summaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/AMDGPUFeatures.td
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/AMDGPU/AMDGPUFeatures.td')
-rw-r--r--lib/Target/AMDGPU/AMDGPUFeatures.td60
1 files changed, 60 insertions, 0 deletions
diff --git a/lib/Target/AMDGPU/AMDGPUFeatures.td b/lib/Target/AMDGPU/AMDGPUFeatures.td
new file mode 100644
index 000000000000..b375cae9018e
--- /dev/null
+++ b/lib/Target/AMDGPU/AMDGPUFeatures.td
@@ -0,0 +1,60 @@
+//===-- AMDGPUFeatures.td - AMDGPU Feature Definitions -----*- tablegen -*-===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+def FeatureFP64 : SubtargetFeature<"fp64",
+ "FP64",
+ "true",
+ "Enable double precision operations"
+>;
+
+def FeatureFMA : SubtargetFeature<"fmaf",
+ "FMA",
+ "true",
+ "Enable single precision FMA (not as fast as mul+add, but fused)"
+>;
+
+class SubtargetFeatureLocalMemorySize <int Value> : SubtargetFeature<
+ "localmemorysize"#Value,
+ "LocalMemorySize",
+ !cast<string>(Value),
+ "The size of local memory in bytes"
+>;
+
+def FeatureLocalMemorySize0 : SubtargetFeatureLocalMemorySize<0>;
+def FeatureLocalMemorySize32768 : SubtargetFeatureLocalMemorySize<32768>;
+def FeatureLocalMemorySize65536 : SubtargetFeatureLocalMemorySize<65536>;
+
+class SubtargetFeatureWavefrontSize <int Value> : SubtargetFeature<
+ "wavefrontsize"#Value,
+ "WavefrontSize",
+ !cast<string>(Value),
+ "The number of threads per wavefront"
+>;
+
+def FeatureWavefrontSize16 : SubtargetFeatureWavefrontSize<16>;
+def FeatureWavefrontSize32 : SubtargetFeatureWavefrontSize<32>;
+def FeatureWavefrontSize64 : SubtargetFeatureWavefrontSize<64>;
+
+class SubtargetFeatureGeneration <string Value, string Subtarget,
+ list<SubtargetFeature> Implies> :
+ SubtargetFeature <Value, "Gen", Subtarget#"::"#Value,
+ Value#" GPU generation", Implies>;
+
+def FeatureDX10Clamp : SubtargetFeature<"dx10-clamp",
+ "DX10Clamp",
+ "true",
+ "clamp modifier clamps NaNs to 0.0"
+>;
+
+def FeaturePromoteAlloca : SubtargetFeature <"promote-alloca",
+ "EnablePromoteAlloca",
+ "true",
+ "Enable promote alloca pass"
+>;
+