[ExpandMemCmp] Improve memcmp optimisation for boolean results #71221
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
|
@llvm/pr-subscribers-backend-x86 @llvm/pr-subscribers-backend-aarch64 Author: Igor Kirillov (igogo-x86) ChangesThis patch enhances the optimization of memcmp calls when only two outcomes Previously, LLVM would generate unnecessary operations even when the user of Patch is 52.15 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/71221.diff 6 Files Affected:
diff --git a/llvm/lib/CodeGen/ExpandMemCmp.cpp b/llvm/lib/CodeGen/ExpandMemCmp.cpp
index 28e258be226a695..a3dd0feea2ff969 100644
--- a/llvm/lib/CodeGen/ExpandMemCmp.cpp
+++ b/llvm/lib/CodeGen/ExpandMemCmp.cpp
@@ -23,6 +23,7 @@
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/PatternMatch.h"
#include "llvm/InitializePasses.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
@@ -31,6 +32,7 @@
#include <optional>
using namespace llvm;
+using namespace llvm::PatternMatch;
namespace llvm {
class TargetLowering;
@@ -656,6 +658,37 @@ Value *MemCmpExpansion::getMemCmpOneBlock() {
const LoadPair Loads = getLoadPair(LoadSizeType, BSwapSizeType, MaxLoadType,
/*Offset*/ 0);
+
+ // If a user of memcmp cares only about two outcomes, for example:
+ // bool result = memcmp(a, b, NBYTES) > 0;
+ // We can generate more optimal code with a smaller number of operations
+ if (auto *U = CI->getUniqueUndroppableUser()) {
+ auto *UI = cast<Instruction>(U);
+ ICmpInst::Predicate Pred = ICmpInst::Predicate::BAD_ICMP_PREDICATE;
+ uint64_t Shift;
+ bool NeedsZExt = false;
+ // This is a special case because instead of checking if the result is less than zero:
+ // bool result = memcmp(a, b, NBYTES) < 0;
+ // Compiler is clever enough to generate the following code:
+ // bool result = memcmp(a, b, NBYTES) >> 31;
+ if (match(UI, m_LShr(m_Value(), m_ConstantInt(Shift))) && Shift == CI->getType()->getIntegerBitWidth() - 1) {
+ Pred = ICmpInst::ICMP_SLT;
+ NeedsZExt = true;
+ } else {
+ // In case of a successful match this call will set `Pred` variable
+ match(UI, m_ICmp(Pred, m_Specific(CI), m_Zero()));
+ }
+ // Generate new code and remove the original memcmp call and the user
+ if (ICmpInst::isSigned(Pred)) {
+ Value *Cmp = Builder.CreateICmp(CmpInst::getUnsignedPredicate(Pred), Loads.Lhs, Loads.Rhs);
+ auto *Result = NeedsZExt ? Builder.CreateZExt(Cmp, U->getType()) : Cmp;
+ UI->replaceAllUsesWith(Result);
+ UI->eraseFromParent();
+ CI->eraseFromParent();
+ return nullptr;
+ }
+ }
+
// The result of memcmp is negative, zero, or positive, so produce that by
// subtracting 2 extended compare bits: sub (ugt, ult).
// If a target prefers to use selects to get -1/0/1, they should be able
@@ -670,7 +703,7 @@ Value *MemCmpExpansion::getMemCmpOneBlock() {
}
// This function expands the memcmp call into an inline expansion and returns
-// the memcmp result.
+// the memcmp result. Returns nullptr if the memcmp is already replaced.
Value *MemCmpExpansion::getMemCmpExpansion() {
// Create the basic block framework for a multi-block expansion.
if (getNumBlocks() != 1) {
@@ -838,11 +871,11 @@ static bool expandMemCmp(CallInst *CI, const TargetTransformInfo *TTI,
NumMemCmpInlined++;
- Value *Res = Expansion.getMemCmpExpansion();
-
- // Replace call with result of expansion and erase call.
- CI->replaceAllUsesWith(Res);
- CI->eraseFromParent();
+ if (Value *Res = Expansion.getMemCmpExpansion()) {
+ // Replace call with result of expansion and erase call.
+ CI->replaceAllUsesWith(Res);
+ CI->eraseFromParent();
+ }
return true;
}
diff --git a/llvm/test/CodeGen/AArch64/memcmp.ll b/llvm/test/CodeGen/AArch64/memcmp.ll
index d13a416a28761ca..4da7c8c95a4e4f0 100644
--- a/llvm/test/CodeGen/AArch64/memcmp.ll
+++ b/llvm/test/CodeGen/AArch64/memcmp.ll
@@ -222,16 +222,28 @@ define i1 @length4_lt(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev w8, w8
; CHECK-NEXT: rev w9, w9
; CHECK-NEXT: cmp w8, w9
-; CHECK-NEXT: cset w8, hi
-; CHECK-NEXT: cset w9, lo
-; CHECK-NEXT: sub w8, w8, w9
-; CHECK-NEXT: lsr w0, w8, #31
+; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%m = tail call i32 @memcmp(ptr %X, ptr %Y, i64 4) nounwind
%c = icmp slt i32 %m, 0
ret i1 %c
}
+define i32 @length4_lt_32(ptr %X, ptr %Y) nounwind {
+; CHECK-LABEL: length4_lt_32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldr w8, [x0]
+; CHECK-NEXT: ldr w9, [x1]
+; CHECK-NEXT: rev w8, w8
+; CHECK-NEXT: rev w9, w9
+; CHECK-NEXT: cmp w8, w9
+; CHECK-NEXT: cset w0, lo
+; CHECK-NEXT: ret
+ %m = tail call i32 @memcmp(ptr %X, ptr %Y, i64 4) nounwind
+ %c = lshr i32 %m, 31
+ ret i32 %c
+}
+
define i1 @length4_gt(ptr %X, ptr %Y) nounwind {
; CHECK-LABEL: length4_gt:
; CHECK: // %bb.0:
@@ -240,11 +252,7 @@ define i1 @length4_gt(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev w8, w8
; CHECK-NEXT: rev w9, w9
; CHECK-NEXT: cmp w8, w9
-; CHECK-NEXT: cset w8, hi
-; CHECK-NEXT: cset w9, lo
-; CHECK-NEXT: sub w8, w8, w9
-; CHECK-NEXT: cmp w8, #0
-; CHECK-NEXT: cset w0, gt
+; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%m = tail call i32 @memcmp(ptr %X, ptr %Y, i64 4) nounwind
%c = icmp sgt i32 %m, 0
@@ -313,10 +321,7 @@ define i1 @length5_lt(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: cset w8, hi
-; CHECK-NEXT: cset w9, lo
-; CHECK-NEXT: sub w8, w8, w9
-; CHECK-NEXT: lsr w0, w8, #31
+; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%m = tail call i32 @memcmp(ptr %X, ptr %Y, i64 5) nounwind
%c = icmp slt i32 %m, 0
@@ -343,6 +348,25 @@ define i32 @length6(ptr %X, ptr %Y) nounwind {
ret i32 %m
}
+define i32 @length6_lt(ptr %X, ptr %Y) nounwind {
+; CHECK-LABEL: length6_lt:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldrh w8, [x0, #4]
+; CHECK-NEXT: ldr w9, [x0]
+; CHECK-NEXT: ldrh w10, [x1, #4]
+; CHECK-NEXT: ldr w11, [x1]
+; CHECK-NEXT: orr x8, x9, x8, lsl #32
+; CHECK-NEXT: orr x9, x11, x10, lsl #32
+; CHECK-NEXT: rev x8, x8
+; CHECK-NEXT: rev x9, x9
+; CHECK-NEXT: cmp x8, x9
+; CHECK-NEXT: cset w0, lo
+; CHECK-NEXT: ret
+ %m = tail call i32 @memcmp(ptr %X, ptr %Y, i64 6) nounwind
+ %r = lshr i32 %m, 31
+ ret i32 %r
+}
+
define i32 @length7(ptr %X, ptr %Y) nounwind {
; CHECK-LABEL: length7:
; CHECK: // %bb.0:
@@ -351,18 +375,18 @@ define i32 @length7(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev w8, w8
; CHECK-NEXT: rev w9, w9
; CHECK-NEXT: cmp w8, w9
-; CHECK-NEXT: b.ne .LBB22_3
+; CHECK-NEXT: b.ne .LBB24_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldur w8, [x0, #3]
; CHECK-NEXT: ldur w9, [x1, #3]
; CHECK-NEXT: rev w8, w8
; CHECK-NEXT: rev w9, w9
; CHECK-NEXT: cmp w8, w9
-; CHECK-NEXT: b.ne .LBB22_3
+; CHECK-NEXT: b.ne .LBB24_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB22_3: // %res_block
+; CHECK-NEXT: .LBB24_3: // %res_block
; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -379,18 +403,18 @@ define i1 @length7_lt(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev w8, w8
; CHECK-NEXT: rev w9, w9
; CHECK-NEXT: cmp w8, w9
-; CHECK-NEXT: b.ne .LBB23_3
+; CHECK-NEXT: b.ne .LBB25_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldur w8, [x0, #3]
; CHECK-NEXT: ldur w9, [x1, #3]
; CHECK-NEXT: rev w8, w8
; CHECK-NEXT: rev w9, w9
; CHECK-NEXT: cmp w8, w9
-; CHECK-NEXT: b.ne .LBB23_3
+; CHECK-NEXT: b.ne .LBB25_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: lsr w0, wzr, #31
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB23_3: // %res_block
+; CHECK-NEXT: .LBB25_3: // %res_block
; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w8, w8, hs
@@ -470,13 +494,13 @@ define i32 @length9(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB28_2
+; CHECK-NEXT: b.ne .LBB30_2
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldrb w8, [x0, #8]
; CHECK-NEXT: ldrb w9, [x1, #8]
; CHECK-NEXT: sub w0, w8, w9
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB28_2: // %res_block
+; CHECK-NEXT: .LBB30_2: // %res_block
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
; CHECK-NEXT: ret
@@ -508,7 +532,7 @@ define i32 @length10(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB30_3
+; CHECK-NEXT: b.ne .LBB32_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldrh w8, [x0, #8]
; CHECK-NEXT: ldrh w9, [x1, #8]
@@ -517,11 +541,11 @@ define i32 @length10(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: lsr w8, w8, #16
; CHECK-NEXT: lsr w9, w9, #16
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB30_3
+; CHECK-NEXT: b.ne .LBB32_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB30_3: // %res_block
+; CHECK-NEXT: .LBB32_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -554,18 +578,18 @@ define i32 @length11(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB32_3
+; CHECK-NEXT: b.ne .LBB34_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldur x8, [x0, #3]
; CHECK-NEXT: ldur x9, [x1, #3]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB32_3
+; CHECK-NEXT: b.ne .LBB34_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB32_3: // %res_block
+; CHECK-NEXT: .LBB34_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -614,18 +638,18 @@ define i32 @length12(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB35_3
+; CHECK-NEXT: b.ne .LBB37_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr w8, [x0, #8]
; CHECK-NEXT: ldr w9, [x1, #8]
; CHECK-NEXT: rev w8, w8
; CHECK-NEXT: rev w9, w9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB35_3
+; CHECK-NEXT: b.ne .LBB37_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB35_3: // %res_block
+; CHECK-NEXT: .LBB37_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -674,18 +698,18 @@ define i32 @length15(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB38_3
+; CHECK-NEXT: b.ne .LBB40_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldur x8, [x0, #7]
; CHECK-NEXT: ldur x9, [x1, #7]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB38_3
+; CHECK-NEXT: b.ne .LBB40_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB38_3: // %res_block
+; CHECK-NEXT: .LBB40_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -702,18 +726,18 @@ define i1 @length15_lt(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB39_3
+; CHECK-NEXT: b.ne .LBB41_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldur x8, [x0, #7]
; CHECK-NEXT: ldur x9, [x1, #7]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB39_3
+; CHECK-NEXT: b.ne .LBB41_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: lsr w0, wzr, #31
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB39_3: // %res_block
+; CHECK-NEXT: .LBB41_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w8, w8, hs
@@ -734,7 +758,7 @@ define i32 @length15_const(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: movk x8, #12594, lsl #48
; CHECK-NEXT: cmp x9, x8
-; CHECK-NEXT: b.ne .LBB40_3
+; CHECK-NEXT: b.ne .LBB42_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: mov x8, #13365 // =0x3435
; CHECK-NEXT: ldur x9, [x0, #7]
@@ -743,11 +767,11 @@ define i32 @length15_const(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: movk x8, #14393, lsl #48
; CHECK-NEXT: cmp x9, x8
-; CHECK-NEXT: b.ne .LBB40_3
+; CHECK-NEXT: b.ne .LBB42_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB40_3: // %res_block
+; CHECK-NEXT: .LBB42_3: // %res_block
; CHECK-NEXT: cmp x9, x8
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -782,7 +806,7 @@ define i1 @length15_gt_const(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: movk x8, #12594, lsl #48
; CHECK-NEXT: cmp x9, x8
-; CHECK-NEXT: b.ne .LBB42_3
+; CHECK-NEXT: b.ne .LBB44_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: mov x8, #13365 // =0x3435
; CHECK-NEXT: ldur x9, [x0, #7]
@@ -791,15 +815,15 @@ define i1 @length15_gt_const(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: movk x8, #14393, lsl #48
; CHECK-NEXT: cmp x9, x8
-; CHECK-NEXT: b.ne .LBB42_3
+; CHECK-NEXT: b.ne .LBB44_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: b .LBB42_4
-; CHECK-NEXT: .LBB42_3: // %res_block
+; CHECK-NEXT: b .LBB44_4
+; CHECK-NEXT: .LBB44_3: // %res_block
; CHECK-NEXT: cmp x9, x8
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w8, w8, hs
-; CHECK-NEXT: .LBB42_4: // %endblock
+; CHECK-NEXT: .LBB44_4: // %endblock
; CHECK-NEXT: cmp w8, #0
; CHECK-NEXT: cset w0, gt
; CHECK-NEXT: ret
@@ -817,18 +841,18 @@ define i32 @length16(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB43_3
+; CHECK-NEXT: b.ne .LBB45_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x1, #8]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB43_3
+; CHECK-NEXT: b.ne .LBB45_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB43_3: // %res_block
+; CHECK-NEXT: .LBB45_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -859,18 +883,18 @@ define i1 @length16_lt(ptr %x, ptr %y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB45_3
+; CHECK-NEXT: b.ne .LBB47_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x1, #8]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB45_3
+; CHECK-NEXT: b.ne .LBB47_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: lsr w0, wzr, #31
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB45_3: // %res_block
+; CHECK-NEXT: .LBB47_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w8, w8, hs
@@ -889,22 +913,22 @@ define i1 @length16_gt(ptr %x, ptr %y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB46_3
+; CHECK-NEXT: b.ne .LBB48_3
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x1, #8]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB46_3
+; CHECK-NEXT: b.ne .LBB48_3
; CHECK-NEXT: // %bb.2:
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: b .LBB46_4
-; CHECK-NEXT: .LBB46_3: // %res_block
+; CHECK-NEXT: b .LBB48_4
+; CHECK-NEXT: .LBB48_3: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w8, w8, hs
-; CHECK-NEXT: .LBB46_4: // %endblock
+; CHECK-NEXT: .LBB48_4: // %endblock
; CHECK-NEXT: cmp w8, #0
; CHECK-NEXT: cset w0, gt
; CHECK-NEXT: ret
@@ -943,25 +967,25 @@ define i32 @length24(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB48_4
+; CHECK-NEXT: b.ne .LBB50_4
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x1, #8]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB48_4
+; CHECK-NEXT: b.ne .LBB50_4
; CHECK-NEXT: // %bb.2: // %loadbb2
; CHECK-NEXT: ldr x8, [x0, #16]
; CHECK-NEXT: ldr x9, [x1, #16]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB48_4
+; CHECK-NEXT: b.ne .LBB50_4
; CHECK-NEXT: // %bb.3:
; CHECK-NEXT: mov w0, wzr
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB48_4: // %res_block
+; CHECK-NEXT: .LBB50_4: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w0, w8, hs
@@ -995,25 +1019,25 @@ define i1 @length24_lt(ptr %x, ptr %y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB50_4
+; CHECK-NEXT: b.ne .LBB52_4
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x1, #8]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB50_4
+; CHECK-NEXT: b.ne .LBB52_4
; CHECK-NEXT: // %bb.2: // %loadbb2
; CHECK-NEXT: ldr x8, [x0, #16]
; CHECK-NEXT: ldr x9, [x1, #16]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB50_4
+; CHECK-NEXT: b.ne .LBB52_4
; CHECK-NEXT: // %bb.3:
; CHECK-NEXT: lsr w0, wzr, #31
; CHECK-NEXT: ret
-; CHECK-NEXT: .LBB50_4: // %res_block
+; CHECK-NEXT: .LBB52_4: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w8, w8, hs
@@ -1032,29 +1056,29 @@ define i1 @length24_gt(ptr %x, ptr %y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB51_4
+; CHECK-NEXT: b.ne .LBB53_4
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x1, #8]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB51_4
+; CHECK-NEXT: b.ne .LBB53_4
; CHECK-NEXT: // %bb.2: // %loadbb2
; CHECK-NEXT: ldr x8, [x0, #16]
; CHECK-NEXT: ldr x9, [x1, #16]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB51_4
+; CHECK-NEXT: b.ne .LBB53_4
; CHECK-NEXT: // %bb.3:
; CHECK-NEXT: mov w8, wzr
-; CHECK-NEXT: b .LBB51_5
-; CHECK-NEXT: .LBB51_4: // %res_block
+; CHECK-NEXT: b .LBB53_5
+; CHECK-NEXT: .LBB53_4: // %res_block
; CHECK-NEXT: cmp x8, x9
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cneg w8, w8, hs
-; CHECK-NEXT: .LBB51_5: // %endblock
+; CHECK-NEXT: .LBB53_5: // %endblock
; CHECK-NEXT: cmp w8, #0
; CHECK-NEXT: cset w0, gt
; CHECK-NEXT: ret
@@ -1098,32 +1122,32 @@ define i32 @length31(ptr %X, ptr %Y) nounwind {
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB53_5
+; CHECK-NEXT: b.ne .LBB55_5
; CHECK-NEXT: // %bb.1: // %loadbb1
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x1, #8]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB53_5
+; CHECK-NEXT: b.ne .LBB55_5
; CHECK-NEXT: // %bb.2: // %loadbb2
; CHECK-NEXT: ldr x8, [x0, #16]
; CHECK-NEXT: ldr x9, [x1, #16]
; CHECK-NEXT: rev x8, x8
; CHECK-NEXT: rev x9, x9
; CHECK-NEXT: cmp x8, x9
-; CHECK-NEXT: b.ne .LBB53_5
+; CHECK-NEXT: b.ne .LBB55_5
; CHECK-NEXT: // %bb.3: // %loadbb3
; CHECK-NEXT: ldur x8, ...
[truncated]
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
nikic
reviewed
Nov 3, 2023
david-arm
reviewed
Nov 6, 2023
There was a problem hiding this comment.
Thanks for this @igogo-x86! It looks good - I just have a couple of minor comments.
igogo-x86
force-pushed
the
expand-memcmp-two-outcome-comparison
branch
from
7ce3e79 to
6974602
Compare
david-arm
approved these changes
Nov 6, 2023
david-arm
approved these changes
Nov 6, 2023
This patch enhances the optimization of memcmp calls when only two outcomes are needed and comparison fits into one block, for example: bool result = memcmp(a, b, 6) > 0; Previously, LLVM would generate unnecessary operations even when the user of memcmp was only interested in a binary outcome.
igogo-x86
force-pushed
the
expand-memcmp-two-outcome-comparison
branch
from
8160d60 to
9fb8ac9
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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 patch enhances the optimization of memcmp calls when only two outcomes
are needed and comparison fits into one block, for example:
Previously, LLVM would generate unnecessary operations even when the user of
memcmp was only interested in a binary outcome.