PowerPC inline asm was emitting two output operands

for a single "m" constraint; this is wrong because the
opcode of a load or store would have to change in parallel.
This patch makes it always compute addresses into a register,
which is correct but not as efficient as possible.  7144566.

llvm-svn: 79292

1 files changed, 22 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/PowerPC/2009-08-17-inline-asm-addr-mode-breakage.ll b/llvm/test/CodeGen/PowerPC/2009-08-17-inline-asm-addr-mode-breakage.ll
new file mode 100644
index 00000000000..9e0c5de39f9
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/2009-08-17-inline-asm-addr-mode-breakage.ll
@@ -0,0 +1,22 @@
+; RUN: llvm-as < %s | llc -march=ppc32 | grep add
+; ModuleID = '<stdin>'
+target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f128:64:128"
+target triple = "powerpc-apple-darwin10.0"
+; It is wrong on powerpc to substitute reg+reg for $0; the stw opcode
+; would have to change.
+
+@x = external global [0 x i32]                    ; <[0 x i32]*> [#uses=1]
+
+define void @foo(i32 %y) nounwind ssp {
+entry:
+  %y_addr = alloca i32                            ; <i32*> [#uses=2]
+  %"alloca point" = bitcast i32 0 to i32          ; <i32> [#uses=0]
+  store i32 %y, i32* %y_addr
+  %0 = load i32* %y_addr, align 4                 ; <i32> [#uses=1]
+  %1 = getelementptr inbounds [0 x i32]* @x, i32 0, i32 %0 ; <i32*> [#uses=1]
+  call void asm sideeffect "isync\0A\09eieio\0A\09stw $1, $0", "=*o,r,~{memory}"(i32* %1, i32 0) nounwind
+  br label %return
+
+return:                                           ; preds = %entry
+  ret void
+}