Disable the load/store optimization pass for Thumb-1.

Moritz's changes have improved codegen a lot, but further testing showed significant correctness problems. Disable by default until these have been worked out.

Patch by Moritz Roth!

llvm-svn: 210789

diff --git a/llvm/lib/Target/ARM/ARMTargetMachine.cpp b/llvm/lib/Target/ARM/ARMTargetMachine.cpp
index 6ef2ea4..c1e866b 100644 (file)
--- a/llvm/lib/Target/ARM/ARMTargetMachine.cpp
+++ b/llvm/lib/Target/ARM/ARMTargetMachine.cpp
@@ -265,7 +265,8 @@ bool ARMPassConfig::addInstSelector() {
 }
 
 bool ARMPassConfig::addPreRegAlloc() {
-  if (getOptLevel() != CodeGenOpt::None)
+  // FIXME: Temporarily disabling Thumb-1 pre-RA Load/Store optimization pass
+  if (getOptLevel() != CodeGenOpt::None && !getARMSubtarget().isThumb1Only())
     addPass(createARMLoadStoreOptimizationPass(true));
   if (getOptLevel() != CodeGenOpt::None && getARMSubtarget().isCortexA9())
     addPass(createMLxExpansionPass());
@@ -280,8 +281,11 @@ bool ARMPassConfig::addPreRegAlloc() {
 
 bool ARMPassConfig::addPreSched2() {
   if (getOptLevel() != CodeGenOpt::None) {
-    addPass(createARMLoadStoreOptimizationPass());
-    printAndVerify("After ARM load / store optimizer");
+    // FIXME: Temporarily disabling Thumb-1 post-RA Load/Store optimization pass
+    if (!getARMSubtarget().isThumb1Only()) {
+      addPass(createARMLoadStoreOptimizationPass());
+      printAndVerify("After ARM load / store optimizer");
+    }
 
     if (getARMSubtarget().hasNEON())
       addPass(createExecutionDependencyFixPass(&ARM::DPRRegClass));

diff --git a/llvm/test/CodeGen/Thumb/2014-06-10-thumb1-ldst-opt-bug.ll b/llvm/test/CodeGen/Thumb/2014-06-10-thumb1-ldst-opt-bug.ll
index 3c43383..ae66369 100644 (file)
--- a/llvm/test/CodeGen/Thumb/2014-06-10-thumb1-ldst-opt-bug.ll
+++ b/llvm/test/CodeGen/Thumb/2014-06-10-thumb1-ldst-opt-bug.ll
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -mtriple=thumbv6m-eabi -o - | FileCheck %s
+; XFAIL: *
 
 define void @foo(i32* %A) #0 {
 entry:

diff --git a/llvm/test/CodeGen/Thumb/dyn-stackalloc.ll b/llvm/test/CodeGen/Thumb/dyn-stackalloc.ll
index 6bc39af..6c6de55 100644 (file)
--- a/llvm/test/CodeGen/Thumb/dyn-stackalloc.ll
+++ b/llvm/test/CodeGen/Thumb/dyn-stackalloc.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -mtriple=thumb-apple-darwin -disable-cgp-branch-opts -disable-post-ra | FileCheck %s -check-prefix=CHECK -check-prefix=RA_GREEDY
-; RUN: llc < %s -mtriple=thumb-apple-darwin -disable-cgp-branch-opts -disable-post-ra -regalloc=basic | FileCheck %s -check-prefix=CHECK -check-prefix=RA_BASIC
+; RUN: llc < %s -mtriple=thumb-apple-darwin -disable-cgp-branch-opts -disable-post-ra | FileCheck %s
+; RUN: llc < %s -mtriple=thumb-apple-darwin -disable-cgp-branch-opts -disable-post-ra -regalloc=basic | FileCheck %s
 
        %struct.state = type { i32, %struct.info*, float**, i32, i32, i32, i32, i32, i32, i32, i32, i32, i64, i64, i64, i64, i64, i64, i8* }
        %struct.info = type { i32, i32, i32, i32, i32, i32, i32, i8* }
@@ -45,8 +45,7 @@ define void @t2(%struct.comment* %vc, i8* %tag, i8* %contents) {
 ; CHECK: sub sp, #
 ; CHECK: mov r[[R0:[0-9]+]], sp
 ; CHECK: str r{{[0-9+]}}, [r[[R0]]
-; RA_GREEDY: str r{{[0-9+]}}, [r[[R0]]
-; RA_BASIC: stm r[[R0]]!
+; CHECK: str r{{[0-9+]}}, [r[[R0]]
 ; CHECK-NOT: ldr r0, [sp
 ; CHECK: mov r[[R1:[0-9]+]], sp
 ; CHECK: subs r[[R2:[0-9]+]], r[[R1]], r{{[0-9]+}}

diff --git a/llvm/test/CodeGen/Thumb/thumb-ldm.ll b/llvm/test/CodeGen/Thumb/thumb-ldm.ll
index dd98e6f..95f3edc 100644 (file)
--- a/llvm/test/CodeGen/Thumb/thumb-ldm.ll
+++ b/llvm/test/CodeGen/Thumb/thumb-ldm.ll
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -mtriple=thumbv6m-eabi -o - | FileCheck %s
+; XFAIL: *
 
 @X = external global [0 x i32]          ; <[0 x i32]*> [#uses=5]
 

diff --git a/llvm/test/CodeGen/Thumb/thumb-memcpy-ldm-stm.ll b/llvm/test/CodeGen/Thumb/thumb-memcpy-ldm-stm.ll
index 06cfd9b..dedc82b 100644 (file)
--- a/llvm/test/CodeGen/Thumb/thumb-memcpy-ldm-stm.ll
+++ b/llvm/test/CodeGen/Thumb/thumb-memcpy-ldm-stm.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -mtriple=thumbv6m-eabi %s -o - | FileCheck %s
+; XFAIL: *
 
 @d = external global [64 x i32]
 @s = external global [64 x i32]