[RISCV] Add isCommutable for VDOTA4 and VDOTA4U#190090
Merged
Conversation
Mark PseudoVDOTA4_VV and PseudoVDOTA4U_VV as commutable since both source operands have the same signedness. VDOTA4SU is left non-commutable because its operands differ in signedness (signed x unsigned). Add findCommutedOpIndices cases for the new commutable pseudos and a test covering commutable and non-commutable dot product variants. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Member
|
@llvm/pr-subscribers-backend-risc-v Author: Jim Lin (tclin914) ChangesMark PseudoVDOTA4_VV and PseudoVDOTA4U_VV as commutable since both Add findCommutedOpIndices cases for the new commutable pseudos and Full diff: https://github.com/llvm/llvm-project/pull/190090.diff 3 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
index 089683a43f800..fd13c0e3607b3 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
@@ -4116,6 +4116,16 @@ bool RISCVInstrInfo::findCommutedOpIndices(const MachineInstr &MI,
case CASE_RVV_OPCODE(VAADD_VV):
case CASE_RVV_OPCODE(VAADDU_VV):
case CASE_RVV_OPCODE(VSMUL_VV):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4_VV, MF2):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4_VV, M1):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4_VV, M2):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4_VV, M4):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4_VV, M8):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4U_VV, MF2):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4U_VV, M1):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4U_VV, M2):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4U_VV, M4):
+ case CASE_RVV_OPCODE_LMUL(VDOTA4U_VV, M8):
// Operands 2 and 3 are commutable.
return fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, 2, 3);
case CASE_VFMA_SPLATS(FMADD):
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZvdot4a8i.td b/llvm/lib/Target/RISCV/RISCVInstrInfoZvdot4a8i.td
index b9fbe3e5286ee..d8c60dc9a584c 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoZvdot4a8i.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZvdot4a8i.td
@@ -61,9 +61,9 @@ let HasPassthruOp = true, HasMaskOp = true in {
// Pseudo Instructions for CodeGen
//===----------------------------------------------------------------------===//
-multiclass VPseudoVDOTA4_VV_VX {
+multiclass VPseudoVDOTA4_VV_VX<bit Commutable = 0> {
foreach m = MxSet<32>.m in {
- defm "" : VPseudoBinaryV_VV<m>,
+ defm "" : VPseudoBinaryV_VV<m, Commutable=Commutable>,
SchedBinary<"WriteVIMulAddV", "ReadVIMulAddV", "ReadVIMulAddV", m.MX,
forcePassthruRead=true>;
defm "" : VPseudoBinaryV_VX<m>,
@@ -72,11 +72,10 @@ multiclass VPseudoVDOTA4_VV_VX {
}
}
-// TODO: Add isCommutable for VDOTA4 and VDOTA4U
let Predicates = [HasStdExtZvdot4a8i], mayLoad = 0, mayStore = 0,
hasSideEffects = 0 in {
- defm PseudoVDOTA4 : VPseudoVDOTA4_VV_VX;
- defm PseudoVDOTA4U : VPseudoVDOTA4_VV_VX;
+ defm PseudoVDOTA4 : VPseudoVDOTA4_VV_VX<Commutable=1>;
+ defm PseudoVDOTA4U : VPseudoVDOTA4_VV_VX<Commutable=1>;
defm PseudoVDOTA4SU : VPseudoVDOTA4_VV_VX;
// VDOTA4US does not have a VV variant
foreach m = MxListVF4 in {
diff --git a/llvm/test/CodeGen/RISCV/rvv/commutable-zvdot4a8i.ll b/llvm/test/CodeGen/RISCV/rvv/commutable-zvdot4a8i.ll
new file mode 100644
index 0000000000000..e5b3324651fce
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/commutable-zvdot4a8i.ll
@@ -0,0 +1,126 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=riscv32 -mattr=+zve64x,+experimental-zvdot4a8i \
+; RUN: -verify-machineinstrs | FileCheck %s
+; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=riscv64 -mattr=+zve64x,+experimental-zvdot4a8i \
+; RUN: -verify-machineinstrs | FileCheck %s
+
+; vdota4.vv - commutable
+define <vscale x 2 x i32> @commutable_vdota4_vv(<vscale x 2 x i32> %0, <vscale x 2 x i32> %1, iXLen %2) nounwind {
+; CHECK-LABEL: commutable_vdota4_vv:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vsetvli zero, a0, e32, m1, ta, ma
+; CHECK-NEXT: vdota4.vv v8, v8, v9
+; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
+; CHECK-NEXT: vadd.vv v8, v8, v8
+; CHECK-NEXT: ret
+entry:
+ %a = call <vscale x 2 x i32> @llvm.riscv.vdota4.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %0,
+ <vscale x 2 x i32> %1,
+ iXLen %2, iXLen 1)
+ %b = call <vscale x 2 x i32> @llvm.riscv.vdota4.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %1,
+ <vscale x 2 x i32> %0,
+ iXLen %2, iXLen 1)
+ %ret = add <vscale x 2 x i32> %a, %b
+ ret <vscale x 2 x i32> %ret
+}
+
+define <vscale x 2 x i32> @commutable_vdota4_vv_masked(<vscale x 2 x i32> %0, <vscale x 2 x i32> %1, <vscale x 2 x i1> %mask, iXLen %2) {
+; CHECK-LABEL: commutable_vdota4_vv_masked:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli zero, a0, e32, m1, ta, ma
+; CHECK-NEXT: vdota4.vv v8, v8, v9, v0.t
+; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
+; CHECK-NEXT: vadd.vv v8, v8, v8
+; CHECK-NEXT: ret
+ %a = call <vscale x 2 x i32> @llvm.riscv.vdota4.mask.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %0,
+ <vscale x 2 x i32> %1,
+ <vscale x 2 x i1> %mask,
+ iXLen %2, iXLen 1)
+ %b = call <vscale x 2 x i32> @llvm.riscv.vdota4.mask.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %1,
+ <vscale x 2 x i32> %0,
+ <vscale x 2 x i1> %mask,
+ iXLen %2, iXLen 1)
+ %ret = add <vscale x 2 x i32> %a, %b
+ ret <vscale x 2 x i32> %ret
+}
+
+; vdota4u.vv - commutable
+define <vscale x 2 x i32> @commutable_vdota4u_vv(<vscale x 2 x i32> %0, <vscale x 2 x i32> %1, iXLen %2) nounwind {
+; CHECK-LABEL: commutable_vdota4u_vv:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vsetvli zero, a0, e32, m1, ta, ma
+; CHECK-NEXT: vdota4u.vv v8, v8, v9
+; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
+; CHECK-NEXT: vadd.vv v8, v8, v8
+; CHECK-NEXT: ret
+entry:
+ %a = call <vscale x 2 x i32> @llvm.riscv.vdota4u.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %0,
+ <vscale x 2 x i32> %1,
+ iXLen %2, iXLen 1)
+ %b = call <vscale x 2 x i32> @llvm.riscv.vdota4u.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %1,
+ <vscale x 2 x i32> %0,
+ iXLen %2, iXLen 1)
+ %ret = add <vscale x 2 x i32> %a, %b
+ ret <vscale x 2 x i32> %ret
+}
+
+define <vscale x 2 x i32> @commutable_vdota4u_vv_masked(<vscale x 2 x i32> %0, <vscale x 2 x i32> %1, <vscale x 2 x i1> %mask, iXLen %2) {
+; CHECK-LABEL: commutable_vdota4u_vv_masked:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli zero, a0, e32, m1, ta, ma
+; CHECK-NEXT: vdota4u.vv v8, v8, v9, v0.t
+; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
+; CHECK-NEXT: vadd.vv v8, v8, v8
+; CHECK-NEXT: ret
+ %a = call <vscale x 2 x i32> @llvm.riscv.vdota4u.mask.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %0,
+ <vscale x 2 x i32> %1,
+ <vscale x 2 x i1> %mask,
+ iXLen %2, iXLen 1)
+ %b = call <vscale x 2 x i32> @llvm.riscv.vdota4u.mask.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %1,
+ <vscale x 2 x i32> %0,
+ <vscale x 2 x i1> %mask,
+ iXLen %2, iXLen 1)
+ %ret = add <vscale x 2 x i32> %a, %b
+ ret <vscale x 2 x i32> %ret
+}
+
+; vdota4su.vv - NOT commutable (signed x unsigned, operand order matters)
+define <vscale x 2 x i32> @commutable_vdota4su_vv(<vscale x 2 x i32> %0, <vscale x 2 x i32> %1, iXLen %2) nounwind {
+; CHECK-LABEL: commutable_vdota4su_vv:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vsetvli zero, a0, e32, m1, ta, ma
+; CHECK-NEXT: vdota4su.vv v10, v8, v9
+; CHECK-NEXT: vdota4su.vv v8, v9, v8
+; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
+; CHECK-NEXT: vadd.vv v8, v10, v8
+; CHECK-NEXT: ret
+entry:
+ %a = call <vscale x 2 x i32> @llvm.riscv.vdota4su.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %0,
+ <vscale x 2 x i32> %1,
+ iXLen %2, iXLen 1)
+ %b = call <vscale x 2 x i32> @llvm.riscv.vdota4su.nxv2i32.nxv2i32(
+ <vscale x 2 x i32> poison,
+ <vscale x 2 x i32> %1,
+ <vscale x 2 x i32> %0,
+ iXLen %2, iXLen 1)
+ %ret = add <vscale x 2 x i32> %a, %b
+ ret <vscale x 2 x i32> %ret
+}
|
wangpc-pp
approved these changes
Apr 2, 2026
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
Mark PseudoVDOTA4_VV and PseudoVDOTA4U_VV as commutable since both
source operands have the same signedness. VDOTA4SU is left
non-commutable because its operands differ in signedness (signed x
unsigned).
Add findCommutedOpIndices cases for the new commutable pseudos and
a test covering commutable and non-commutable dot product variants.