Skip to content

[X86][AMX] Support AMX-AVX512 #114070

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 7 commits into from
Nov 8, 2024
Merged

[X86][AMX] Support AMX-AVX512 #114070

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
587d010
[X86][AMX] Support AMX-AVX512
phoebewang Oct 29, 2024
65b821f
Merge remote-tracking branch 'origin/main' into AMX-AVX512
phoebewang Nov 6, 2024
c38da4e
Address review comments
phoebewang Nov 6, 2024
1b93a9a
Add avx10.2-512 predicate
phoebewang Nov 7, 2024
6560806
Add missing parts
phoebewang Nov 8, 2024
6a78b18
Add avx10.2-512
phoebewang Nov 8, 2024
ffcb6a7
Add HasAVX10_2_512
phoebewang Nov 8, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions clang/docs/ReleaseNotes.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -679,6 +679,7 @@ X86 Support
* Supported intrinsics of ``_mm(256|512)_(mask(z))_loadrs_epi(8|16|32|64)``.
- Support ISA of ``AMX-FP8``.
- Support ISA of ``AMX-TRANSPOSE``.
- Support ISA of ``AMX-AVX512``.

Arm and AArch64 Support
^^^^^^^^^^^^^^^^^^^^^^^
Expand Down
13 changes: 13 additions & 0 deletions clang/include/clang/Basic/BuiltinsX86_64.def
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,6 +133,12 @@ TARGET_BUILTIN(__builtin_ia32_t2rpntlvwz0t1_internal, "vUsUsUsV256i*V256i*vC*z",
TARGET_BUILTIN(__builtin_ia32_t2rpntlvwz1_internal, "vUsUsUsV256i*V256i*vC*z", "n", "amx-transpose")
TARGET_BUILTIN(__builtin_ia32_t2rpntlvwz1t1_internal, "vUsUsUsV256i*V256i*vC*z", "n", "amx-transpose")
TARGET_BUILTIN(__builtin_ia32_ttransposed_internal, "V256iUsUsV256i", "n", "amx-transpose")
TARGET_BUILTIN(__builtin_ia32_tcvtrowd2ps_internal, "V16fUsUsV256iUi", "n", "amx-avx512")
Copy link
Contributor

@fzou1 fzou1 Nov 7, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is "avx10.2-512" feature needed to be added for the intrinsics here and there?

Copy link
Contributor Author

@phoebewang phoebewang Nov 7, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good catch! Done.

Copy link
Contributor

@fzou1 fzou1 Nov 8, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is it necessary to add avx10.2-512 feature for these internal APIs? With that, we may detect errors for new APIs if there is no AVX10.2-512 target feature.

Copy link
Contributor Author

@phoebewang phoebewang Nov 8, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, sorry I missed that parts.

TARGET_BUILTIN(__builtin_ia32_tcvtrowps2pbf16h_internal, "V32yUsUsV256iUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tcvtrowps2pbf16l_internal, "V32yUsUsV256iUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tcvtrowps2phh_internal, "V32xUsUsV256iUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tcvtrowps2phl_internal, "V32xUsUsV256iUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tilemovrow_internal, "V16iUsUsV256iUi", "n", "amx-avx512")
// AMX
TARGET_BUILTIN(__builtin_ia32_tile_loadconfig, "vvC*", "n", "amx-tile")
TARGET_BUILTIN(__builtin_ia32_tile_storeconfig, "vvC*", "n", "amx-tile")
Expand All @@ -159,6 +165,13 @@ TARGET_BUILTIN(__builtin_ia32_t2rpntlvwz1, "vIUcvC*z", "n", "amx-transpose")
TARGET_BUILTIN(__builtin_ia32_t2rpntlvwz1t1, "vIUcvC*z", "n","amx-transpose")
TARGET_BUILTIN(__builtin_ia32_ttransposed, "vIUcIUc", "n", "amx-transpose")

TARGET_BUILTIN(__builtin_ia32_tcvtrowd2ps, "V16fIUcUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tcvtrowps2pbf16h, "V32yIUcUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tcvtrowps2pbf16l, "V32yIUcUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tcvtrowps2phh, "V32xIUcUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tcvtrowps2phl, "V32xIUcUi", "n", "amx-avx512")
TARGET_BUILTIN(__builtin_ia32_tilemovrow, "V16iIUcUi", "n", "amx-avx512")

TARGET_BUILTIN(__builtin_ia32_prefetchi, "vvC*Ui", "nc", "prefetchi")
TARGET_BUILTIN(__builtin_ia32_cmpccxadd32, "Siv*SiSiIi", "n", "cmpccxadd")
TARGET_BUILTIN(__builtin_ia32_cmpccxadd64, "SLLiSLLi*SLLiSLLiIi", "n", "cmpccxadd")
Expand Down
2 changes: 2 additions & 0 deletions clang/include/clang/Driver/Options.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6289,6 +6289,8 @@ def mno_80387 : Flag<["-"], "mno-80387">, Alias<mno_x87>;
def mno_fp_ret_in_387 : Flag<["-"], "mno-fp-ret-in-387">, Alias<mno_x87>;
def mmmx : Flag<["-"], "mmmx">, Group<m_x86_Features_Group>;
def mno_mmx : Flag<["-"], "mno-mmx">, Group<m_x86_Features_Group>;
def mamx_avx512 : Flag<["-"], "mamx-avx512">, Group<m_x86_Features_Group>;
def mno_amx_avx512 : Flag<["-"], "mno-amx-avx512">, Group<m_x86_Features_Group>;
def mamx_bf16 : Flag<["-"], "mamx-bf16">, Group<m_x86_Features_Group>;
def mno_amx_bf16 : Flag<["-"], "mno-amx-bf16">, Group<m_x86_Features_Group>;
def mamx_complex : Flag<["-"], "mamx-complex">, Group<m_x86_Features_Group>;
Expand Down
6 changes: 6 additions & 0 deletions clang/lib/Basic/Targets/X86.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -432,6 +432,8 @@ bool X86TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
HasAMXFP8 = true;
} else if (Feature == "+amx-transpose") {
HasAMXTRANSPOSE = true;
} else if (Feature == "+amx-avx512") {
HasAMXAVX512 = true;
} else if (Feature == "+cmpccxadd") {
HasCMPCCXADD = true;
} else if (Feature == "+raoint") {
Expand Down Expand Up @@ -955,6 +957,8 @@ void X86TargetInfo::getTargetDefines(const LangOptions &Opts,
Builder.defineMacro("__AMX_FP8__");
if (HasAMXTRANSPOSE)
Builder.defineMacro("__AMX_TRANSPOSE__");
if (HasAMXAVX512)
Builder.defineMacro("__AMX_AVX512__");
if (HasCMPCCXADD)
Builder.defineMacro("__CMPCCXADD__");
if (HasRAOINT)
Expand Down Expand Up @@ -1080,6 +1084,7 @@ bool X86TargetInfo::isValidFeatureName(StringRef Name) const {
return llvm::StringSwitch<bool>(Name)
.Case("adx", true)
.Case("aes", true)
.Case("amx-avx512", true)
.Case("amx-bf16", true)
.Case("amx-complex", true)
.Case("amx-fp16", true)
Expand Down Expand Up @@ -1200,6 +1205,7 @@ bool X86TargetInfo::hasFeature(StringRef Feature) const {
return llvm::StringSwitch<bool>(Feature)
.Case("adx", HasADX)
.Case("aes", HasAES)
.Case("amx-avx512", HasAMXAVX512)
.Case("amx-bf16", HasAMXBF16)
.Case("amx-complex", HasAMXCOMPLEX)
.Case("amx-fp16", HasAMXFP16)
Expand Down
1 change: 1 addition & 0 deletions clang/lib/Basic/Targets/X86.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,6 +159,7 @@ class LLVM_LIBRARY_VISIBILITY X86TargetInfo : public TargetInfo {
bool HasAMXCOMPLEX = false;
bool HasAMXFP8 = false;
bool HasAMXTRANSPOSE = false;
bool HasAMXAVX512 = false;
bool HasSERIALIZE = false;
bool HasTSXLDTRK = false;
bool HasUSERMSR = false;
Expand Down
1 change: 1 addition & 0 deletions clang/lib/Headers/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,6 +146,7 @@ set(x86_files
adcintrin.h
adxintrin.h
ammintrin.h
amxavx512intrin.h
amxcomplexintrin.h
amxfp16intrin.h
amxfp8intrin.h
Expand Down
Loading
Loading