1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 04:02:41 +01:00
llvm-mirror/test/CodeGen/Blackfin/promote-logic.ll
Chris Lattner 52a02205d8 Change the scheduler from adding nodes in allnodes order
to adding them in a determinstic order (bottom up from 
the root) based on the structure of the graph itself.

This updates tests for some random changes, interesting
bits: CodeGen/Blackfin/promote-logic.ll no longer crashes.
I have no idea why, but that's good right?

CodeGen/X86/2009-07-16-LoadFoldingBug.ll also fails, but
now compiles to have one fewer constant pool entry, making
the expected load that was being folded disappear.  Since it
is an unreduced mass of gnast, I just removed it.

This fixes PR6370

llvm-svn: 97023
2010-02-24 06:11:37 +00:00

43 lines
1.0 KiB
LLVM

; RUN: llc < %s -march=bfin
; DAGCombiner::SimplifyBinOpWithSameOpcodeHands can produce an illegal i16 OR
; operation after LegalizeOps.
define void @mng_display_bgr565() {
entry:
br i1 false, label %bb.preheader, label %return
bb.preheader:
br i1 false, label %cond_true48, label %cond_next80
cond_true48:
%tmp = load i8* null
%tmp51 = zext i8 %tmp to i16
%tmp99 = load i8* null
%tmp54 = bitcast i8 %tmp99 to i8
%tmp54.upgrd.1 = zext i8 %tmp54 to i32
%tmp55 = lshr i32 %tmp54.upgrd.1, 3
%tmp55.upgrd.2 = trunc i32 %tmp55 to i16
%tmp52 = shl i16 %tmp51, 5
%tmp56 = and i16 %tmp55.upgrd.2, 28
%tmp57 = or i16 %tmp56, %tmp52
%tmp60 = zext i16 %tmp57 to i32
%tmp62 = xor i32 0, 65535
%tmp63 = mul i32 %tmp60, %tmp62
%tmp65 = add i32 0, %tmp63
%tmp69 = add i32 0, %tmp65
%tmp70 = lshr i32 %tmp69, 16
%tmp70.upgrd.3 = trunc i32 %tmp70 to i16
%tmp75 = lshr i16 %tmp70.upgrd.3, 8
%tmp75.upgrd.4 = trunc i16 %tmp75 to i8
%tmp76 = lshr i8 %tmp75.upgrd.4, 5
store i8 %tmp76, i8* null
ret void
cond_next80:
ret void
return:
ret void
}