Skip to content

[ExpandMemCmp] Recognize canonical form of (icmp sle/sge X, 0) in getMemCmpOneBlock. #121540

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 4 commits into from
Jan 3, 2025
Merged

[ExpandMemCmp] Recognize canonical form of (icmp sle/sge X, 0) in getMemCmpOneBlock. #121540

merged 4 commits into from
Jan 3, 2025

Conversation

topperc
Copy link
Collaborator

@topperc topperc commented Jan 3, 2025 •
edited
Loading

This code recognizes special cases where the result of memcmp is
compared with 0. If the compare is sle/sge, then InstCombine
canonicalizes to (icmp slt X, 1) or (icmp sgt X, -1). We should
recognize those patterns too.

topperc added 2 commits January 2, 2025 17:57
@topperc
Add tests.
a885001
@topperc
[ExpandMemCmp] Recognize canonical form of (icmp sle/sge X, 0) in get…
0d562a0 
…MemCmpOneBlock.

This code recognizes special cases where the result of memcmp is
compared with 0. If the compare is sle/sge, then InstCombine
canonicalizes to (icmp slt X, 1) or (icmp sgt X, -1).

I'm not sure if this is the best fix. After llvm#121530 we will emit
a ucmp in the default case instead of (sub (zext (icmp ugt)), (zext (icmp ult))).
With that it would be easy to optimize (setgt (ucmp X, Y), -1) in DAGCombine.

Tests have not been pre-committed yet.
@topperc topperc requested review from nikic, RKSimon and dtcxzyw January 3, 2025 03:36
@llvmbot
Copy link
Member

llvmbot commented Jan 3, 2025 •
edited
Loading

@llvm/pr-subscribers-backend-x86

@llvm/pr-subscribers-backend-aarch64

Author: Craig Topper (topperc)

Changes

This code recognizes special cases where the result of memcmp is
compared with 0. If the compare is sle/sge, then InstCombine
canonicalizes to (icmp slt X, 1) or (icmp sgt X, -1). We should
recognize those patterns too.

I'm not sure if this is the best fix. After #121530 we will emit
a ucmp in the default case instead of (sub (zext (icmp ugt)), (zext (icmp ult))).
With that it would be easy to optimize (setgt (ucmp X, Y), -1) in DAGCombine.

Tests have not been pre-committed yet.

Patch is 145.21 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121540.diff

4 Files Affected:

  • (modified) llvm/lib/CodeGen/ExpandMemCmp.cpp (+13-3)
  • (modified) llvm/test/CodeGen/AArch64/memcmp.ll (+201-171)
  • (modified) llvm/test/CodeGen/RISCV/memcmp.ll (+756-170)
  • (modified) llvm/test/CodeGen/X86/memcmp.ll (+74-44)
diff --git a/llvm/lib/CodeGen/ExpandMemCmp.cpp b/llvm/lib/CodeGen/ExpandMemCmp.cpp
index f8ca7e370f6ef9..fc832c3e668ea8 100644
--- a/llvm/lib/CodeGen/ExpandMemCmp.cpp
+++ b/llvm/lib/CodeGen/ExpandMemCmp.cpp
@@ -668,7 +668,7 @@ Value *MemCmpExpansion::getMemCmpOneBlock() {
   // We can generate more optimal code with a smaller number of operations
   if (CI->hasOneUser()) {
     auto *UI = cast<Instruction>(*CI->user_begin());
-    CmpPredicate Pred = ICmpInst::Predicate::BAD_ICMP_PREDICATE;
+    CmpPredicate Pred;
     uint64_t Shift;
     bool NeedsZExt = false;
     // This is a special case because instead of checking if the result is less
@@ -680,9 +680,19 @@ Value *MemCmpExpansion::getMemCmpOneBlock() {
         Shift == (CI->getType()->getIntegerBitWidth() - 1)) {
       Pred = ICmpInst::ICMP_SLT;
       NeedsZExt = true;
+    } else if (match(UI, m_ICmp(Pred, m_Specific(CI), m_Zero()))) {
+      // Compare with 0. Pred is already set.
+    } else if (match(UI, m_ICmp(Pred, m_Specific(CI), m_AllOnes())) &&
+               Pred == ICmpInst::ICMP_SGT) {
+      // Adjust predicate as if it compared with 0.
+      Pred = ICmpInst::ICMP_SGE;
+    } else if (match(UI, m_ICmp(Pred, m_Specific(CI), m_One())) &&
+               Pred == ICmpInst::ICMP_SLT) {
+      // Adjust predicate as if it compared with 0.
+      Pred = ICmpInst::ICMP_SLE;
     } else {
-      // In case of a successful match this call will set `Pred` variable
-      match(UI, m_ICmp(Pred, m_Specific(CI), m_Zero()));
+      // No match, invalidate Pred.
+      Pred = ICmpInst::Predicate::BAD_ICMP_PREDICATE;
     }
     // Generate new code and remove the original memcmp call and the user
     if (ICmpInst::isSigned(Pred)) {
diff --git a/llvm/test/CodeGen/AArch64/memcmp.ll b/llvm/test/CodeGen/AArch64/memcmp.ll
index 4da7c8c95a4e4f..e08cd32560b5f2 100644
--- a/llvm/test/CodeGen/AArch64/memcmp.ll
+++ b/llvm/test/CodeGen/AArch64/memcmp.ll
@@ -259,6 +259,36 @@ define i1 @length4_gt(ptr %X, ptr %Y) nounwind {
   ret i1 %c
 }
 
+define i1 @length4_le(ptr %X, ptr %Y) nounwind {
+; CHECK-LABEL: length4_le:
+; 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, ls
+; CHECK-NEXT:    ret
+  %m = tail call i32 @memcmp(ptr %X, ptr %Y, i64 4) nounwind
+  %c = icmp slt i32 %m, 1
+  ret i1 %c
+}
+
+define i1 @length4_ge(ptr %X, ptr %Y) nounwind {
+; CHECK-LABEL: length4_ge:
+; 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, hs
+; CHECK-NEXT:    ret
+  %m = tail call i32 @memcmp(ptr %X, ptr %Y, i64 4) nounwind
+  %c = icmp sgt i32 %m, -1
+  ret i1 %c
+}
+
 define i1 @length4_eq_const(ptr %X) nounwind {
 ; CHECK-LABEL: length4_eq_const:
 ; CHECK:       // %bb.0:
@@ -375,18 +405,18 @@ 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 .LBB24_3
+; CHECK-NEXT:    b.ne .LBB26_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 .LBB24_3
+; CHECK-NEXT:    b.ne .LBB26_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    mov w0, wzr
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB24_3: // %res_block
+; CHECK-NEXT:  .LBB26_3: // %res_block
 ; CHECK-NEXT:    cmp w8, w9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w0, w8, hs
@@ -403,18 +433,18 @@ 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 .LBB25_3
+; CHECK-NEXT:    b.ne .LBB27_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 .LBB25_3
+; CHECK-NEXT:    b.ne .LBB27_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    lsr w0, wzr, #31
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB25_3: // %res_block
+; CHECK-NEXT:  .LBB27_3: // %res_block
 ; CHECK-NEXT:    cmp w8, w9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
@@ -494,13 +524,13 @@ 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 .LBB30_2
+; CHECK-NEXT:    b.ne .LBB32_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:  .LBB30_2: // %res_block
+; CHECK-NEXT:  .LBB32_2: // %res_block
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w0, w8, hs
 ; CHECK-NEXT:    ret
@@ -532,7 +562,7 @@ 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 .LBB32_3
+; CHECK-NEXT:    b.ne .LBB34_3
 ; CHECK-NEXT:  // %bb.1: // %loadbb1
 ; CHECK-NEXT:    ldrh w8, [x0, #8]
 ; CHECK-NEXT:    ldrh w9, [x1, #8]
@@ -541,11 +571,11 @@ 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 .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
@@ -578,18 +608,18 @@ 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 .LBB34_3
+; CHECK-NEXT:    b.ne .LBB36_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 .LBB34_3
+; CHECK-NEXT:    b.ne .LBB36_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    mov w0, wzr
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB34_3: // %res_block
+; CHECK-NEXT:  .LBB36_3: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w0, w8, hs
@@ -638,18 +668,18 @@ 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 .LBB37_3
+; CHECK-NEXT:    b.ne .LBB39_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 .LBB37_3
+; CHECK-NEXT:    b.ne .LBB39_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    mov w0, wzr
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB37_3: // %res_block
+; CHECK-NEXT:  .LBB39_3: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w0, w8, hs
@@ -698,18 +728,18 @@ 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 .LBB40_3
+; CHECK-NEXT:    b.ne .LBB42_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 .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 x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w0, w8, hs
@@ -726,18 +756,18 @@ 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 .LBB41_3
+; CHECK-NEXT:    b.ne .LBB43_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 .LBB41_3
+; CHECK-NEXT:    b.ne .LBB43_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    lsr w0, wzr, #31
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB41_3: // %res_block
+; CHECK-NEXT:  .LBB43_3: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
@@ -758,7 +788,7 @@ 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 .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]
@@ -767,11 +797,11 @@ 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 .LBB42_3
+; CHECK-NEXT:    b.ne .LBB44_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    mov w0, wzr
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB42_3: // %res_block
+; CHECK-NEXT:  .LBB44_3: // %res_block
 ; CHECK-NEXT:    cmp x9, x8
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w0, w8, hs
@@ -806,7 +836,7 @@ 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 .LBB44_3
+; CHECK-NEXT:    b.ne .LBB46_3
 ; CHECK-NEXT:  // %bb.1: // %loadbb1
 ; CHECK-NEXT:    mov x8, #13365 // =0x3435
 ; CHECK-NEXT:    ldur x9, [x0, #7]
@@ -815,15 +845,15 @@ 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 .LBB44_3
+; CHECK-NEXT:    b.ne .LBB46_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    mov w8, wzr
-; CHECK-NEXT:    b .LBB44_4
-; CHECK-NEXT:  .LBB44_3: // %res_block
+; CHECK-NEXT:    b .LBB46_4
+; CHECK-NEXT:  .LBB46_3: // %res_block
 ; CHECK-NEXT:    cmp x9, x8
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
-; CHECK-NEXT:  .LBB44_4: // %endblock
+; CHECK-NEXT:  .LBB46_4: // %endblock
 ; CHECK-NEXT:    cmp w8, #0
 ; CHECK-NEXT:    cset w0, gt
 ; CHECK-NEXT:    ret
@@ -841,18 +871,18 @@ 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 .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:    mov w0, wzr
 ; 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 w0, w8, hs
@@ -883,18 +913,18 @@ 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 .LBB47_3
+; CHECK-NEXT:    b.ne .LBB49_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 .LBB47_3
+; CHECK-NEXT:    b.ne .LBB49_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    lsr w0, wzr, #31
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB47_3: // %res_block
+; CHECK-NEXT:  .LBB49_3: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
@@ -913,22 +943,22 @@ 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 .LBB48_3
+; CHECK-NEXT:    b.ne .LBB50_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 .LBB48_3
+; CHECK-NEXT:    b.ne .LBB50_3
 ; CHECK-NEXT:  // %bb.2:
 ; CHECK-NEXT:    mov w8, wzr
-; CHECK-NEXT:    b .LBB48_4
-; CHECK-NEXT:  .LBB48_3: // %res_block
+; CHECK-NEXT:    b .LBB50_4
+; CHECK-NEXT:  .LBB50_3: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
-; CHECK-NEXT:  .LBB48_4: // %endblock
+; CHECK-NEXT:  .LBB50_4: // %endblock
 ; CHECK-NEXT:    cmp w8, #0
 ; CHECK-NEXT:    cset w0, gt
 ; CHECK-NEXT:    ret
@@ -967,25 +997,25 @@ 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 .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:    mov w0, wzr
 ; 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 w0, w8, hs
@@ -1019,25 +1049,25 @@ 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 .LBB52_4
+; CHECK-NEXT:    b.ne .LBB54_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 .LBB52_4
+; CHECK-NEXT:    b.ne .LBB54_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 .LBB52_4
+; CHECK-NEXT:    b.ne .LBB54_4
 ; CHECK-NEXT:  // %bb.3:
 ; CHECK-NEXT:    lsr w0, wzr, #31
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB52_4: // %res_block
+; CHECK-NEXT:  .LBB54_4: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
@@ -1056,29 +1086,29 @@ 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 .LBB53_4
+; CHECK-NEXT:    b.ne .LBB55_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 .LBB53_4
+; CHECK-NEXT:    b.ne .LBB55_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 .LBB53_4
+; CHECK-NEXT:    b.ne .LBB55_4
 ; CHECK-NEXT:  // %bb.3:
 ; CHECK-NEXT:    mov w8, wzr
-; CHECK-NEXT:    b .LBB53_5
-; CHECK-NEXT:  .LBB53_4: // %res_block
+; CHECK-NEXT:    b .LBB55_5
+; CHECK-NEXT:  .LBB55_4: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
-; CHECK-NEXT:  .LBB53_5: // %endblock
+; CHECK-NEXT:  .LBB55_5: // %endblock
 ; CHECK-NEXT:    cmp w8, #0
 ; CHECK-NEXT:    cset w0, gt
 ; CHECK-NEXT:    ret
@@ -1122,32 +1152,32 @@ 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 .LBB55_5
+; CHECK-NEXT:    b.ne .LBB57_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 .LBB55_5
+; CHECK-NEXT:    b.ne .LBB57_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 .LBB55_5
+; CHECK-NEXT:    b.ne .LBB57_5
 ; CHECK-NEXT:  // %bb.3: // %loadbb3
 ; CHECK-NEXT:    ldur x8, [x0, #23]
 ; CHECK-NEXT:    ldur x9, [x1, #23]
 ; CHECK-NEXT:    rev x8, x8
 ; CHECK-NEXT:    rev x9, x9
 ; CHECK-NEXT:    cmp x8, x9
-; CHECK-NEXT:    b.ne .LBB55_5
+; CHECK-NEXT:    b.ne .LBB57_5
 ; CHECK-NEXT:  // %bb.4:
 ; CHECK-NEXT:    mov w0, wzr
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB55_5: // %res_block
+; CHECK-NEXT:  .LBB57_5: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w0, w8, hs
@@ -1184,32 +1214,32 @@ define i1 @length31_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 .LBB57_5
+; CHECK-NEXT:    b.ne .LBB59_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 .LBB57_5
+; CHECK-NEXT:    b.ne .LBB59_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 .LBB57_5
+; CHECK-NEXT:    b.ne .LBB59_5
 ; CHECK-NEXT:  // %bb.3: // %loadbb3
 ; CHECK-NEXT:    ldur x8, [x0, #23]
 ; CHECK-NEXT:    ldur x9, [x1, #23]
 ; CHECK-NEXT:    rev x8, x8
 ; CHECK-NEXT:    rev x9, x9
 ; CHECK-NEXT:    cmp x8, x9
-; CHECK-NEXT:    b.ne .LBB57_5
+; CHECK-NEXT:    b.ne .LBB59_5
 ; CHECK-NEXT:  // %bb.4:
 ; CHECK-NEXT:    lsr w0, wzr, #31
 ; CHECK-NEXT:    ret
-; CHECK-NEXT:  .LBB57_5: // %res_block
+; CHECK-NEXT:  .LBB59_5: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
@@ -1228,36 +1258,36 @@ define i1 @length31_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 .LBB58_5
+; CHECK-NEXT:    b.ne .LBB60_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 .LBB58_5
+; CHECK-NEXT:    b.ne .LBB60_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 .LBB58_5
+; CHECK-NEXT:    b.ne .LBB60_5
 ; CHECK-NEXT:  // %bb.3: // %loadbb3
 ; CHECK-NEXT:    ldur x8, [x0, #23]
 ; CHECK-NEXT:    ldur x9, [x1, #23]
 ; CHECK-NEXT:    rev x8, x8
 ; CHECK-NEXT:    rev x9, x9
 ; CHECK-NEXT:    cmp x8, x9
-; CHECK-NEXT:    b.ne .LBB58_5
+; CHECK-NEXT:    b.ne .LBB60_5
 ; CHECK-NEXT:  // %bb.4:
 ; CHECK-NEXT:    mov w8, wzr
-; CHECK-NEXT:    b .LBB58_6
-; CHECK-NEXT:  .LBB58_5: // %res_block
+; CHECK-NEXT:    b .LBB60_6
+; CHECK-NEXT:  .LBB60_5: // %res_block
 ; CHECK-NEXT:    cmp x8, x9
 ; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
 ; CHECK-NEXT:    cneg w8, w8, hs
-; CHECK-NEXT:  .LBB58_6: // %endblock
+; CHECK-NEXT:  .LBB60_6: // %endblock
 ; CHECK-NEXT:    cmp w8, #0
 ; CHECK-NEXT:    cset w0, gt
 ; CHECK-NEXT:    ret
@@ -1327,32 +135...
[truncated]

nikic
nikic reviewed Jan 3, 2025
View reviewed changes
Copy link
Contributor

@nikic nikic left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it's okay to special-case these for now. Longer term the plan was to move ExpandMemCmp to the middle-end pipeline (I think there's a PR for that...), in which case we could drop this an let InstCombine deal with it.

llvm/lib/CodeGen/ExpandMemCmp.cpp Outdated
} else if (match(UI, m_ICmp(Pred, m_Specific(CI), m_Zero()))) {
// Compare with 0. Pred is already set.
} else if (match(UI, m_ICmp(Pred, m_Specific(CI), m_AllOnes())) &&
Pred == ICmpInst::ICMP_SGT) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Use m_SpecificICmp instead.

topperc added a commit that referenced this pull request Jan 3, 2025
@topperc
[ExpandMemCmp][AArch][RISCV][X86] Pre-commit tests for recognizing ca…
bad0f98 
…nonical form of (icmp sle/sge X, 0). NFC

Pre-commit for #121540.
nikic
nikic approved these changes Jan 3, 2025
View reviewed changes
Copy link
Contributor

@nikic nikic left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@topperc topperc merged commit a4e4758 into llvm:main Jan 3, 2025
5 of 8 checks passed
@topperc topperc deleted the pr/memcmp-zero branch January 3, 2025 18:23
@preames
Copy link
Collaborator

preames commented Jan 3, 2025

We seem to have UCMP and SCMP SelectionDAG nodes, are these cases simply missing generic DAG combines? No objection to this patch, just wondering about possible followup.

@topperc
Copy link
Collaborator Author

topperc commented Jan 3, 2025

We seem to have UCMP and SCMP SelectionDAG nodes, are these cases simply missing generic DAG combines? No objection to this patch, just wondering about possible followup.

They are missing. It was mentioned in the description originally as a question about whether to do this here or in DAGCombine. I removed it when I committed based on Nikita's comment #121540 (review)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@nikic nikic nikic approved these changes

@RKSimon RKSimon Awaiting requested review from RKSimon

@dtcxzyw dtcxzyw Awaiting requested review from dtcxzyw

Assignees
No one assigned
Labels
backend:AArch64 backend:X86
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@topperc @llvmbot @preames @nikic