[PATCH] D50834: AMDGPU: Add feature for fast f32 denormals

[Matt Arsenault via Phabricator via llvm-commits]

arsenm created this revision.
arsenm added reviewers: rampitec, kzhuravl, b-sumner.
Herald added subscribers: jfb, t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng.

https://reviews.llvm.org/D50834

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


Index: lib/Target/AMDGPU/AMDGPUSubtarget.h
===================================================================
--- lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -310,6 +310,7 @@
 
   // Possibly statically set by tablegen, but may want to be overridden.
   bool FastFMAF32;
+  bool FastDenormalF32;
   bool HalfRate64Ops;
 
   // Dynamially set bits that enable features.
Index: lib/Target/AMDGPU/AMDGPUSubtarget.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -154,6 +154,7 @@
     MaxPrivateElementSize(0),
 
     FastFMAF32(false),
+    FastDenormalF32(false),
     HalfRate64Ops(false),
 
     FP64FP16Denormals(false),
Index: lib/Target/AMDGPU/AMDGPU.td
===================================================================
--- lib/Target/AMDGPU/AMDGPU.td
+++ lib/Target/AMDGPU/AMDGPU.td
@@ -21,6 +21,12 @@
   "Assuming f32 fma is at least as fast as mul + add"
 >;
 
+def FeatureFastDenormalF32 : SubtargetFeature<"fast-denormal-f32",
+  "FastDenormalF32",
+  "true",
+  "Enabling denormals does not cause f32 instructions to run at f64 rates"
+>;
+
 def FeatureMIMG_R128 : SubtargetFeature<"mimg-r128",
   "MIMG_R128",
   "true",
@@ -431,7 +437,7 @@
    FeatureSMemRealTime, FeatureVGPRIndexMode, FeatureMovrel,
    FeatureScalarStores, FeatureInv2PiInlineImm,
    FeatureSDWA, FeatureSDWAOutModsVOPC, FeatureSDWAMac, FeatureDPP,
-   FeatureIntClamp
+   FeatureIntClamp, FeatureFastDenormalF32
   ]
 >;
 
@@ -444,7 +450,7 @@
    FeatureFastFMAF32, FeatureDPP, FeatureIntClamp,
    FeatureSDWA, FeatureSDWAOmod, FeatureSDWAScalar, FeatureSDWASdst,
    FeatureFlatInstOffsets, FeatureFlatGlobalInsts, FeatureFlatScratchInsts,
-   FeatureAddNoCarryInsts, FeatureScalarAtomics
+   FeatureAddNoCarryInsts, FeatureScalarAtomics, FeatureFastDenormalF32
   ]
 >;
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D50834.160975.patch
Type: text/x-patch
Size: 1906 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180816/32274d04/attachment.bin>