[llvm] [AMDGPU] Use different values for SISrcMods::NEG and SISrcMods::SEXT (PR #147964)
Frederik Harwath via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 10 06:42:27 PDT 2025
https://github.com/frederik-h created https://github.com/llvm/llvm-project/pull/147964
The SISrcMods::NEG and SISrcMods::SEXT enumerators share the same enum value. At the time when they were introduced, it was assumed that the "floating point" "neg"/"abs" and the "integer" "sext" source modifiers are mutually exclusive. This can lead to miscompilation as a "sext" modifier may right now be accepted erroneously on some instructions which are encoded as floating point instructions (see the test case modified by this PR). The encoding will then use the "neg" modifier. Furthermore, the "neg"/"abs" and the "sext" modifiers are not necessarily mutually exclusive, i.e. the hardware may support both. This cannot be handled correctly with the current representation.
This patch changes the SISrcMods enum to use different values for NEG and SEXT. This is meant as a first step to allow their coexistence on the same instruction.
>From 84b5c7e5da3ff282b2e5757e8b0d42c18135338f Mon Sep 17 00:00:00 2001
From: Frederik Harwath <fharwath at amd.com>
Date: Wed, 4 Jun 2025 11:54:46 -0400
Subject: [PATCH] [AMDGPU] Use different values for SISrcMods::Neg and ::Sext
The SISrcMods::Neg and SISrcMods::Sext enumerators share the same enum
value. At the time when they were introduced, it was assumed that the
"floating point" "neg" and "abs" and the "integer" "sext" source
modifiers are mutually exclusive. This can lead to miscompilation as a
"sext" modifier may right now be accepted erroneously on some
instructions which are encoded as floating point instructions (see the
test case modified by this PR). The encoding will then use the "neg"
modifier. Furthermore, the "neg"/"abs" and the "sext" modifiers are
not necessarily mutually exclusive, i.e. the hardware may support
both. This cannot be handled correctly with the current
representation.
This patch changes the SISrcMods enum to use different values for
them. This is meant as a first step to allow their coexistence on the
same instruction.
---
llvm/lib/Target/AMDGPU/SIDefines.h | 2 +-
llvm/lib/Target/AMDGPU/VOPCInstructions.td | 4 +--
llvm/lib/Target/AMDGPU/VOPInstructions.td | 26 +++++++--------
.../sdwa-peephole-instr-combine-sel-src.mir | 32 +++++++++----------
4 files changed, 32 insertions(+), 32 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/SIDefines.h b/llvm/lib/Target/AMDGPU/SIDefines.h
index baf74dbdde20e..6d4f812829ebe 100644
--- a/llvm/lib/Target/AMDGPU/SIDefines.h
+++ b/llvm/lib/Target/AMDGPU/SIDefines.h
@@ -266,7 +266,7 @@ namespace SISrcMods {
NONE = 0,
NEG = 1 << 0, // Floating-point negate modifier
ABS = 1 << 1, // Floating-point absolute modifier
- SEXT = 1 << 0, // Integer sign-extend modifier
+ SEXT = 1 << 4, // Integer sign-extend modifier
NEG_HI = ABS, // Floating-point negate high packed component modifier.
OP_SEL_0 = 1 << 2,
OP_SEL_1 = 1 << 3,
diff --git a/llvm/lib/Target/AMDGPU/VOPCInstructions.td b/llvm/lib/Target/AMDGPU/VOPCInstructions.td
index ca5ed5cd24603..26aaec5ae7d4b 100644
--- a/llvm/lib/Target/AMDGPU/VOPCInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOPCInstructions.td
@@ -1450,9 +1450,9 @@ class VOPC_DPP_Base<bits<8> op, string OpName, VOPProfile P>
VOPC_DPPe_Common<op> {
Instruction Opcode = !cast<Instruction>(NAME);
- bits<2> src0_modifiers;
+ bits<5> src0_modifiers;
bits<8> src0;
- bits<2> src1_modifiers;
+ bits<5> src1_modifiers;
bits<9> dpp_ctrl;
bits<1> bound_ctrl;
bits<4> bank_mask;
diff --git a/llvm/lib/Target/AMDGPU/VOPInstructions.td b/llvm/lib/Target/AMDGPU/VOPInstructions.td
index 1e47acb5fde4f..e81f22c221fc2 100644
--- a/llvm/lib/Target/AMDGPU/VOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOPInstructions.td
@@ -405,11 +405,11 @@ class VOP3Interp_gfx11<bits<10> op, VOPProfile p> : VOP3Interp_gfx10<op, p>;
class VOP3be <VOPProfile P> : Enc64 {
bits<8> vdst;
- bits<2> src0_modifiers;
+ bits<5> src0_modifiers;
bits<9> src0;
- bits<2> src1_modifiers;
+ bits<5> src1_modifiers;
bits<9> src1;
- bits<2> src2_modifiers;
+ bits<5> src2_modifiers;
bits<9> src2;
bits<7> sdst;
bits<2> omod;
@@ -626,9 +626,9 @@ def SDWA {
class VOP_SDWAe<VOPProfile P> : Enc64 {
bits<8> src0;
bits<3> src0_sel;
- bits<2> src0_modifiers; // float: {abs,neg}, int {sext}
+ bits<5> src0_modifiers; // float: {abs,neg}, int {sext}
bits<3> src1_sel;
- bits<2> src1_modifiers;
+ bits<5> src1_modifiers;
bits<3> dst_sel;
bits<2> dst_unused;
bits<1> clamp;
@@ -638,10 +638,10 @@ class VOP_SDWAe<VOPProfile P> : Enc64 {
let Inst{44-43} = !if(P.EmitDstSel, dst_unused{1-0}, ?);
let Inst{45} = !if(P.HasSDWAClamp, clamp{0}, 0);
let Inst{50-48} = !if(P.HasSrc0, src0_sel{2-0}, 0);
- let Inst{51} = !if(P.HasSrc0IntMods, src0_modifiers{0}, 0);
+ let Inst{51} = !if(P.HasSrc0IntMods, src0_modifiers{4}, 0);
let Inst{53-52} = !if(P.HasSrc0FloatMods, src0_modifiers{1-0}, 0);
let Inst{58-56} = !if(P.HasSrc1, src1_sel{2-0}, 0);
- let Inst{59} = !if(P.HasSrc1IntMods, src1_modifiers{0}, 0);
+ let Inst{59} = !if(P.HasSrc1IntMods, src1_modifiers{4}, 0);
let Inst{61-60} = !if(P.HasSrc1FloatMods, src1_modifiers{1-0}, 0);
}
@@ -662,18 +662,18 @@ class VOP_SDWAe<VOPProfile P> : Enc64 {
class VOP_SDWA9e<VOPProfile P> : Enc64 {
bits<9> src0; // {src0_sgpr{0}, src0{7-0}}
bits<3> src0_sel;
- bits<2> src0_modifiers; // float: {abs,neg}, int {sext}
+ bits<5> src0_modifiers; // float: {abs,neg}, int {sext}
bits<3> src1_sel;
- bits<2> src1_modifiers;
+ bits<5> src1_modifiers;
bits<1> src1_sgpr;
let Inst{39-32} = !if(P.HasSrc0, src0{7-0}, 0);
let Inst{50-48} = !if(P.HasSrc0, src0_sel{2-0}, 0);
- let Inst{51} = !if(P.HasSrc0IntMods, src0_modifiers{0}, 0);
+ let Inst{51} = !if(P.HasSrc0IntMods, src0_modifiers{4}, 0);
let Inst{53-52} = !if(P.HasSrc0FloatMods, src0_modifiers{1-0}, 0);
let Inst{55} = !if(P.HasSrc0, src0{8}, 0);
let Inst{58-56} = !if(P.HasSrc1, src1_sel{2-0}, 0);
- let Inst{59} = !if(P.HasSrc1IntMods, src1_modifiers{0}, 0);
+ let Inst{59} = !if(P.HasSrc1IntMods, src1_modifiers{4}, 0);
let Inst{61-60} = !if(P.HasSrc1FloatMods, src1_modifiers{1-0}, 0);
let Inst{63} = 0; // src1_sgpr - should be specified in subclass
}
@@ -819,9 +819,9 @@ class VOP_SDWA10_Real<VOP_SDWA_Pseudo ps> :
Base_VOP_SDWA10_Real<ps>, SIMCInstr<ps.PseudoInstr, SIEncodingFamily.SDWA10>;
class VOP_DPPe<VOPProfile P, bit IsDPP16=0> : Enc64 {
- bits<2> src0_modifiers;
+ bits<5> src0_modifiers;
bits<8> src0;
- bits<2> src1_modifiers;
+ bits<5> src1_modifiers;
bits<9> dpp_ctrl;
bits<1> bound_ctrl;
bits<4> bank_mask;
diff --git a/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-combine-sel-src.mir b/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-combine-sel-src.mir
index 14ba8fccb172d..1c20db9577695 100644
--- a/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-combine-sel-src.mir
+++ b/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-combine-sel-src.mir
@@ -484,7 +484,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 16, 8, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -572,7 +572,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 16, 8, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -638,7 +638,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 24, 8, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -704,7 +704,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 24, 8, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -792,7 +792,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 16, 16, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -814,7 +814,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 16, 16, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -836,7 +836,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 16, 16, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -902,7 +902,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 16, 16, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -924,7 +924,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 16, 16, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -946,7 +946,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 0, 32, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 6, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 6, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -968,7 +968,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 0, 32, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 5, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -990,7 +990,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 0, 32, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 4, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 4, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -1012,7 +1012,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 0, 32, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 3, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -1034,7 +1034,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 0, 32, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 2, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -1056,7 +1056,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 0, 32, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 1, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 1, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
@@ -1078,7 +1078,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; CHECK-NEXT: [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[COPY]], 0, [[COPY]], 0, 1, 0, 5, 0, implicit $exec
; CHECK-NEXT: [[V_BFE_I32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_I32_e64 [[V_LSHRREV_B32_sdwa]], 0, 32, implicit $exec
- ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 1, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 0, implicit $exec
+ ; CHECK-NEXT: [[V_LSHRREV_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_LSHRREV_B32_sdwa]], 16, [[V_LSHRREV_B32_sdwa]], 0, 1, 0, 6, 0, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
%1:vgpr_32 = COPY $vgpr0
%2:vgpr_32 = V_LSHRREV_B32_sdwa 0, %1, 0, %1, 0, 1, 0, 5, 0, implicit $exec
More information about the llvm-commits
mailing list