def VCC" from declarations of AMDGPU VOPC instructions since they do not implicitly write to VCC in SDWA mode.
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D89168
Files:
llvm/lib/Target/AMDGPU/VOPCInstructions.td
llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir
llvm/test/CodeGen/AMDGPU/wave32.ll
Index: llvm/test/CodeGen/AMDGPU/wave32.ll
===================================================================
--- llvm/test/CodeGen/AMDGPU/wave32.ll
+++ llvm/test/CodeGen/AMDGPU/wave32.ll
@@ -44,10 +44,10 @@
}
; GCN-LABEL: {{^}}test_vopc_2xf16:
-; GFX1032: v_cmp_le_f16_sdwa [[SC:s[0-9]+]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
-; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]]
-; GFX1064: v_cmp_le_f16_sdwa [[SC:s\[[0-9:]+\]]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
-; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]]
+; GFX1032: v_cmp_le_f16_sdwa [[SC:vcc_lo|s[0-9]+]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
+; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]]
+; GFX1064: v_cmp_le_f16_sdwa [[SC:vcc|s\[[0-9:]+\]]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
+; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]]
define amdgpu_kernel void @test_vopc_2xf16(<2 x half> addrspace(1)* %arg) {
%lid = tail call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %arg, i32 %lid
Index: llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir
===================================================================
--- llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir
+++ llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir
@@ -325,39 +325,39 @@
# GCN-LABEL: {{^}}name: vopc_instructions
# GFX89: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_e32 123, implicit $exec
-# GFX89: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec
-# GFX89: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# GFX89: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $exec
-# GFX89: $vcc = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $exec
+# GFX89: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $mode, implicit $exec
+# GFX89: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# GFX89: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $exec
+# GFX89: $vcc = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $exec
-# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec
+# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $mode, implicit $exec
# VI: %{{[0-9]+}}:sreg_64 = V_CMPX_GT_F32_e64 0, 23, 0, killed %{{[0-9]+}}, 0, implicit-def $exec, implicit $mode, implicit $exec
-# VI: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %3, 0, 6, 4, implicit-def $vcc, implicit $exec
+# VI: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %3, 0, 6, 4, implicit $exec
# VI: %{{[0-9]+}}:sreg_64 = V_CMPX_EQ_I32_e64 23, killed %{{[0-9]+}}, implicit-def $exec, implicit $exec
-# GFX9: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec
+# GFX9: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $mode, implicit $exec
# GFX9: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_e32 23, implicit $exec
-# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# GFX9: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $exec
+# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# GFX9: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $exec
# GFX9: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_e32 23, implicit $exec
-# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $exec
+# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $exec
-# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec
-# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec
-# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 1, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
+# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit $mode, implicit $exec
+# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit $mode, implicit $exec
+# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 1, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
# GFX9: $vcc = V_CMP_EQ_F32_e64 0, %{{[0-9]+}}, 0, killed %{{[0-9]+}}, 1, implicit $mode, implicit $exec
-# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
+# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
# GFX9: $vcc = V_CMP_EQ_F32_e64 0, %{{[0-9]+}}, 0, killed %{{[0-9]+}}, 1, implicit $mode, implicit $exec
-# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
-# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec
+# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
+# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec
# GFX9: $vcc = V_CMPX_GT_F32_e64 1, 23, 1, killed %{{[0-9]+}}, 1, implicit-def $exec, implicit $mode, implicit $exec
Index: llvm/lib/Target/AMDGPU/VOPCInstructions.td
===================================================================
--- llvm/lib/Target/AMDGPU/VOPCInstructions.td
+++ llvm/lib/Target/AMDGPU/VOPCInstructions.td
@@ -229,7 +229,7 @@
foreach _ = BoolToList<P.HasExtSDWA>.ret in
def _sdwa : VOPC_SDWA_Pseudo <opName, P> {
- let Defs = !if(DefExec, [VCC, EXEC], [VCC]);
+ let Defs = !if(DefExec, [EXEC], []);
let SchedRW = P.Schedule;
let isConvergent = DefExec;
let isCompare = 1;
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D89168.297350.patch
Type: text/x-patch
Size: 8037 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20201009/ef240fde/attachment.bin>
More information about the llvm-commits
mailing list