[llvm] r241462 - AMDGPU/SI: Add debugging subtarget feature for DS offsets

[Matt Arsenault]

Author: arsenm
Date: Mon Jul  6 11:01:58 2015
New Revision: 241462

URL: http://llvm.org/viewvc/llvm-project?rev=241462&view=rev
Log:
AMDGPU/SI: Add debugging subtarget feature for DS offsets

We don't have a good way to detect most situations where
DS offsets are usable on SI, so add an option to force using
them even if unsafe for debugging performance problems.

Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPU.td
    llvm/trunk/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
    llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
    llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h
    llvm/trunk/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPU.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPU.td?rev=241462&r1=241461&r2=241462&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPU.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPU.td Mon Jul  6 11:01:58 2015
@@ -98,6 +98,16 @@ def FeatureEnableLoadStoreOpt : Subtarge
         "true",
         "Enable SI load/store optimizer pass">;
 
+// Performance debugging feature. Allow using DS instruction immediate
+// offsets even if the base pointer can't be proven to be base. On SI,
+// base pointer values that won't give the same result as a 16-bit add
+// are not safe to fold, but this will override the conservative test
+// for the base pointer.
+def FeatureEnableUnsafeDSOffsetFolding : SubtargetFeature <"unsafe-ds-offset-folding",
+        "EnableUnsafeDSOffsetFolding",
+        "true",
+        "Force using DS instruction immediate offsets on SI">;
+
 def FeatureFlatAddressSpace : SubtargetFeature<"flat-address-space",
         "FlatAddressSpace",
         "true",

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp?rev=241462&r1=241461&r2=241462&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp Mon Jul  6 11:01:58 2015
@@ -859,7 +859,8 @@ bool AMDGPUDAGToDAGISel::isDSOffsetLegal
       (OffsetBits == 8 && !isUInt<8>(Offset)))
     return false;
 
-  if (Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS)
+  if (Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS ||
+      Subtarget->unsafeDSOffsetFoldingEnabled())
     return true;
 
   // On Southern Islands instruction with a negative base value and an offset

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.cpp?rev=241462&r1=241461&r2=241462&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.cpp Mon Jul  6 11:01:58 2015
@@ -69,6 +69,7 @@ AMDGPUSubtarget::AMDGPUSubtarget(const T
       FP64Denormals(false), FP32Denormals(false), FastFMAF32(false),
       CaymanISA(false), FlatAddressSpace(false), EnableIRStructurizer(true),
       EnablePromoteAlloca(false), EnableIfCvt(true), EnableLoadStoreOpt(false),
+      EnableUnsafeDSOffsetFolding(false),
       WavefrontSize(0), CFALUBug(false), LocalMemorySize(0),
       EnableVGPRSpilling(false), SGPRInitBug(false), IsGCN(false),
       GCN1Encoding(false), GCN3Encoding(false), CIInsts(false), LDSBankCount(0),

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h?rev=241462&r1=241461&r2=241462&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h Mon Jul  6 11:01:58 2015
@@ -76,6 +76,7 @@ private:
   bool EnablePromoteAlloca;
   bool EnableIfCvt;
   bool EnableLoadStoreOpt;
+  bool EnableUnsafeDSOffsetFolding;
   unsigned WavefrontSize;
   bool CFALUBug;
   int LocalMemorySize;
@@ -222,6 +223,10 @@ public:
     return EnableLoadStoreOpt;
   }
 
+  bool unsafeDSOffsetFoldingEnabled() const {
+    return EnableUnsafeDSOffsetFolding;
+  }
+
   unsigned getWavefrontSize() const {
     return WavefrontSize;
   }

Modified: llvm/trunk/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll?rev=241462&r1=241461&r2=241462&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/ds-negative-offset-addressing-mode-loop.ll Mon Jul  6 11:01:58 2015
@@ -1,5 +1,6 @@
-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs -mattr=+load-store-opt -enable-misched < %s | FileCheck -check-prefix=SI --check-prefix=CHECK %s
-; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -mattr=+load-store-opt -enable-misched < %s | FileCheck -check-prefix=CI --check-prefix=CHECK %s
+; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs -mattr=+load-store-opt < %s | FileCheck -check-prefix=SI --check-prefix=CHECK %s
+; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -mattr=+load-store-opt < %s | FileCheck -check-prefix=CI --check-prefix=CHECK %s
+; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs -mattr=+load-store-opt,+unsafe-ds-offset-folding < %s | FileCheck -check-prefix=CI --check-prefix=CHECK %s
 
 declare i32 @llvm.r600.read.tidig.x() #0
 declare void @llvm.AMDGPU.barrier.local() #1