InstCombine: Turn (zext A) == (B & (1<<X)-1) into A == (trunc B), narrowing the compare.

This saves a cast, and zext is more expensive on platforms with subreg support
than trunc is. This occurs in the BSD implementation of memchr(3), see PR12750.
On the synthetic benchmark from that bug stupid_memchr and bsd_memchr have the
same performance now when not inlining either function.

stupid_memchr: 323.0us
bsd_memchr: 321.0us
memchr: 479.0us

where memchr is the llvm-gcc compiled bsd_memchr from osx lion's libc. When
inlining is enabled bsd_memchr still regresses down to llvm-gcc memchr time,
I haven't fully understood the issue yet, something is grossly mangling the
loop after inlining.

llvm-svn: 158297

1 files changed, 22 insertions, 0 deletions

diff --git a/llvm/test/Transforms/InstCombine/icmp.ll b/llvm/test/Transforms/InstCombine/icmp.ll
index a9ae221d8f9..eaff87d695e 100644
--- a/llvm/test/Transforms/InstCombine/icmp.ll
+++ b/llvm/test/Transforms/InstCombine/icmp.ll
@@ -637,3 +637,25 @@ define i1 @test62(i8* %a) {
 ; CHECK: @test62
 ; CHECK-NEXT: ret i1 true
 }
+
+define i1 @test63(i8 %a, i32 %b) nounwind {
+  %z = zext i8 %a to i32
+  %t = and i32 %b, 255
+  %c = icmp eq i32 %z, %t
+  ret i1 %c
+; CHECK: @test63
+; CHECK-NEXT: %1 = trunc i32 %b to i8
+; CHECK-NEXT: %c = icmp eq i8 %1, %a
+; CHECK-NEXT: ret i1 %c
+}
+
+define i1 @test64(i8 %a, i32 %b) nounwind {
+  %t = and i32 %b, 255
+  %z = zext i8 %a to i32
+  %c = icmp eq i32 %t, %z
+  ret i1 %c
+; CHECK: @test64
+; CHECK-NEXT: %1 = trunc i32 %b to i8
+; CHECK-NEXT: %c = icmp eq i8 %1, %a
+; CHECK-NEXT: ret i1 %c
+}