[llvm] [SystemZ] Fix code in widening vector multiplication (PR #150836)
via llvm-commits
llvm-commits at lists.llvm.org
Sun Jul 27 08:54:17 PDT 2025
https://github.com/sujianIBM created https://github.com/llvm/llvm-project/pull/150836
Commit cdc7864 has an error which would wrongly fold widening multiplications into an even/odd widening operation.
This PR fixes it and adds tests to check scenarios which should not be folded into an even/odd widening operation are actually not.
>From 43455f040bd8f0883650af04bea0f596d23b8219 Mon Sep 17 00:00:00 2001
From: Jian Su <jian.su at ibm.com>
Date: Sun, 27 Jul 2025 15:34:38 +0000
Subject: [PATCH] Fix logic in detectEvenOddMultiplyOperand().
---
.../Target/SystemZ/SystemZISelLowering.cpp | 2 +-
llvm/test/CodeGen/SystemZ/vec-mul-07.ll | 114 ++++++++++++++++++
2 files changed, 115 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp b/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp
index e30d7235b81b6..fb0a47dc9dc47 100644
--- a/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp
+++ b/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp
@@ -9044,7 +9044,7 @@ static unsigned detectEvenOddMultiplyOperand(const SelectionDAG &DAG,
if (unsigned(ShuffleMask[Elt]) != 2 * Elt)
CanUseEven = false;
if (unsigned(ShuffleMask[Elt]) != 2 * Elt + 1)
- CanUseEven = true;
+ CanUseOdd = false;
}
Op = Op.getOperand(0);
if (CanUseEven)
diff --git a/llvm/test/CodeGen/SystemZ/vec-mul-07.ll b/llvm/test/CodeGen/SystemZ/vec-mul-07.ll
index 73c7a8dec5dfc..583561625cfc0 100644
--- a/llvm/test/CodeGen/SystemZ/vec-mul-07.ll
+++ b/llvm/test/CodeGen/SystemZ/vec-mul-07.ll
@@ -3,6 +3,23 @@
;
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
+; Test a v16i8 -> v8i16 unsigned widening multiplication
+; which is not folded into an even/odd widening operation.
+define <8 x i16> @f1_not(<16 x i8> %val1, <16 x i8> %val2) {
+; CHECK-LABEL: f1_not:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vuplhb %v0, %v24
+; CHECK-NEXT: vuplhb %v1, %v26
+; CHECK-NEXT: vmlhw %v24, %v0, %v1
+; CHECK-NEXT: br %r14
+ %shuf1 = shufflevector <16 x i8> %val1, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %zext1 = zext <8 x i8> %shuf1 to <8 x i16>
+ %shuf2 = shufflevector <16 x i8> %val2, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %zext2 = zext <8 x i8> %shuf2 to <8 x i16>
+ %ret = mul <8 x i16> %zext1, %zext2
+ ret <8 x i16> %ret
+}
+
; Test a v16i8 (even) -> v8i16 unsigned widening multiplication.
define <8 x i16> @f1(<16 x i8> %val1, <16 x i8> %val2) {
; CHECK-LABEL: f1:
@@ -31,6 +48,23 @@ define <8 x i16> @f2(<16 x i8> %val1, <16 x i8> %val2) {
ret <8 x i16> %ret
}
+; Test a v16i8 -> v8i16 signed widening multiplication
+; which is not folded into an even/odd widening operation.
+define <8 x i16> @f3_not(<16 x i8> %val1, <16 x i8> %val2) {
+; CHECK-LABEL: f3_not:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vuphb %v0, %v26
+; CHECK-NEXT: vuphb %v1, %v24
+; CHECK-NEXT: vmlhw %v24, %v1, %v0
+; CHECK-NEXT: br %r14
+ %shuf1 = shufflevector <16 x i8> %val1, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %sext1 = sext <8 x i8> %shuf1 to <8 x i16>
+ %shuf2 = shufflevector <16 x i8> %val2, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %sext2 = sext <8 x i8> %shuf2 to <8 x i16>
+ %ret = mul <8 x i16> %sext1, %sext2
+ ret <8 x i16> %ret
+}
+
; Test a v16i8 (even) -> v8i16 signed widening multiplication.
define <8 x i16> @f3(<16 x i8> %val1, <16 x i8> %val2) {
; CHECK-LABEL: f3:
@@ -59,6 +93,23 @@ define <8 x i16> @f4(<16 x i8> %val1, <16 x i8> %val2) {
ret <8 x i16> %ret
}
+; Test a v8i16 -> v4i32 unsigned widening multiplication
+; which is not folded into an even/odd widening operation.
+define <4 x i32> @f5_not(<8 x i16> %val1, <8 x i16> %val2) {
+; CHECK-LABEL: f5_not:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vuplhh %v0, %v24
+; CHECK-NEXT: vuplhh %v1, %v26
+; CHECK-NEXT: vmlf %v24, %v0, %v1
+; CHECK-NEXT: br %r14
+ %shuf1 = shufflevector <8 x i16> %val1, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %zext1 = zext <4 x i16> %shuf1 to <4 x i32>
+ %shuf2 = shufflevector <8 x i16> %val2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %zext2 = zext <4 x i16> %shuf2 to <4 x i32>
+ %ret = mul <4 x i32> %zext1, %zext2
+ ret <4 x i32> %ret
+}
+
; Test a v8i16 (even) -> v4i32 unsigned widening multiplication.
define <4 x i32> @f5(<8 x i16> %val1, <8 x i16> %val2) {
; CHECK-LABEL: f5:
@@ -87,6 +138,23 @@ define <4 x i32> @f6(<8 x i16> %val1, <8 x i16> %val2) {
ret <4 x i32> %ret
}
+; Test a v8i16 -> v4i32 signed widening multiplication
+; which is not folded into an even/odd widening operation.
+define <4 x i32> @f7_not(<8 x i16> %val1, <8 x i16> %val2) {
+; CHECK-LABEL: f7_not:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vuphh %v0, %v26
+; CHECK-NEXT: vuphh %v1, %v24
+; CHECK-NEXT: vmlf %v24, %v1, %v0
+; CHECK-NEXT: br %r14
+ %shuf1 = shufflevector <8 x i16> %val1, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %sext1 = sext <4 x i16> %shuf1 to <4 x i32>
+ %shuf2 = shufflevector <8 x i16> %val2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %sext2 = sext <4 x i16> %shuf2 to <4 x i32>
+ %ret = mul <4 x i32> %sext1, %sext2
+ ret <4 x i32> %ret
+}
+
; Test a v8i16 (even) -> v4i32 signed widening multiplication.
define <4 x i32> @f7(<8 x i16> %val1, <8 x i16> %val2) {
; CHECK-LABEL: f7:
@@ -115,6 +183,29 @@ define <4 x i32> @f8(<8 x i16> %val1, <8 x i16> %val2) {
ret <4 x i32> %ret
}
+; Test a v4i32 -> v2i64 unsigned widening multiplication
+; which is not folded into an even/odd widening operation.
+define <2 x i64> @f9_not(<4 x i32> %val1, <4 x i32> %val2) {
+; CHECK-LABEL: f9_not:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vuplhf %v0, %v24
+; CHECK-NEXT: vuplhf %v1, %v26
+; CHECK-NEXT: vlgvg %r0, %v1, 1
+; CHECK-NEXT: vlgvg %r1, %v0, 1
+; CHECK-NEXT: msgr %r1, %r0
+; CHECK-NEXT: vlgvg %r0, %v1, 0
+; CHECK-NEXT: vlgvg %r2, %v0, 0
+; CHECK-NEXT: msgr %r2, %r0
+; CHECK-NEXT: vlvgp %v24, %r2, %r1
+; CHECK-NEXT: br %r14
+ %shuf1 = shufflevector <4 x i32> %val1, <4 x i32> poison, <2 x i32> <i32 0, i32 1>
+ %zext1 = zext <2 x i32> %shuf1 to <2 x i64>
+ %shuf2 = shufflevector <4 x i32> %val2, <4 x i32> poison, <2 x i32> <i32 0, i32 1>
+ %zext2 = zext <2 x i32> %shuf2 to <2 x i64>
+ %ret = mul <2 x i64> %zext1, %zext2
+ ret <2 x i64> %ret
+}
+
; Test a v4i32 (even) -> v2i64 unsigned widening multiplication.
define <2 x i64> @f9(<4 x i32> %val1, <4 x i32> %val2) {
; CHECK-LABEL: f9:
@@ -143,6 +234,29 @@ define <2 x i64> @f10(<4 x i32> %val1, <4 x i32> %val2) {
ret <2 x i64> %ret
}
+; Test a v4i32 -> v2i64 signed widening multiplication
+; which is not folded into an even/odd widening operation.
+define <2 x i64> @f11_not(<4 x i32> %val1, <4 x i32> %val2) {
+; CHECK-LABEL: f11_not:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vuphf %v0, %v24
+; CHECK-NEXT: vuphf %v1, %v26
+; CHECK-NEXT: vlgvg %r0, %v1, 1
+; CHECK-NEXT: vlgvg %r1, %v0, 1
+; CHECK-NEXT: msgr %r1, %r0
+; CHECK-NEXT: vlgvg %r0, %v1, 0
+; CHECK-NEXT: vlgvg %r2, %v0, 0
+; CHECK-NEXT: msgr %r2, %r0
+; CHECK-NEXT: vlvgp %v24, %r2, %r1
+; CHECK-NEXT: br %r14
+ %shuf1 = shufflevector <4 x i32> %val1, <4 x i32> poison, <2 x i32> <i32 0, i32 1>
+ %sext1 = sext <2 x i32> %shuf1 to <2 x i64>
+ %shuf2 = shufflevector <4 x i32> %val2, <4 x i32> poison, <2 x i32> <i32 0, i32 1>
+ %sext2 = sext <2 x i32> %shuf2 to <2 x i64>
+ %ret = mul <2 x i64> %sext1, %sext2
+ ret <2 x i64> %ret
+}
+
; Test a v4i32 (even) -> v2i64 signed widening multiplication.
define <2 x i64> @f11(<4 x i32> %val1, <4 x i32> %val2) {
; CHECK-LABEL: f11:
More information about the llvm-commits
mailing list