[PATCH] D16172: AMDGPU: Add subtarget feature for instruction rates

[Matt Arsenault via llvm-commits]

arsenm created this revision.
arsenm added a reviewer: tstellarAMD.
arsenm added a subscriber: llvm-commits.
Herald added a subscriber: arsenm.

http://reviews.llvm.org/D16172

Files:
  lib/Target/AMDGPU/AMDGPU.td
  lib/Target/AMDGPU/AMDGPUSubtarget.cpp
  lib/Target/AMDGPU/AMDGPUSubtarget.h
  lib/Target/AMDGPU/Processors.td

Index: lib/Target/AMDGPU/Processors.td
===================================================================
--- lib/Target/AMDGPU/Processors.td
+++ lib/Target/AMDGPU/Processors.td
@@ -84,11 +84,11 @@
 //===----------------------------------------------------------------------===//
 
 def : ProcessorModel<"SI", SIFullSpeedModel,
-  [FeatureSouthernIslands, FeatureFastFMAF32]
+  [FeatureSouthernIslands, FeatureFastFMAF32, HalfRate64Ops]
 >;
 
-def : ProcessorModel<"tahiti",   SIFullSpeedModel,
-  [FeatureSouthernIslands, FeatureFastFMAF32]
+def : ProcessorModel<"tahiti", SIFullSpeedModel,
+  [FeatureSouthernIslands, FeatureFastFMAF32, HalfRate64Ops]
 >;
 
 def : ProcessorModel<"pitcairn", SIQuarterSpeedModel, [FeatureSouthernIslands]>;
@@ -116,8 +116,8 @@
 >;
 
 def : ProcessorModel<"hawaii", SIFullSpeedModel,
-  [FeatureSeaIslands, FeatureFastFMAF32, FeatureLDSBankCount32,
-   FeatureISAVersion7_0_1]
+  [FeatureSeaIslands, FeatureFastFMAF32, HalfRate64Ops,
+   FeatureLDSBankCount32, FeatureISAVersion7_0_1]
 >;
 
 def : ProcessorModel<"mullins",    SIQuarterSpeedModel,
Index: lib/Target/AMDGPU/AMDGPUSubtarget.h
===================================================================
--- lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -68,6 +68,7 @@
   bool FP64Denormals;
   bool FP32Denormals;
   bool FastFMAF32;
+  bool HalfRate64Ops;
   bool CaymanISA;
   bool FlatAddressSpace;
   bool FlatForGlobal;
@@ -157,6 +158,10 @@
     return FastFMAF32;
   }
 
+  bool hasHalfRate64Ops() const {
+    return HalfRate64Ops;
+  }
+
   bool hasFlatAddressSpace() const {
     return FlatAddressSpace;
   }
Index: lib/Target/AMDGPU/AMDGPUSubtarget.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -70,11 +70,14 @@
       DumpCode(false), R600ALUInst(false), HasVertexCache(false),
       TexVTXClauseSize(0), Gen(AMDGPUSubtarget::R600), FP64(false),
       FP64Denormals(false), FP32Denormals(false), FastFMAF32(false),
-      CaymanISA(false), FlatAddressSpace(false), FlatForGlobal(false),
-      EnableIRStructurizer(true), EnablePromoteAlloca(false), EnableIfCvt(true),
-      EnableLoadStoreOpt(false), EnableUnsafeDSOffsetFolding(false),
+      HalfRate64Ops(false), CaymanISA(false), FlatAddressSpace(false),
+      FlatForGlobal(false), EnableIRStructurizer(true),
+      EnablePromoteAlloca(false),
+      EnableIfCvt(true), EnableLoadStoreOpt(false),
+      EnableUnsafeDSOffsetFolding(false),
       EnableXNACK(false),
-      WavefrontSize(0), CFALUBug(false), LocalMemorySize(0),
+      WavefrontSize(0), CFALUBug(false),
+      LocalMemorySize(0),
       EnableVGPRSpilling(false), SGPRInitBug(false), IsGCN(false),
       GCN1Encoding(false), GCN3Encoding(false), CIInsts(false), LDSBankCount(0),
       IsaVersion(ISAVersion0_0_0), EnableHugeScratchBuffer(false),
Index: lib/Target/AMDGPU/AMDGPU.td
===================================================================
--- lib/Target/AMDGPU/AMDGPU.td
+++ lib/Target/AMDGPU/AMDGPU.td
@@ -59,6 +59,12 @@
         "Assuming f32 fma is at least as fast as mul + add",
         []>;
 
+def HalfRate64Ops : SubtargetFeature<"half-rate-64-ops",
+        "HalfRate64Ops",
+        "true",
+        "Most fp64 instructions are half rate instead of quarter",
+        []>;
+
 // Some instructions do not support denormals despite this flag. Using
 // fp32 denormals also causes instructions to run at the double
 // precision rate for the device.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D16172.44827.patch
Type: text/x-patch
Size: 3547 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160114/5263ef45/attachment.bin>