shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
llvm-project/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td
Diana Picus ac005e16f6
Reapply "[AMDGPU] Intrinsic for launching whole wave functions" (#153584) 
This reverts commit 14cd1339318b16e08c1363ec6896bd7d1e4ae281. The
buildbot failure seems to have been a cmake issue which has been
discussed in more detail in this Discourse post:

https://discourse.llvm.org/t/cmake-doesnt-regenerate-all-tablegen-target-files/87901

If any buildbots fail to select arbitrary intrinsics with this patch,
it's worth considering using clean builds with ccache instead of
incremental builds, as recommended here:

https://llvm.org/docs/HowToAddABuilder.html#:~:text=Use%20CCache%20and%20NOT%20incremental%20builds

The original commit message for this patch:
Add the llvm.amdgcn.call.whole.wave intrinsic for calling whole wave
functions. This will take as its first argument the callee with the
amdgpu_gfx_whole_wave calling convention, followed by the call
parameters which must match the signature of the callee except for the
first function argument (the i1 original EXEC mask, which doesn't need
to be passed in). Indirect calls are not allowed.

Make direct calls to amdgpu_gfx_whole_wave functions a verifier error.

Tail calls are handled in a future patch.
2025-08-15 10:12:47 +02:00

408 lines
28 KiB
TableGen
Raw Blame History

//===-- AMDGPUSearchableTables.td - ------------------------*- tablegen -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Resource intrinsics table.
//===----------------------------------------------------------------------===//
class RsrcIntrinsic<AMDGPURsrcIntrinsic intr> {
Intrinsic Intr = !cast<Intrinsic>(intr);
bits<8> RsrcArg = intr.RsrcArg;
bit IsImage = intr.IsImage;
}
def RsrcIntrinsics : GenericTable {
let FilterClass = "RsrcIntrinsic";
let Fields = ["Intr", "RsrcArg", "IsImage"];
let PrimaryKey = ["Intr"];
let PrimaryKeyName = "lookupRsrcIntrinsic";
}
foreach intr = !listconcat(AMDGPUBufferIntrinsics,
AMDGPUImageDimIntrinsics,
AMDGPUImageDimAtomicIntrinsics) in {
def : RsrcIntrinsic<!cast<AMDGPURsrcIntrinsic>(intr)>;
}
class GcnBufferFormatBase<bits<8> f, bits<8> bpc, bits<8> numc, bits<8> nfmt, bits<8> dfmt> {
bits<8> Format = f;
bits<8> BitsPerComp = bpc;
bits<8> NumComponents = numc;
bits<8> NumFormat = nfmt;
bits<8> DataFormat = dfmt;
}
class Gfx9BufferFormat<bits<8> f, bits<8> bpc, bits<8> numc, bits<8> nfmt, bits<8> dfmt> : GcnBufferFormatBase<f, bpc, numc, nfmt, dfmt>;
class Gfx10BufferFormat<bits<8> f, bits<8> bpc, bits<8> numc, bits<8> nfmt, bits<8> dfmt> : GcnBufferFormatBase<f, bpc, numc, nfmt, dfmt>;
class Gfx11PlusBufferFormat<bits<8> f, bits<8> bpc, bits<8> numc, bits<8> nfmt, bits<8> dfmt> : GcnBufferFormatBase<f, bpc, numc, nfmt, dfmt>;
class GcnBufferFormatTable : GenericTable {
let CppTypeName = "GcnBufferFormatInfo";
let Fields = ["Format", "BitsPerComp", "NumComponents", "NumFormat", "DataFormat"];
let PrimaryKey = ["BitsPerComp", "NumComponents", "NumFormat"];
}
def Gfx9BufferFormat : GcnBufferFormatTable {
let FilterClass = "Gfx9BufferFormat";
let PrimaryKeyName = "getGfx9BufferFormatInfo";
}
def Gfx10BufferFormat : GcnBufferFormatTable {
let FilterClass = "Gfx10BufferFormat";
let PrimaryKeyName = "getGfx10BufferFormatInfo";
}
def Gfx11PlusBufferFormat : GcnBufferFormatTable {
let FilterClass = "Gfx11PlusBufferFormat";
let PrimaryKeyName = "getGfx11PlusBufferFormatInfo";
}
def getGfx9BufferFormatInfo : SearchIndex {
let Table = Gfx9BufferFormat;
let Key = ["Format"];
}
def getGfx10BufferFormatInfo : SearchIndex {
let Table = Gfx10BufferFormat;
let Key = ["Format"];
}
def getGfx11PlusBufferFormatInfo : SearchIndex {
let Table = Gfx11PlusBufferFormat;
let Key = ["Format"];
}
// Buffer formats with equal component sizes (GFX9 and earlier)
def : Gfx9BufferFormat< /*FORMAT_8_UNORM*/ 0x01, 8, 1, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_8*/ 1>;
def : Gfx9BufferFormat< /*FORMAT_8_SNORM*/ 0x11, 8, 1, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_8*/ 1>;
def : Gfx9BufferFormat< /*FORMAT_8_USCALED*/ 0x21, 8, 1, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_8*/ 1>;
def : Gfx9BufferFormat< /*FORMAT_8_SSCALED*/ 0x31, 8, 1, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_8*/ 1>;
def : Gfx9BufferFormat< /*FORMAT_8_UINT*/ 0x41, 8, 1, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_8*/ 1>;
def : Gfx9BufferFormat< /*FORMAT_8_SINT*/ 0x51, 8, 1, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_8*/ 1>;
def : Gfx9BufferFormat< /*FORMAT_16_UNORM*/ 0x02, 16, 1, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_16*/ 2>;
def : Gfx9BufferFormat< /*FORMAT_16_SNORM*/ 0x12, 16, 1, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_16*/ 2>;
def : Gfx9BufferFormat< /*FORMAT_16_USCALED*/ 0x22, 16, 1, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_16*/ 2>;
def : Gfx9BufferFormat< /*FORMAT_16_SSCALED*/ 0x32, 16, 1, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_16*/ 2>;
def : Gfx9BufferFormat< /*FORMAT_16_UINT*/ 0x42, 16, 1, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_16*/ 2>;
def : Gfx9BufferFormat< /*FORMAT_16_SINT*/ 0x52, 16, 1, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_16*/ 2>;
def : Gfx9BufferFormat< /*FORMAT_16_FLOAT*/ 0x72, 16, 1, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_16*/ 2>;
def : Gfx9BufferFormat< /*FORMAT_8_8_UNORM*/ 0x03, 8, 2, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_8_8*/ 3>;
def : Gfx9BufferFormat< /*FORMAT_8_8_SNORM*/ 0x13, 8, 2, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_8_8*/ 3>;
def : Gfx9BufferFormat< /*FORMAT_8_8_USCALED*/ 0x23, 8, 2, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_8_8*/ 3>;
def : Gfx9BufferFormat< /*FORMAT_8_8_SSCALED*/ 0x33, 8, 2, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_8_8*/ 3>;
def : Gfx9BufferFormat< /*FORMAT_8_8_UINT*/ 0x43, 8, 2, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_8_8*/ 3>;
def : Gfx9BufferFormat< /*FORMAT_8_8_SINT*/ 0x53, 8, 2, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_8_8*/ 3>;
def : Gfx9BufferFormat< /*FORMAT_32_UINT*/ 0x44, 32, 1, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32*/ 4>;
def : Gfx9BufferFormat< /*FORMAT_32_SINT*/ 0x54, 32, 1, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32*/ 4>;
def : Gfx9BufferFormat< /*FORMAT_32_FLOAT*/ 0x74, 32, 1, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32*/ 4>;
def : Gfx9BufferFormat< /*FORMAT_16_16_UNORM*/ 0x05, 16, 2, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_16_16*/ 5>;
def : Gfx9BufferFormat< /*FORMAT_16_16_SNORM*/ 0x15, 16, 2, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_16_16*/ 5>;
def : Gfx9BufferFormat< /*FORMAT_16_16_USCALED*/ 0x25, 16, 2, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_16_16*/ 5>;
def : Gfx9BufferFormat< /*FORMAT_16_16_SSCALED*/ 0x35, 16, 2, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_16_16*/ 5>;
def : Gfx9BufferFormat< /*FORMAT_16_16_UINT*/ 0x45, 16, 2, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_16_16*/ 5>;
def : Gfx9BufferFormat< /*FORMAT_16_16_SINT*/ 0x55, 16, 2, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_16_16*/ 5>;
def : Gfx9BufferFormat< /*FORMAT_16_16_FLOAT*/ 0x75, 16, 2, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_16_16*/ 5>;
def : Gfx9BufferFormat< /*FORMAT_8_8_8_8_UNORM*/ 0x0A, 8, 4, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx9BufferFormat< /*FORMAT_8_8_8_8_SNORM*/ 0x1A, 8, 4, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx9BufferFormat< /*FORMAT_8_8_8_8_USCALED*/ 0x2A, 8, 4, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx9BufferFormat< /*FORMAT_8_8_8_8_SSCALED*/ 0x3A, 8, 4, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx9BufferFormat< /*FORMAT_8_8_8_8_UINT*/ 0x4A, 8, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx9BufferFormat< /*FORMAT_8_8_8_8_SINT*/ 0x5A, 8, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx9BufferFormat< /*FORMAT_32_32_UINT*/ 0x4B, 32, 2, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx9BufferFormat< /*FORMAT_32_32_SINT*/ 0x5B, 32, 2, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx9BufferFormat< /*FORMAT_32_32_FLOAT*/ 0x7B, 32, 2, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx9BufferFormat< /*FORMAT_16_16_16_16_UNORM*/ 0x0C, 16, 4, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx9BufferFormat< /*FORMAT_16_16_16_16_SNORM*/ 0x1C, 16, 4, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx9BufferFormat< /*FORMAT_16_16_16_16_USCALED*/ 0x2C, 16, 4, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx9BufferFormat< /*FORMAT_16_16_16_16_SSCALED*/ 0x3C, 16, 4, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx9BufferFormat< /*FORMAT_16_16_16_16_UINT*/ 0x4C, 16, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx9BufferFormat< /*FORMAT_16_16_16_16_SINT*/ 0x5C, 16, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx9BufferFormat< /*FORMAT_16_16_16_16_FLOAT*/ 0x7C, 16, 4, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx9BufferFormat< /*FORMAT_32_32_32_UINT*/ 0x4D, 32, 3, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx9BufferFormat< /*FORMAT_32_32_32_SINT*/ 0x5D, 32, 3, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx9BufferFormat< /*FORMAT_32_32_32_FLOAT*/ 0x7D, 32, 3, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx9BufferFormat< /*FORMAT_32_32_32_32_UINT*/ 0x4E, 32, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32_32_32*/ 14>;
def : Gfx9BufferFormat< /*FORMAT_32_32_32_32_SINT*/ 0x5E, 32, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32_32_32*/ 14>;
def : Gfx9BufferFormat< /*FORMAT_32_32_32_32_FLOAT*/ 0x7E, 32, 4, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32_32_32*/ 14>;
// Buffer formats with equal component sizes (GFX10 and later)
multiclass Gfx10PlusBufferFormat<bits<8> f, bits<8> bpc, bits<8> numc, bits<8> nfmt, bits<8> dfmt> {
def : Gfx10BufferFormat<f, bpc, numc, nfmt, dfmt>;
def : Gfx11PlusBufferFormat<f, bpc, numc, nfmt, dfmt>;
}
defm : Gfx10PlusBufferFormat< /*FORMAT_8_UNORM*/ 0x01, 8, 1, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_8*/ 1>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_SNORM*/ 0x02, 8, 1, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_8*/ 1>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_USCALED*/ 0x03, 8, 1, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_8*/ 1>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_SSCALED*/ 0x04, 8, 1, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_8*/ 1>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_UINT*/ 0x05, 8, 1, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_8*/ 1>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_SINT*/ 0x06, 8, 1, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_8*/ 1>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_UNORM*/ 0x07, 16, 1, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_16*/ 2>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_SNORM*/ 0x08, 16, 1, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_16*/ 2>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_USCALED*/ 0x09, 16, 1, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_16*/ 2>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_SSCALED*/ 0x0A, 16, 1, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_16*/ 2>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_UINT*/ 0x0B, 16, 1, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_16*/ 2>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_SINT*/ 0x0C, 16, 1, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_16*/ 2>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_FLOAT*/ 0x0D, 16, 1, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_16*/ 2>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_8_UNORM*/ 0x0E, 8, 2, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_8_8*/ 3>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_8_SNORM*/ 0x0F, 8, 2, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_8_8*/ 3>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_8_USCALED*/ 0x10, 8, 2, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_8_8*/ 3>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_8_SSCALED*/ 0x11, 8, 2, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_8_8*/ 3>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_8_UINT*/ 0x12, 8, 2, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_8_8*/ 3>;
defm : Gfx10PlusBufferFormat< /*FORMAT_8_8_SINT*/ 0x13, 8, 2, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_8_8*/ 3>;
defm : Gfx10PlusBufferFormat< /*FORMAT_32_UINT*/ 0x14, 32, 1, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32*/ 4>;
defm : Gfx10PlusBufferFormat< /*FORMAT_32_SINT*/ 0x15, 32, 1, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32*/ 4>;
defm : Gfx10PlusBufferFormat< /*FORMAT_32_FLOAT*/ 0x16, 32, 1, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32*/ 4>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_16_UNORM*/ 0x17, 16, 2, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_16_16*/ 5>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_16_SNORM*/ 0x18, 16, 2, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_16_16*/ 5>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_16_USCALED*/ 0x19, 16, 2, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_16_16*/ 5>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_16_SSCALED*/ 0x1A, 16, 2, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_16_16*/ 5>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_16_UINT*/ 0x1B, 16, 2, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_16_16*/ 5>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_16_SINT*/ 0x1C, 16, 2, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_16_16*/ 5>;
defm : Gfx10PlusBufferFormat< /*FORMAT_16_16_FLOAT*/ 0x1D, 16, 2, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_16_16*/ 5>;
// Buffer formats with equal component sizes (GFX10 only)
def : Gfx10BufferFormat< /*FORMAT_8_8_8_8_UNORM*/ 0x38, 8, 4, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx10BufferFormat< /*FORMAT_8_8_8_8_SNORM*/ 0x39, 8, 4, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx10BufferFormat< /*FORMAT_8_8_8_8_USCALED*/ 0x3A, 8, 4, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx10BufferFormat< /*FORMAT_8_8_8_8_SSCALED*/ 0x3B, 8, 4, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx10BufferFormat< /*FORMAT_8_8_8_8_UINT*/ 0x3C, 8, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx10BufferFormat< /*FORMAT_8_8_8_8_SINT*/ 0x3D, 8, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx10BufferFormat< /*FORMAT_32_32_UINT*/ 0x3E, 32, 2, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx10BufferFormat< /*FORMAT_32_32_SINT*/ 0x3F, 32, 2, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx10BufferFormat< /*FORMAT_32_32_FLOAT*/ 0x40, 32, 2, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx10BufferFormat< /*FORMAT_16_16_16_16_UNORM*/ 0x41, 16, 4, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx10BufferFormat< /*FORMAT_16_16_16_16_SNORM*/ 0x42, 16, 4, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx10BufferFormat< /*FORMAT_16_16_16_16_USCALED*/ 0x43, 16, 4, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx10BufferFormat< /*FORMAT_16_16_16_16_SSCALED*/ 0x44, 16, 4, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx10BufferFormat< /*FORMAT_16_16_16_16_UINT*/ 0x45, 16, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx10BufferFormat< /*FORMAT_16_16_16_16_SINT*/ 0x46, 16, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx10BufferFormat< /*FORMAT_16_16_16_16_FLOAT*/ 0x47, 16, 4, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx10BufferFormat< /*FORMAT_32_32_32_UINT*/ 0x48, 32, 3, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx10BufferFormat< /*FORMAT_32_32_32_SINT*/ 0x49, 32, 3, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx10BufferFormat< /*FORMAT_32_32_32_FLOAT*/ 0x4A, 32, 3, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx10BufferFormat< /*FORMAT_32_32_32_32_UINT*/ 0x4B, 32, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32_32_32*/ 14>;
def : Gfx10BufferFormat< /*FORMAT_32_32_32_32_SINT*/ 0x4C, 32, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32_32_32*/ 14>;
def : Gfx10BufferFormat< /*FORMAT_32_32_32_32_FLOAT*/ 0x4D, 32, 4, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32_32_32*/ 14>;
// Buffer formats with equal component sizes (GFX11 and later)
def : Gfx11PlusBufferFormat< /*FORMAT_8_8_8_8_UNORM*/ 0x2A, 8, 4, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx11PlusBufferFormat< /*FORMAT_8_8_8_8_SNORM*/ 0x2B, 8, 4, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx11PlusBufferFormat< /*FORMAT_8_8_8_8_USCALED*/ 0x2C, 8, 4, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx11PlusBufferFormat< /*FORMAT_8_8_8_8_SSCALED*/ 0x2D, 8, 4, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx11PlusBufferFormat< /*FORMAT_8_8_8_8_UINT*/ 0x2E, 8, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx11PlusBufferFormat< /*FORMAT_8_8_8_8_SINT*/ 0x2F, 8, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_8_8_8_8*/ 10>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_UINT*/ 0x30, 32, 2, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_SINT*/ 0x31, 32, 2, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_FLOAT*/ 0x32, 32, 2, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32*/ 11>;
def : Gfx11PlusBufferFormat< /*FORMAT_16_16_16_16_UNORM*/ 0x33, 16, 4, /*NUM_FORMAT_UNORM*/ 0, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx11PlusBufferFormat< /*FORMAT_16_16_16_16_SNORM*/ 0x34, 16, 4, /*NUM_FORMAT_SNORM*/ 1, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx11PlusBufferFormat< /*FORMAT_16_16_16_16_USCALED*/ 0x35, 16, 4, /*NUM_FORMAT_USCALED*/ 2, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx11PlusBufferFormat< /*FORMAT_16_16_16_16_SSCALED*/ 0x36, 16, 4, /*NUM_FORMAT_SSCALED*/ 3, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx11PlusBufferFormat< /*FORMAT_16_16_16_16_UINT*/ 0x37, 16, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx11PlusBufferFormat< /*FORMAT_16_16_16_16_SINT*/ 0x38, 16, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx11PlusBufferFormat< /*FORMAT_16_16_16_16_FLOAT*/ 0x39, 16, 4, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_16_16_16_16*/ 12>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_32_UINT*/ 0x3A, 32, 3, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_32_SINT*/ 0x3B, 32, 3, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_32_FLOAT*/ 0x3C, 32, 3, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32_32*/ 13>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_32_32_UINT*/ 0x3D, 32, 4, /*NUM_FORMAT_UINT*/ 4, /*DATA_FORMAT_32_32_32_32*/ 14>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_32_32_SINT*/ 0x3E, 32, 4, /*NUM_FORMAT_SINT*/ 5, /*DATA_FORMAT_32_32_32_32*/ 14>;
def : Gfx11PlusBufferFormat< /*FORMAT_32_32_32_32_FLOAT*/ 0x3F, 32, 4, /*NUM_FORMAT_FLOAT*/ 7, /*DATA_FORMAT_32_32_32_32*/ 14>;
class SourceOfDivergence<Intrinsic intr> {
Intrinsic Intr = intr;
}
def SourcesOfDivergence : GenericTable {
let FilterClass = "SourceOfDivergence";
let Fields = ["Intr"];
let PrimaryKey = ["Intr"];
let PrimaryKeyName = "lookupSourceOfDivergence";
}
def : SourceOfDivergence<int_amdgcn_workitem_id_x>;
def : SourceOfDivergence<int_amdgcn_workitem_id_y>;
def : SourceOfDivergence<int_amdgcn_workitem_id_z>;
def : SourceOfDivergence<int_amdgcn_interp_mov>;
def : SourceOfDivergence<int_amdgcn_interp_p1>;
def : SourceOfDivergence<int_amdgcn_interp_p2>;
def : SourceOfDivergence<int_amdgcn_interp_p1_f16>;
def : SourceOfDivergence<int_amdgcn_interp_p2_f16>;
def : SourceOfDivergence<int_amdgcn_lds_direct_load>;
def : SourceOfDivergence<int_amdgcn_lds_param_load>;
def : SourceOfDivergence<int_amdgcn_mbcnt_hi>;
def : SourceOfDivergence<int_amdgcn_mbcnt_lo>;
def : SourceOfDivergence<int_r600_read_tidig_x>;
def : SourceOfDivergence<int_r600_read_tidig_y>;
def : SourceOfDivergence<int_r600_read_tidig_z>;
def : SourceOfDivergence<int_amdgcn_atomic_cond_sub_u32>;
def : SourceOfDivergence<int_amdgcn_global_atomic_csub>;
def : SourceOfDivergence<int_amdgcn_global_atomic_fmin_num>;
def : SourceOfDivergence<int_amdgcn_global_atomic_fmax_num>;
def : SourceOfDivergence<int_amdgcn_global_atomic_ordered_add_b64>;
def : SourceOfDivergence<int_amdgcn_flat_atomic_fmin_num>;
def : SourceOfDivergence<int_amdgcn_flat_atomic_fmax_num>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_swap>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_add>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_sub>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_smin>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_umin>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_smax>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_umax>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_and>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_or>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_xor>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_inc>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_dec>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_fadd>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_fmin>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_fmax>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_cmpswap>;
def : SourceOfDivergence<int_amdgcn_raw_buffer_atomic_cond_sub_u32>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_swap>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_add>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_sub>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_smin>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_umin>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_smax>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_umax>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_and>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_or>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_xor>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_inc>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_dec>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_fadd>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_fmin>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_fmax>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_cmpswap>;
def : SourceOfDivergence<int_amdgcn_raw_ptr_buffer_atomic_cond_sub_u32>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_swap>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_add>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_sub>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_smin>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_umin>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_smax>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_umax>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_and>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_or>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_xor>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_inc>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_dec>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_fadd>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_fmin>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_fmax>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_cmpswap>;
def : SourceOfDivergence<int_amdgcn_struct_buffer_atomic_cond_sub_u32>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_swap>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_add>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_sub>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_smin>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_umin>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_smax>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_umax>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_and>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_or>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_xor>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_inc>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_dec>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_fadd>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_fmin>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_fmax>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_cmpswap>;
def : SourceOfDivergence<int_amdgcn_struct_ptr_buffer_atomic_cond_sub_u32>;
def : SourceOfDivergence<int_amdgcn_ps_live>;
def : SourceOfDivergence<int_amdgcn_live_mask>;
def : SourceOfDivergence<int_amdgcn_ds_swizzle>;
def : SourceOfDivergence<int_amdgcn_ds_ordered_add>;
def : SourceOfDivergence<int_amdgcn_ds_ordered_swap>;
def : SourceOfDivergence<int_amdgcn_permlane16>;
def : SourceOfDivergence<int_amdgcn_permlanex16>;
def : SourceOfDivergence<int_amdgcn_permlane16_var>;
def : SourceOfDivergence<int_amdgcn_permlanex16_var>;
def : SourceOfDivergence<int_amdgcn_permlane_bcast>;
def : SourceOfDivergence<int_amdgcn_permlane_up>;
def : SourceOfDivergence<int_amdgcn_permlane_down>;
def : SourceOfDivergence<int_amdgcn_permlane_xor>;
def : SourceOfDivergence<int_amdgcn_permlane_idx_gen>;
def : SourceOfDivergence<int_amdgcn_mov_dpp>;
def : SourceOfDivergence<int_amdgcn_mov_dpp8>;
def : SourceOfDivergence<int_amdgcn_update_dpp>;
def : SourceOfDivergence<int_amdgcn_writelane>;
def : SourceOfDivergence<int_amdgcn_init_whole_wave>;
def : SourceOfDivergence<int_amdgcn_permlane16_swap>;
def : SourceOfDivergence<int_amdgcn_permlane32_swap>;
foreach intr = AMDGPUMFMAIntrinsics908 in
def : SourceOfDivergence<intr>;
foreach intr = AMDGPUMFMAIntrinsics90A in
def : SourceOfDivergence<intr>;
foreach intr = AMDGPUMFMAIntrinsics940 in
def : SourceOfDivergence<intr>;
foreach intr = AMDGPUMFMAIntrinsics950 in
def : SourceOfDivergence<intr>;
foreach intr = AMDGPUWMMAIntrinsicsGFX11 in
def : SourceOfDivergence<intr>;
foreach intr = AMDGPUWMMAIntrinsicsGFX12 in
def : SourceOfDivergence<intr>;
foreach intr = AMDGPUWMMAIntrinsicsGFX1250 in
def : SourceOfDivergence<intr>;
foreach intr = AMDGPUSWMMACIntrinsicsGFX1250 in
def : SourceOfDivergence<intr>;
def : SourceOfDivergence<int_amdgcn_global_load_tr_b64>;
def : SourceOfDivergence<int_amdgcn_global_load_tr_b128>;
def : SourceOfDivergence<int_amdgcn_global_load_tr4_b64>;
def : SourceOfDivergence<int_amdgcn_global_load_tr6_b96>;
def : SourceOfDivergence<int_amdgcn_ds_load_tr8_b64>;
def : SourceOfDivergence<int_amdgcn_ds_load_tr16_b128>;
def : SourceOfDivergence<int_amdgcn_ds_load_tr4_b64>;
def : SourceOfDivergence<int_amdgcn_ds_load_tr6_b96>;
def : SourceOfDivergence<int_amdgcn_ds_read_tr4_b64>;
def : SourceOfDivergence<int_amdgcn_ds_read_tr6_b96>;
def : SourceOfDivergence<int_amdgcn_ds_read_tr8_b64>;
def : SourceOfDivergence<int_amdgcn_ds_read_tr16_b64>;
// The dummy boolean output is divergent from the IR's perspective,
// but the mask results are uniform. These produce a divergent and
// uniform result, so the returned struct is collectively divergent.
// isAlwaysUniform can override the extract of the uniform component.
def : SourceOfDivergence<int_amdgcn_if>;
def : SourceOfDivergence<int_amdgcn_else>;
def : SourceOfDivergence<int_amdgcn_loop>;
def : SourceOfDivergence<int_amdgcn_inverse_ballot>;
foreach intr = AMDGPUImageDimAtomicIntrinsics in
def : SourceOfDivergence<intr>;
def : SourceOfDivergence<int_amdgcn_dead>;
def : SourceOfDivergence<int_amdgcn_call_whole_wave>;
class AlwaysUniform<Intrinsic intr> {
Intrinsic Intr = intr;
}
def UniformIntrinsics : GenericTable {
let FilterClass = "AlwaysUniform";
let Fields = ["Intr"];
let PrimaryKey = ["Intr"];
let PrimaryKeyName = "lookupAlwaysUniform";
}
def : AlwaysUniform<int_amdgcn_readfirstlane>;
def : AlwaysUniform<int_amdgcn_readlane>;
def : AlwaysUniform<int_amdgcn_icmp>;
def : AlwaysUniform<int_amdgcn_fcmp>;
def : AlwaysUniform<int_amdgcn_ballot>;
def : AlwaysUniform<int_amdgcn_if_break>;
def : AlwaysUniform<int_amdgcn_workgroup_id_x>;
def : AlwaysUniform<int_amdgcn_workgroup_id_y>;
def : AlwaysUniform<int_amdgcn_workgroup_id_z>;
def : AlwaysUniform<int_amdgcn_s_getpc>;
def : AlwaysUniform<int_amdgcn_s_getreg>;
def : AlwaysUniform<int_amdgcn_s_memrealtime>;
def : AlwaysUniform<int_amdgcn_s_memtime>;
Reference in New Issue View Git Blame Copy Permalink