[llvm] [LLVM][DAGCombiner] fold (shl (X * vscale(C0)), C1) -> (X * vscale(C0 << C1)). (PR #150651)
Paul Walker via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 31 06:27:25 PDT 2025
https://github.com/paulwalker-arm updated https://github.com/llvm/llvm-project/pull/150651
>From 568aa4a2f098ac077a8ffc57d2795d6c80a457ce Mon Sep 17 00:00:00 2001
From: Paul Walker <paul.walker at arm.com>
Date: Fri, 25 Jul 2025 17:15:32 +0100
Subject: [PATCH 1/3] Regeneate sve-vscale-combine.ll. Add tests to show
missing combine.
---
.../CodeGen/AArch64/sve-vscale-combine.ll | 109 ++++++++++++------
1 file changed, 72 insertions(+), 37 deletions(-)
diff --git a/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll b/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll
index 9306c208c3b55..47ea20313c7ea 100644
--- a/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll
+++ b/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll
@@ -1,14 +1,14 @@
-; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve --asm-verbose=false < %s |FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mattr=+sve < %s | FileCheck %s
-declare i32 @llvm.vscale.i32()
-declare i64 @llvm.vscale.i64()
+target triple = "aarch64-unknown-linux-gnu"
; Fold (add (vscale * C0), (vscale * C1)) to (vscale * (C0 + C1)).
define i64 @combine_add_vscale_i64() nounwind {
; CHECK-LABEL: combine_add_vscale_i64:
-; CHECK-NOT: add
-; CHECK-NEXT: cntd x0
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: cntd x0
+; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%add = add i64 %vscale, %vscale
ret i64 %add
@@ -16,9 +16,10 @@ define i64 @combine_add_vscale_i64() nounwind {
define i32 @combine_add_vscale_i32() nounwind {
; CHECK-LABEL: combine_add_vscale_i32:
-; CHECK-NOT: add
-; CHECK-NEXT: cntd x0
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: cntd x0
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
+; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%add = add i32 %vscale, %vscale
ret i32 %add
@@ -28,9 +29,9 @@ define i32 @combine_add_vscale_i32() nounwind {
; In this test, C0 = 1, C1 = 32.
define i64 @combine_mul_vscale_i64() nounwind {
; CHECK-LABEL: combine_mul_vscale_i64:
-; CHECK-NOT: mul
-; CHECK-NEXT: rdvl x0, #2
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x0, #2
+; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%mul = mul i64 %vscale, 32
ret i64 %mul
@@ -38,9 +39,10 @@ define i64 @combine_mul_vscale_i64() nounwind {
define i32 @combine_mul_vscale_i32() nounwind {
; CHECK-LABEL: combine_mul_vscale_i32:
-; CHECK-NOT: mul
-; CHECK-NEXT: rdvl x0, #3
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x0, #3
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
+; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%mul = mul i32 %vscale, 48
ret i32 %mul
@@ -49,11 +51,11 @@ define i32 @combine_mul_vscale_i32() nounwind {
; Canonicalize (sub X, (vscale * C)) to (add X, (vscale * -C))
define i64 @combine_sub_vscale_i64(i64 %in) nounwind {
; CHECK-LABEL: combine_sub_vscale_i64:
-; CHECK-NOT: sub
-; CHECK-NEXT: rdvl x8, #-1
-; CHECK-NEXT: asr x8, x8, #4
-; CHECK-NEXT: add x0, x0, x8
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x8, #-1
+; CHECK-NEXT: asr x8, x8, #4
+; CHECK-NEXT: add x0, x0, x8
+; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%sub = sub i64 %in, %vscale
ret i64 %sub
@@ -61,11 +63,11 @@ define i64 @combine_sub_vscale_i64(i64 %in) nounwind {
define i32 @combine_sub_vscale_i32(i32 %in) nounwind {
; CHECK-LABEL: combine_sub_vscale_i32:
-; CHECK-NOT: sub
-; CHECK-NEXT: rdvl x8, #-1
-; CHECK-NEXT: asr x8, x8, #4
-; CHECK-NEXT: add w0, w0, w8
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x8, #-1
+; CHECK-NEXT: asr x8, x8, #4
+; CHECK-NEXT: add w0, w0, w8
+; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%sub = sub i32 %in, %vscale
ret i32 %sub
@@ -75,12 +77,13 @@ define i32 @combine_sub_vscale_i32(i32 %in) nounwind {
; (sub X, (vscale * C)) to (add X, (vscale * -C))
define i64 @multiple_uses_sub_vscale_i64(i64 %x, i64 %y) nounwind {
; CHECK-LABEL: multiple_uses_sub_vscale_i64:
-; CHECK-NEXT: rdvl x8, #1
-; CHECK-NEXT: lsr x8, x8, #4
-; CHECK-NEXT: sub x9, x0, x8
-; CHECK-NEXT: add x8, x1, x8
-; CHECK-NEXT: mul x0, x9, x8
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x8, #1
+; CHECK-NEXT: lsr x8, x8, #4
+; CHECK-NEXT: sub x9, x0, x8
+; CHECK-NEXT: add x8, x1, x8
+; CHECK-NEXT: mul x0, x9, x8
+; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%sub = sub i64 %x, %vscale
%add = add i64 %y, %vscale
@@ -95,9 +98,9 @@ define i64 @multiple_uses_sub_vscale_i64(i64 %x, i64 %y) nounwind {
; Hence, the immediate for RDVL is #1.
define i64 @combine_shl_vscale_i64() nounwind {
; CHECK-LABEL: combine_shl_vscale_i64:
-; CHECK-NOT: shl
-; CHECK-NEXT: rdvl x0, #1
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x0, #1
+; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%shl = shl i64 %vscale, 4
ret i64 %shl
@@ -105,10 +108,42 @@ define i64 @combine_shl_vscale_i64() nounwind {
define i32 @combine_shl_vscale_i32() nounwind {
; CHECK-LABEL: combine_shl_vscale_i32:
-; CHECK-NOT: shl
-; CHECK-NEXT: rdvl x0, #1
-; CHECK-NEXT: ret
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x0, #1
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
+; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%shl = shl i32 %vscale, 4
ret i32 %shl
}
+
+define i64 @combine_shl_mul_vscale(i64 %a) nounwind {
+; CHECK-LABEL: combine_shl_mul_vscale:
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x8, #1
+; CHECK-NEXT: lsr x8, x8, #4
+; CHECK-NEXT: mul x8, x0, x8
+; CHECK-NEXT: lsl x0, x8, #3
+; CHECK-NEXT: ret
+ %vscale = tail call i64 @llvm.vscale.i64()
+ %mul = mul i64 %a, %vscale
+ %shl = shl i64 %mul, 3
+ ret i64 %shl
+}
+
+define i64 @combine_shl_mul_vscale_commuted(i64 %a) nounwind {
+; CHECK-LABEL: combine_shl_mul_vscale_commuted:
+; CHECK: // %bb.0:
+; CHECK-NEXT: rdvl x8, #1
+; CHECK-NEXT: lsr x8, x8, #4
+; CHECK-NEXT: mul x8, x8, x0
+; CHECK-NEXT: lsl x0, x8, #3
+; CHECK-NEXT: ret
+ %vscale = tail call i64 @llvm.vscale.i64()
+ %mul = mul i64 %vscale, %a
+ %shl = shl i64 %mul, 3
+ ret i64 %shl
+}
+
+declare i32 @llvm.vscale.i32()
+declare i64 @llvm.vscale.i64()
>From d921291d2982696d6e323a9ce52f23608a54ad24 Mon Sep 17 00:00:00 2001
From: Paul Walker <paul.walker at arm.com>
Date: Fri, 25 Jul 2025 17:21:18 +0100
Subject: [PATCH 2/3] [LLVM][DAGCombiner] fold (shl (X * vscale(C0)), C1) -> (X
* vscale(C0 << C1)).
---
llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 9 +++++++++
llvm/test/CodeGen/AArch64/sve-vscale-combine.ll | 12 ++++--------
.../LoopStrengthReduce/AArch64/vscale-fixups.ll | 8 ++++----
3 files changed, 17 insertions(+), 12 deletions(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index d3df43473013e..8dc0ab74d25c6 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -10615,6 +10615,15 @@ SDValue DAGCombiner::visitSHL(SDNode *N) {
return DAG.getVScale(DL, VT, C0 << C1);
}
+ SDValue X;
+ APInt VS0;
+
+ // fold (shl (X * vscale(VS0)), C1) -> (X * vscale(VS0 << C1))
+ if (N1C && sd_match(N0, m_Mul(m_Value(X), m_VScale(m_ConstInt(VS0))))) {
+ SDValue VScale = DAG.getVScale(DL, VT, VS0 << N1C->getAPIntValue());
+ return DAG.getNode(ISD::MUL, DL, VT, X, VScale);
+ }
+
// Fold (shl step_vector(C0), C1) to (step_vector(C0 << C1)).
APInt ShlVal;
if (N0.getOpcode() == ISD::STEP_VECTOR &&
diff --git a/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll b/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll
index 47ea20313c7ea..7dcd56c3e7dd7 100644
--- a/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll
+++ b/llvm/test/CodeGen/AArch64/sve-vscale-combine.ll
@@ -120,10 +120,8 @@ define i32 @combine_shl_vscale_i32() nounwind {
define i64 @combine_shl_mul_vscale(i64 %a) nounwind {
; CHECK-LABEL: combine_shl_mul_vscale:
; CHECK: // %bb.0:
-; CHECK-NEXT: rdvl x8, #1
-; CHECK-NEXT: lsr x8, x8, #4
-; CHECK-NEXT: mul x8, x0, x8
-; CHECK-NEXT: lsl x0, x8, #3
+; CHECK-NEXT: cnth x8
+; CHECK-NEXT: mul x0, x0, x8
; CHECK-NEXT: ret
%vscale = tail call i64 @llvm.vscale.i64()
%mul = mul i64 %a, %vscale
@@ -134,10 +132,8 @@ define i64 @combine_shl_mul_vscale(i64 %a) nounwind {
define i64 @combine_shl_mul_vscale_commuted(i64 %a) nounwind {
; CHECK-LABEL: combine_shl_mul_vscale_commuted:
; CHECK: // %bb.0:
-; CHECK-NEXT: rdvl x8, #1
-; CHECK-NEXT: lsr x8, x8, #4
-; CHECK-NEXT: mul x8, x8, x0
-; CHECK-NEXT: lsl x0, x8, #3
+; CHECK-NEXT: cnth x8
+; CHECK-NEXT: mul x0, x0, x8
; CHECK-NEXT: ret
%vscale = tail call i64 @llvm.vscale.i64()
%mul = mul i64 %vscale, %a
diff --git a/llvm/test/Transforms/LoopStrengthReduce/AArch64/vscale-fixups.ll b/llvm/test/Transforms/LoopStrengthReduce/AArch64/vscale-fixups.ll
index aa954aeb0ad07..9003072f5fcdf 100644
--- a/llvm/test/Transforms/LoopStrengthReduce/AArch64/vscale-fixups.ll
+++ b/llvm/test/Transforms/LoopStrengthReduce/AArch64/vscale-fixups.ll
@@ -383,14 +383,14 @@ define void @vscale_squared_offset(ptr %alloc) #0 {
; COMMON-LABEL: vscale_squared_offset:
; COMMON: // %bb.0: // %entry
; COMMON-NEXT: rdvl x9, #1
+; COMMON-NEXT: rdvl x10, #4
; COMMON-NEXT: fmov z0.s, #4.00000000
-; COMMON-NEXT: mov x8, xzr
; COMMON-NEXT: lsr x9, x9, #4
; COMMON-NEXT: fmov z1.s, #8.00000000
-; COMMON-NEXT: cntw x10
+; COMMON-NEXT: mov x8, xzr
; COMMON-NEXT: ptrue p0.s, vl1
-; COMMON-NEXT: umull x9, w9, w9
-; COMMON-NEXT: lsl x9, x9, #6
+; COMMON-NEXT: umull x9, w9, w10
+; COMMON-NEXT: cntw x10
; COMMON-NEXT: cmp x8, x10
; COMMON-NEXT: b.ge .LBB6_2
; COMMON-NEXT: .LBB6_1: // %for.body
>From 13d8d4ef22c29da90e3a853e891299ec07da5bf3 Mon Sep 17 00:00:00 2001
From: Paul Walker <paul.walker at arm.com>
Date: Thu, 31 Jul 2025 13:05:24 +0000
Subject: [PATCH 3/3] Prefer NUW flags as per
https://alive2.llvm.org/ce/z/_V5s8p
---
llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 8dc0ab74d25c6..b6bf9b80a0b6f 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -10620,8 +10620,12 @@ SDValue DAGCombiner::visitSHL(SDNode *N) {
// fold (shl (X * vscale(VS0)), C1) -> (X * vscale(VS0 << C1))
if (N1C && sd_match(N0, m_Mul(m_Value(X), m_VScale(m_ConstInt(VS0))))) {
+ SDNodeFlags Flags;
+ Flags.setNoUnsignedWrap(N->getFlags().hasNoUnsignedWrap() &&
+ N0->getFlags().hasNoUnsignedWrap());
+
SDValue VScale = DAG.getVScale(DL, VT, VS0 << N1C->getAPIntValue());
- return DAG.getNode(ISD::MUL, DL, VT, X, VScale);
+ return DAG.getNode(ISD::MUL, DL, VT, X, VScale, Flags);
}
// Fold (shl step_vector(C0), C1) to (step_vector(C0 << C1)).
More information about the llvm-commits
mailing list