[llvm] r348750 - [X86] Fix AvoidStoreForwardingBlocks pass for negative displacements

[Nikita Popov via llvm-commits]

Author: nikic
Date: Mon Dec 10 02:16:50 2018
New Revision: 348750

URL: http://llvm.org/viewvc/llvm-project?rev=348750&view=rev
Log:
[X86] Fix AvoidStoreForwardingBlocks pass for negative displacements

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.

Differential Revision: https://reviews.llvm.org/D55485

Modified:
    llvm/trunk/lib/Target/X86/X86AvoidStoreForwardingBlocks.cpp
    llvm/trunk/test/CodeGen/X86/pr39926.ll

Modified: llvm/trunk/lib/Target/X86/X86AvoidStoreForwardingBlocks.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86AvoidStoreForwardingBlocks.cpp?rev=348750&r1=348749&r2=348750&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86AvoidStoreForwardingBlocks.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86AvoidStoreForwardingBlocks.cpp Mon Dec 10 02:16:50 2018
@@ -586,7 +586,7 @@ void X86AvoidSFBPass::breakBlockedCopies
       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.

Modified: llvm/trunk/test/CodeGen/X86/pr39926.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pr39926.ll?rev=348750&r1=348749&r2=348750&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/pr39926.ll (original)
+++ llvm/trunk/test/CodeGen/X86/pr39926.ll Mon Dec 10 02:16:50 2018
@@ -8,9 +8,9 @@ define i8 @test_offset(i8* %base) {
 ; 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)