release/23.x: [AArch64] Do not allow anyext load in isEquivalentMaskless (#216288) - #216937
Open
llvmbot wants to merge 1 commit into
Open
release/23.x: [AArch64] Do not allow anyext load in isEquivalentMaskless (#216288)#216937llvmbot wants to merge 1 commit into
llvmbot wants to merge 1 commit into
Conversation
isEquivalentMaskless tries to prove that the `and` in `cmp(and(add(x, C1), 0xff), C2)` is redundant and can be removed, providing that the input x is know to be a sext or zext. AFAICT it should not apply to anyext, which could take any value. Fixes llvm#215839 (cherry picked from commit 90c0c67)
Member
Author
|
@efriedma-quic What do you think about merging this PR to the release branch? |
|
@llvm/pr-subscribers-backend-aarch64 Author: llvmbot ChangesBackport 90c0c67 Requested by: @davemgreen Full diff: https://github.com/llvm/llvm-project/pull/216937.diff 2 Files Affected:
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 507817e29ce2b..14488c04c25f4 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -28010,6 +28010,9 @@ bool checkValueWidth(SDValue V, unsigned width, ISD::LoadExtType &ExtType) {
LoadSDNode *LoadNode = cast<LoadSDNode>(V.getNode());
if ((LoadNode->getMemoryVT() == MVT::i8 && width == 8)
|| (LoadNode->getMemoryVT() == MVT::i16 && width == 16)) {
+ if (LoadNode->getExtensionType() != ISD::SEXTLOAD &&
+ LoadNode->getExtensionType() != ISD::ZEXTLOAD)
+ return false;
ExtType = LoadNode->getExtensionType();
return true;
}
diff --git a/llvm/test/CodeGen/AArch64/pr215839.ll b/llvm/test/CodeGen/AArch64/pr215839.ll
new file mode 100644
index 0000000000000..2671057f2166e
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/pr215839.ll
@@ -0,0 +1,152 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=aarch64 %s -o - | FileCheck %s
+
+define i32 @reduced(<4 x i16> %vecins, i32 %conv) {
+; CHECK-LABEL: reduced:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: sub sp, sp, #16
+; CHECK-NEXT: .cfi_def_cfa_offset 16
+; CHECK-NEXT: add x8, sp, #8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: mov w9, #26065 // =0x65d1
+; CHECK-NEXT: mov w10, #9878 // =0x2696
+; CHECK-NEXT: bfi x8, x0, #1, #2
+; CHECK-NEXT: .LBB0_1: // %lbl_b59
+; CHECK-NEXT: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: str d0, [sp, #8]
+; CHECK-NEXT: ldrh w11, [x8]
+; CHECK-NEXT: add w11, w11, #1
+; CHECK-NEXT: and w11, w11, #0xffff
+; CHECK-NEXT: cmp w11, w9
+; CHECK-NEXT: ccmp w11, w10, #4, ne
+; CHECK-NEXT: b.ne .LBB0_1
+; CHECK-NEXT: // %bb.2: // %sw.bb
+; CHECK-NEXT: mov w0, wzr
+; CHECK-NEXT: add sp, sp, #16
+; CHECK-NEXT: ret
+entry:
+ br label %lbl_b59
+
+lbl_b59: ; preds = %lbl_b59, %entry
+ %vecext = extractelement <4 x i16> %vecins, i32 %conv
+ %add4 = add i16 %vecext, 1
+ switch i16 %add4, label %lbl_b59 [
+ i16 9878, label %sw.bb
+ i16 26065, label %sw.bb
+ ]
+
+sw.bb: ; preds = %lbl_b59, %lbl_b59
+ ret i32 0
+}
+
+
+@g1 = dso_local local_unnamed_addr global i32 1662431956, align 4
+@g30 = dso_local local_unnamed_addr global i16 0, align 4
+@g19 = dso_local local_unnamed_addr global i8 0, align 4
+@f29_c9 = dso_local local_unnamed_addr global i8 0, align 1
+
+define i32 @original() {
+; CHECK-LABEL: original:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 16
+; CHECK-NEXT: .cfi_offset w30, -16
+; CHECK-NEXT: adrp x8, g1
+; CHECK-NEXT: mov x10, #4611686018427387904 // =0x4000000000000000
+; CHECK-NEXT: mov w14, #24726 // =0x6096
+; CHECK-NEXT: ldr w8, [x8, :lo12:g1]
+; CHECK-NEXT: mov w11, #26065 // =0x65d1
+; CHECK-NEXT: mov w12, #9878 // =0x2696
+; CHECK-NEXT: add x8, x8, x8
+; CHECK-NEXT: lsl x9, x8, #31
+; CHECK-NEXT: cmp x8, x9, asr #31
+; CHECK-NEXT: add x8, x9, x10
+; CHECK-NEXT: mov w9, #860684288 // =0x334d0000
+; CHECK-NEXT: fmov d0, x9
+; CHECK-NEXT: ccmp x8, #0, #8, eq
+; CHECK-NEXT: adrp x8, g19
+; CHECK-NEXT: cset w9, mi
+; CHECK-NEXT: mov w10, #-28618 // =0xffff9036
+; CHECK-NEXT: .LBB1_1: // %lbl_b59
+; CHECK-NEXT: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: ext v1.8b, v0.8b, v0.8b, #6
+; CHECK-NEXT: trn1 v0.4h, v0.4h, v1.4h
+; CHECK-NEXT: fmov x13, d0
+; CHECK-NEXT: lsr x13, x13, #61
+; CHECK-NEXT: cmp x13, #1
+; CHECK-NEXT: rev16 w13, w14
+; CHECK-NEXT: b.ne .LBB1_4
+; CHECK-NEXT: // %bb.2: // %lbl_br68
+; CHECK-NEXT: // in Loop: Header=BB1_1 Depth=1
+; CHECK-NEXT: mov v0.h[0], w13
+; CHECK-NEXT: and w14, w13, #0xffff
+; CHECK-NEXT: cmp w9, #0
+; CHECK-NEXT: csel w14, wzr, w14, ne
+; CHECK-NEXT: add x15, sp, #8
+; CHECK-NEXT: bfi x15, x14, #1, #2
+; CHECK-NEXT: str d0, [sp, #8]
+; CHECK-NEXT: ldrsh w15, [x15]
+; CHECK-NEXT: add w14, w15, w10
+; CHECK-NEXT: cmp w15, #1
+; CHECK-NEXT: and w15, w14, #0xffff
+; CHECK-NEXT: cset w16, lt
+; CHECK-NEXT: cmp w15, w11
+; CHECK-NEXT: strb w16, [x8, :lo12:g19]
+; CHECK-NEXT: ccmp w15, w12, #4, ne
+; CHECK-NEXT: b.ne .LBB1_1
+; CHECK-NEXT: // %bb.3: // %sw.bb
+; CHECK-NEXT: adrp x8, g30
+; CHECK-NEXT: mov w0, wzr
+; CHECK-NEXT: strh w13, [x8, :lo12:g30]
+; CHECK-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+; CHECK-NEXT: .LBB1_4: // %if.end
+; CHECK-NEXT: adrp x8, g30
+; CHECK-NEXT: strh w13, [x8, :lo12:g30]
+; CHECK-NEXT: bl abort
+entry:
+ %0 = load i32, ptr @g1, align 4
+ %1 = zext i32 %0 to i33
+ %2 = tail call { i33, i1 } @llvm.sadd.with.overflow.i33(i33 %1, i33 %1)
+ %3 = extractvalue { i33, i1 } %2, 1
+ %4 = extractvalue { i33, i1 } %2, 0
+ %5 = add i33 %4, 2147483648
+ %6 = icmp slt i33 %5, 0
+ %7 = or i1 %3, %6
+ br label %lbl_b59
+
+lbl_b59: ; preds = %lbl_br68, %entry
+ %v10.0 = phi i16 [ 24726, %entry ], [ %add, %lbl_br68 ]
+ %bc4.0 = phi <4 x i16> [ <i16 0, i16 13133, i16 0, i16 0>, %entry ], [ %vecins, %lbl_br68 ]
+ %8 = tail call i16 @llvm.bswap.i16(i16 %v10.0)
+ %shuffle = shufflevector <4 x i16> %bc4.0, <4 x i16> poison, <4 x i32> <i32 0, i32 3, i32 2, i32 1>
+ %9 = bitcast <4 x i16> %shuffle to i64
+ %10 = and i64 %9, -2305843009213693952
+ %cmp = icmp eq i64 %10, 2305843009213693952
+ br i1 %cmp, label %lbl_br68, label %if.end
+
+if.end: ; preds = %lbl_b59
+ store i16 %8, ptr @g30, align 4
+ tail call void @abort() #3
+ unreachable
+
+lbl_br68: ; preds = %lbl_b59
+ %vecins = insertelement <4 x i16> %shuffle, i16 %8, i64 0
+ %conv = zext i16 %8 to i32
+ %cond = select i1 %7, i32 0, i32 %conv
+ %vecext = extractelement <4 x i16> %vecins, i32 %cond
+ %cmp2 = icmp slt i16 %vecext, 1
+ %storedv = zext i1 %cmp2 to i8
+ store i8 %storedv, ptr @g19, align 4
+ %add = add i16 %vecext, -28618
+ switch i16 %add, label %lbl_b59 [
+ i16 9878, label %sw.bb
+ i16 26065, label %sw.bb
+ ]
+
+sw.bb: ; preds = %lbl_br68, %lbl_br68
+ store i16 %8, ptr @g30, align 4
+ ret i32 0
+}
+
+declare void @abort()
|
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
3 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.
Backport 90c0c67
Requested by: @davemgreen