[PATCH] D18246: X86: Use push-pop for materializing 8-bit immediates for minsize (take 2)

Hans Wennborg via llvm-commits llvm-commits at lists.llvm.org
Tue Mar 22 13:52:24 PDT 2016

hans added inline comments.

Comment at: test/CodeGen/X86/powi.ll:32-34
@@ -31,5 +31,5 @@
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movl  $15, %edi
+; CHECK-NEXT:    movl  $128, %edi
 ; CHECK-NEXT:    jmp
-  %ret = tail call double @llvm.powi.f64(double %a, i32 15) nounwind ; <double> [#uses=1]
+  %ret = tail call double @llvm.powi.f64(double %a, i32 128) nounwind ; <double> [#uses=1]
   ret double %ret
majnemer wrote:
> Why did the constant in this test change?
Because with the old patch we'd materialize 15 with push/pop, and I figured just tweaking the constant was more clear and less intrusive than updating the expected instructions.

However, in this new patch we conservatively don't use push/pop here because maybe the function could write to the redzone. I'll restore the test.


More information about the llvm-commits mailing list