[X86] Fix avoid store forwarding blocks pass for negative displacements

[nikic]

Summary

Fixes https://bugs.llvm.org/show_bug.cgi?id=39926.

The size of the first copy was computed as std::abs(std::abs(LdDisp2) - std::abs(LdDisp1)), which results in skipped bytes if the signs of LdDisp2 and LdDisp1 differ. As far as I can see, this should just be LdDisp2 - LdDisp1. The case where LdDisp1 > LdDisp2 is already handled in the code above, in which case LdDisp2 is set to LdDisp1 and this subtraction will evaluate to Size1 = 0, which is the correct value to skip an overlapping copy.

[RKSimon]

test/CodeGen/X86/pr39926.ll

3 ↗

(On Diff #177406)

Please commit this with trunk's current codegen and rebase so the patch shows the codegen diff.

[nikic]

Rebase over committed test.

llvm/trunk/lib/Target/X86/X86AvoidStoreForwardingBlocks.cpp

for (auto DispSizePair : BlockingStoresDispSizeMap) {
     Size2 = DispSizePair.second;
     // Avoid copying overlapping areas.
     if (LdDisp2 < LdDisp1) {
       int OverlapDelta = LdDisp1 - LdDisp2;
       LdDisp2 += OverlapDelta;
       StDisp2 += OverlapDelta;
       Size2 -= OverlapDelta;
     }
-    Size1 = std::abs(std::abs(LdDisp2) - std::abs(LdDisp1));
+    Size1 = LdDisp2 - LdDisp1;
 
     // Build a copy for the point until the current blocking store's
     // displacement.
     buildCopies(Size1, LoadInst, LdDisp1, StoreInst, StDisp1, LMMOffset,
                 SMMOffset);
     // Build a copy for the current blocking store.
     buildCopies(Size2, LoadInst, LdDisp2, StoreInst, StDisp2, LMMOffset + Size1,
                 SMMOffset + Size1);

llvm/trunk/test/CodeGen/X86/pr39926.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=avx | FileCheck %s
 define i8 @test_offset(i8* %base) {
 ; CHECK-LABEL: test_offset:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    pushq %rax
 ; CHECK-NEXT:    .cfi_def_cfa_offset 16
 ; CHECK-NEXT:    movb $0, 7(%rdi)
 ; CHECK-NEXT:    movw $0, 5(%rdi)
 ; CHECK-NEXT:    movl $0, 1(%rdi)
-; CHECK-NEXT:    movzwl -4(%rdi), %eax
-; CHECK-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
-; CHECK-NEXT:    movb -2(%rdi), %al
+; CHECK-NEXT:    movl -4(%rdi), %eax
+; CHECK-NEXT:    movl %eax, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movb (%rdi), %al
 ; CHECK-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movl 1(%rdi), %eax
 ; CHECK-NEXT:    movl %eax, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movzwl 5(%rdi), %eax
 ; CHECK-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movb 7(%rdi), %al
 ; CHECK-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movl 8(%rdi), %eax