DAGCombine: fold "(zext x) == C" into "x == (trunc C)" if the trunc is lossless.

On x86 this allows to fold a load into the cmp, greatly reducing register pressure.
  movzbl	(%rdi), %eax
  cmpl	$47, %eax
->
  cmpb	$47, (%rdi)

This shaves 8k off gcc.o on i386. I'll leave applying the patch in README.txt to Chris :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130005 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 1 insertions, 30 deletions

diff --git a/lib/Target/README.txt b/lib/Target/README.txt
index 5032b2bbbb..c345ce50c0 100644
--- a/lib/Target/README.txt
+++ b/lib/Target/README.txt
@@ -2259,34 +2259,6 @@ icmp transform.
 
 //===---------------------------------------------------------------------===//
 
-These functions:
-int foo(int *X) {
-  if ((*X & 255) == 47)
-    bar();
-}
-int foo2(int X) {
-  if ((X & 255) == 47)
-    bar();
-}
-
-codegen to:
-
-  movzbl	(%rdi), %eax
-  cmpl	$47, %eax
-  jne	LBB0_2
-
-and:
-  movzbl	%dil, %eax
-  cmpl	$47, %eax
-  jne	LBB1_2
-
-If a dag combine shrunk the compare to a byte compare, then we'd fold the load
-in the first example, and eliminate the movzbl in the second, saving a register.
-This can be a target independent dag combine that works on ISD::SETCC, it would
-catch this before the legalize ops pass.
-
-//===---------------------------------------------------------------------===//
-
 We should optimize this:
 
   %tmp = load i16* %arrayidx, align 4, !tbaa !0
@@ -2329,8 +2301,7 @@ Index: InstCombine/InstCombineCompares.cpp
    {
 
 
-but we can't do that until the dag combine above is added.  Not having this
-is blocking resolving PR6627.
+Not having this is blocking resolving PR6627.
 
 //===---------------------------------------------------------------------===//