teach codegen to turn trunc(zextload) into load when possible.

This doesn't occur much at all, it only seems to formed in the case
when the trunc optimization kicks in due to phase ordering.  In that
case it is saves a few bytes on x86-32.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101350 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -3637,7 +3637,7 @@ SDValue DAGCombiner::ReduceLoadWidth(SDNode *N) {
   // Do not generate loads of non-round integer types since these can
   // be expensive (and would be wrong if the type is not byte sized).
   if (isa<LoadSDNode>(N0) && N0.hasOneUse() && ExtVT.isRound() &&
-      cast<LoadSDNode>(N0)->getMemoryVT().getSizeInBits() > EVTBits &&
+      cast<LoadSDNode>(N0)->getMemoryVT().getSizeInBits() >= EVTBits &&
       // Do not change the width of a volatile load.
       !cast<LoadSDNode>(N0)->isVolatile()) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);

test/CodeGen/X86/store-narrow.ll

@@ -1,5 +1,6 @@
 ; rdar://7860110
-; RUN: llc < %s | FileCheck %s
+; RUN: llc < %s | FileCheck %s -check-prefix=X64
+; RUN: llc -march=x86 < %s | FileCheck %s -check-prefix=X32
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-darwin10.2"
 
@@ -12,8 +13,12 @@ entry:
   store i32 %D, i32* %a0, align 4
   ret void
   
-; CHECK: test1:
+; X64: test1:
-; CHECK: movb	%sil, (%rdi)
+; X64: movb	%sil, (%rdi)
+
+; X32: test1:
+; X32: movb	8(%esp), %al
+; X32: movb	%al, (%{{.*}})
 }
 
 define void @test2(i32* nocapture %a0, i8 zeroext %a1) nounwind ssp {
@@ -25,8 +30,12 @@ entry:
   %D = or i32 %B, %CS
   store i32 %D, i32* %a0, align 4
   ret void
-; CHECK: test2:
+; X64: test2:
-; CHECK: movb	%sil, 1(%rdi)
+; X64: movb	%sil, 1(%rdi)
+
+; X32: test2:
+; X32: movb	8(%esp), %al
+; X32: movb	%al, 1(%{{.*}})
 }
 
 define void @test3(i32* nocapture %a0, i16 zeroext %a1) nounwind ssp {
@@ -37,8 +46,12 @@ entry:
   %D = or i32 %B, %C
   store i32 %D, i32* %a0, align 4
   ret void
-; CHECK: test3:
+; X64: test3:
-; CHECK: movw	%si, (%rdi)
+; X64: movw	%si, (%rdi)
+
+; X32: test3:
+; X32: movw	8(%esp), %ax
+; X32: movw	%ax, (%{{.*}})
 }
 
 define void @test4(i32* nocapture %a0, i16 zeroext %a1) nounwind ssp {
@@ -50,8 +63,12 @@ entry:
   %D = or i32 %B, %CS
   store i32 %D, i32* %a0, align 4
   ret void
-; CHECK: test4:
+; X64: test4:
-; CHECK: movw	%si, 2(%rdi)
+; X64: movw	%si, 2(%rdi)
+
+; X32: test4:
+; X32: movw	8(%esp), %ax
+; X32: movw	%ax, 2(%{{.*}})
 }
 
 define void @test5(i64* nocapture %a0, i16 zeroext %a1) nounwind ssp {
@@ -63,8 +80,12 @@ entry:
   %D = or i64 %B, %CS
   store i64 %D, i64* %a0, align 4
   ret void
-; CHECK: test5:
+; X64: test5:
-; CHECK: movw	%si, 2(%rdi)
+; X64: movw	%si, 2(%rdi)
+
+; X32: test5:
+; X32: movw	8(%esp), %ax
+; X32: movw	%ax, 2(%{{.*}})
 }
 
 define void @test6(i64* nocapture %a0, i8 zeroext %a1) nounwind ssp {
@@ -76,6 +97,11 @@ entry:
   %D = or i64 %B, %CS
   store i64 %D, i64* %a0, align 4
   ret void
-; CHECK: test6:
+; X64: test6:
-; CHECK: movb	%sil, 5(%rdi)
+; X64: movb	%sil, 5(%rdi)
+
+
+; X32: test6:
+; X32: movb	8(%esp), %al
+; X32: movb	%al, 5(%{{.*}})
 }