[RISCV] Use ISD::XOR instead of RISCVISD::VMXOR_VL in lowerVectorMaskVecReduction of scalable ISD::VECREDUCE_AND #121812
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…VecReduction of scalable ISD::VECREDUCE_AND This allows combining the XOR with earlier ISD::ANDs inserted by type legalization.
Member
|
@llvm/pr-subscribers-backend-risc-v Author: Craig Topper (topperc) ChangesThis allows combining the XOR with earlier ISD::ANDs inserted by type legalization. Full diff: https://github.com/llvm/llvm-project/pull/121812.diff 2 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 7efe3732d8be13..0d443cf7ec5c83 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -10154,7 +10154,10 @@ SDValue RISCVTargetLowering::lowerVectorMaskVecReduction(SDValue Op,
case ISD::VP_REDUCE_AND: {
// vcpop ~x == 0
SDValue TrueMask = DAG.getNode(RISCVISD::VMSET_VL, DL, ContainerVT, VL);
- Vec = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Vec, TrueMask, VL);
+ if (IsVP || VecVT.isFixedLengthVector())
+ Vec = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Vec, TrueMask, VL);
+ else
+ Vec = DAG.getNode(ISD::XOR, DL, ContainerVT, Vec, TrueMask);
Vec = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Vec, Mask, VL);
CC = ISD::SETEQ;
break;
diff --git a/llvm/test/CodeGen/RISCV/rvv/vreductions-mask.ll b/llvm/test/CodeGen/RISCV/rvv/vreductions-mask.ll
index d99fd036b4fc92..ce9d6c5ab91a8a 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vreductions-mask.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vreductions-mask.ll
@@ -785,8 +785,7 @@ define zeroext i1 @vreduce_and_nxv128i1(<vscale x 128 x i1> %v) {
; CHECK-LABEL: vreduce_and_nxv128i1:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a0, zero, e8, m8, ta, ma
-; CHECK-NEXT: vmand.mm v8, v0, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v0, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -814,8 +813,7 @@ define zeroext i1 @vreduce_smax_nxv128i1(<vscale x 128 x i1> %v) {
; CHECK-LABEL: vreduce_smax_nxv128i1:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a0, zero, e8, m8, ta, ma
-; CHECK-NEXT: vmand.mm v8, v0, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v0, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -829,8 +827,7 @@ define zeroext i1 @vreduce_umin_nxv128i1(<vscale x 128 x i1> %v) {
; CHECK-LABEL: vreduce_umin_nxv128i1:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a0, zero, e8, m8, ta, ma
-; CHECK-NEXT: vmand.mm v8, v0, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v0, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -892,8 +889,7 @@ define zeroext i1 @vreduce_and_nxv256i1(<vscale x 256 x i1> %v) {
; CHECK-NEXT: vsetvli a0, zero, e8, m8, ta, ma
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v0, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -925,8 +921,7 @@ define zeroext i1 @vreduce_smax_nxv256i1(<vscale x 256 x i1> %v) {
; CHECK-NEXT: vsetvli a0, zero, e8, m8, ta, ma
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v0, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -942,8 +937,7 @@ define zeroext i1 @vreduce_umin_nxv256i1(<vscale x 256 x i1> %v) {
; CHECK-NEXT: vsetvli a0, zero, e8, m8, ta, ma
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v0, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -1019,8 +1013,7 @@ define zeroext i1 @vreduce_and_nxv512i1(<vscale x 512 x i1> %v) {
; CHECK-NEXT: vmand.mm v11, v0, v11
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v11, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -1060,8 +1053,7 @@ define zeroext i1 @vreduce_smax_nxv512i1(<vscale x 512 x i1> %v) {
; CHECK-NEXT: vmand.mm v11, v0, v11
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v11, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -1081,8 +1073,7 @@ define zeroext i1 @vreduce_umin_nxv512i1(<vscale x 512 x i1> %v) {
; CHECK-NEXT: vmand.mm v11, v0, v11
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v11, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -1186,8 +1177,7 @@ define zeroext i1 @vreduce_and_nxv1024i1(<vscale x 1024 x i1> %v) {
; CHECK-NEXT: vmand.mm v11, v15, v11
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v11, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -1243,8 +1233,7 @@ define zeroext i1 @vreduce_smax_nxv1024i1(<vscale x 1024 x i1> %v) {
; CHECK-NEXT: vmand.mm v11, v15, v11
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v11, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
@@ -1272,8 +1261,7 @@ define zeroext i1 @vreduce_umin_nxv1024i1(<vscale x 1024 x i1> %v) {
; CHECK-NEXT: vmand.mm v11, v15, v11
; CHECK-NEXT: vmand.mm v8, v8, v10
; CHECK-NEXT: vmand.mm v9, v11, v9
-; CHECK-NEXT: vmand.mm v8, v9, v8
-; CHECK-NEXT: vmnot.m v8, v8
+; CHECK-NEXT: vmnand.mm v8, v9, v8
; CHECK-NEXT: vcpop.m a0, v8
; CHECK-NEXT: seqz a0, a0
; CHECK-NEXT: ret
|
preames
approved these changes
Jan 6, 2025
Collaborator
preames
left a comment
preames
left a comment
There was a problem hiding this comment.
LGTM w/optional comment.
| // vcpop ~x == 0 | ||
| SDValue TrueMask = DAG.getNode(RISCVISD::VMSET_VL, DL, ContainerVT, VL); | ||
| Vec = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Vec, TrueMask, VL); | ||
| if (IsVP || VecVT.isFixedLengthVector()) |
Collaborator
There was a problem hiding this comment.
Do we need the special case? I'm somewhat surprised that the plain ISD::XOR doesn't work out fine for the fixed case at least. Any idea why? Even for the VP case, I think we only need the mask and VL on the vcpop itself?
Collaborator
Author
There was a problem hiding this comment.
It caused an extra vtype toggle for fixed vectors. I didn't dig any further.
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/88/builds/6462 Here is the relevant piece of the build log for the reference |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This allows combining the XOR with earlier ISD::ANDs inserted by type legalization.