https://github.com/KungFuDonkey updated https://github.com/llvm/llvm-project/pull/185383
>From b7b66542448dd5b3b66fe428ad3b98c39f2317a7 Mon Sep 17 00:00:00 2001 From: KungFuDonkey <[email protected]> Date: Sun, 8 Mar 2026 16:32:23 +0100 Subject: [PATCH 1/5] Added GroupMemoryBarrier --- clang/include/clang/Basic/Builtins.td | 285 ++++++++++-------- clang/lib/CodeGen/CGHLSLBuiltins.cpp | 5 + clang/lib/CodeGen/CGHLSLRuntime.h | 2 + .../lib/Headers/hlsl/hlsl_alias_intrinsics.h | 11 + .../builtins/GroupMemoryBarrier.hlsl | 20 ++ .../SemaHLSL/BuiltIns/GroupMemoryBarrier.hlsl | 6 + llvm/include/llvm/IR/IntrinsicsDirectX.td | 3 + llvm/include/llvm/IR/IntrinsicsSPIRV.td | 1 + llvm/lib/Target/DirectX/DXIL.td | 2 + .../Target/SPIRV/SPIRVInstructionSelector.cpp | 178 ++++++----- .../CodeGen/DirectX/group_memory_barrier.ll | 8 + .../hlsl-intrinsics/group_memory_barrier.ll | 14 + 12 files changed, 333 insertions(+), 202 deletions(-) create mode 100644 clang/test/CodeGenHLSL/builtins/GroupMemoryBarrier.hlsl create mode 100644 clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier.hlsl create mode 100644 llvm/test/CodeGen/DirectX/group_memory_barrier.ll create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td index 4981711fe786d..c98236e4258d4 100644 --- a/clang/include/clang/Basic/Builtins.td +++ b/clang/include/clang/Basic/Builtins.td @@ -8,68 +8,64 @@ include "clang/Basic/BuiltinsBase.td" -class FPMathTemplate : Template<["float", "double", "long double"], - ["f", "", "l"]>; +class FPMathTemplate + : Template<["float", "double", "long double"], ["f", "", "l"]>; -class FPMathWithF16Template : - Template<["float", "double", "long double", "__fp16"], - ["f", "", "l", "f16"]>; +class FPMathWithF16Template + : Template<["float", "double", "long double", "__fp16"], ["f", "", "l", + "f16"]>; -class FPMathWithF16F128Template : - Template<["float", "double", "long double", "__fp16", "__float128"], - ["f", "", "l", "f16", "f128"]>; +class FPMathWithF16F128Template + : Template<["float", "double", "long double", "__fp16", "__float128"], + ["f", "", "l", "f16", "f128"]>; -class FPMathWithF128Template : - Template<["float", "double", "long double", "__float128"], - ["f", "", "l", "f128"]>; +class FPMathWithF128Template + : Template<["float", "double", "long double", "__float128"], ["f", "", "l", + "f128"]>; -class F16F128MathTemplate : Template<["__fp16", "__float128"], - ["f16", "f128"]>; +class F16F128MathTemplate : Template<["__fp16", "__float128"], ["f16", "f128"]>; -class IntMathTemplate : Template<["int", "long int", "long long int"], - ["", "l", "ll"], /*AsPrefix=*/1>; +class IntMathTemplate + : Template<["int", "long int", "long long int"], ["", "l", "ll"], + /*AsPrefix=*/1>; -class MSInt8_16_32Template : Template<["char", "short", "msint32_t"], - ["8", "16", ""]>; +class MSInt8_16_32Template + : Template<["char", "short", "msint32_t"], ["8", "16", ""]>; class Int8_16_32_64Template - : Template<["char", "short", "int", "long long int"], - ["8", "16", "32", "64"]>; + : Template<["char", "short", "int", "long long int"], ["8", "16", "32", + "64"]>; class MSInt8_16_32_64Template - : Template<["char", "short", "msint32_t", "long long int"], - ["8", "16", "", "64"]>; + : Template<["char", "short", "msint32_t", "long long int"], ["8", "16", "", + "64"]>; -class MSInt16_32Template : Template<["short", "msint32_t"], - ["16", ""]>; +class MSInt16_32Template : Template<["short", "msint32_t"], ["16", ""]>; -class MSUInt16_32_64Template : - Template<["unsigned short", "unsigned int", "uint64_t"], - ["16", "", "64"]>; +class MSUInt16_32_64Template + : Template<["unsigned short", "unsigned int", "uint64_t"], ["16", "", + "64"]>; -class MSInt32_64Template : Template<["msint32_t", "int64_t"], - ["", "64"]>; +class MSInt32_64Template : Template<["msint32_t", "int64_t"], ["", "64"]>; -class FloatDoubleTemplate : Template<["float", "double"], - ["f", ""]>; +class FloatDoubleTemplate : Template<["float", "double"], ["f", ""]>; // FIXME: These assume that char -> i8, short -> i16, int -> i32, // long long -> i64. -class SyncBuiltinsTemplate : - Template<["char", "short", "int", "long long int", "__int128_t"], - ["1", "2", "4", "8", "16"]>; +class SyncBuiltinsTemplate + : Template<["char", "short", "int", "long long int", "__int128_t"], + ["1", "2", "4", "8", "16"]>; -class BitInt8_16_32_64BuiltinsTemplate : - Template<["unsigned char", "unsigned short", "uint32_t", "uint64_t"], - ["8", "16", "32", "64"]>; +class BitInt8_16_32_64BuiltinsTemplate + : Template<["unsigned char", "unsigned short", "uint32_t", "uint64_t"], + ["8", "16", "32", "64"]>; -class BitShort_Int_Long_LongLongTemplate : - Template<["short", "int", "long int", "long long int"], - ["s", "", "l", "ll"]>; +class BitShort_Int_Long_LongLongTemplate + : Template<["short", "int", "long int", "long long int"], ["s", "", "l", + "ll"]>; -class BitInt_Long_LongLongTemplate : - Template<["int", "long int", "long long int"], - ["", "l", "ll"]>; +class BitInt_Long_LongLongTemplate + : Template<["int", "long int", "long long int"], ["", "l", "ll"]>; // Most of the types used in the prototypes are types from C, C++ or ObjC. There // are a few builtin-specific types and qualifiers. @@ -139,55 +135,64 @@ def CeilF16F128 : Builtin, F16F128MathTemplate { def CosF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_cos"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def CoshF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_cosh"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def ErfF128 : Builtin { let Spellings = ["__builtin_erff128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def ErfcF128 : Builtin { let Spellings = ["__builtin_erfcf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def ExpF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_exp"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Exp2F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_exp2"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Exp10F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_exp10"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Expm1F128 : Builtin { let Spellings = ["__builtin_expm1f128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def FdimF128 : Builtin { let Spellings = ["__builtin_fdimf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } @@ -199,7 +204,8 @@ def FloorF16F128 : Builtin, F16F128MathTemplate { def FmaF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_fma"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T, T)"; } @@ -229,7 +235,8 @@ def FminimumNumF16F128 : Builtin, F16F128MathTemplate { def Atan2F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_atan2"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T)"; } @@ -259,7 +266,8 @@ def FabsF128 : Builtin { def FmodF16F128 : F16F128MathTemplate, Builtin { let Spellings = ["__builtin_fmod"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T)"; } @@ -276,7 +284,7 @@ def HugeVal : Builtin, FPMathWithF128Template { } def HugeValF16 : Builtin { - let Spellings = ["__builtin_huge_valf16"]; + let Spellings = ["__builtin_huge_valf16"]; let Attributes = [NoThrow, Const, Constexpr]; let Prototype = "_Float16()"; } @@ -295,7 +303,8 @@ def InfF16 : Builtin { def LdexpF16F128 : F16F128MathTemplate, Builtin { let Spellings = ["__builtin_ldexp"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, int)"; } @@ -319,9 +328,10 @@ def NanF128 : Builtin { let Prototype = "__float128(char const*)"; } -def Nans : Builtin, - Template<["float", "double", "long double", "_Float16", "__float128"], - ["f", "", "l", "f16", "f128"]> { +def Nans + : Builtin, + Template<["float", "double", "long double", "_Float16", "__float128"], + ["f", "", "l", "f16", "f128"]> { let Spellings = ["__builtin_nans"]; let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Pure, Constexpr]; let Prototype = "T(char const*)"; @@ -335,19 +345,22 @@ def PowI : Builtin, FPMathTemplate { def PowF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_pow"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T)"; } def HypotF128 : Builtin { let Spellings = ["__builtin_hypotf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } def ILogbF128 : Builtin { let Spellings = ["__builtin_ilogbf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "int(__float128)"; } @@ -359,55 +372,64 @@ def LgammaF128 : Builtin { def LLrintF128 : Builtin { let Spellings = ["__builtin_llrintf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "long long int(__float128)"; } def LLroundF128 : Builtin { let Spellings = ["__builtin_llroundf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "long long int(__float128)"; } def Log10F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_log10"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Log1pF128 : Builtin { let Spellings = ["__builtin_log1pf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def Log2F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_log2"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def LogbF128 : Builtin { let Spellings = ["__builtin_logbf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def LogF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_log"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def LrintF128 : Builtin { let Spellings = ["__builtin_lrintf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "long int(__float128)"; } def LroundF128 : Builtin { let Spellings = ["__builtin_lroundf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "long int(__float128)"; } @@ -419,19 +441,22 @@ def NearbyintF128 : Builtin { def NextafterF128 : Builtin { let Spellings = ["__builtin_nextafterf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } def NexttowardF128 : Builtin { let Spellings = ["__builtin_nexttowardf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } def RemainderF128 : Builtin { let Spellings = ["__builtin_remainderf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, + ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } @@ -670,15 +695,13 @@ def Signbit : Builtin { def SignbitF : Builtin { let Spellings = ["__builtin_signbitf"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, - Constexpr]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, Constexpr]; let Prototype = "int(float)"; } def SignbitL : Builtin { let Spellings = ["__builtin_signbitl"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, - Constexpr]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, Constexpr]; let Prototype = "int(long double)"; } @@ -748,8 +771,9 @@ def Clrsb : Builtin, BitInt_Long_LongLongTemplate { // there exists native types on the target that are 32- and 64-bits wide, unless // these conditions are fulfilled these builtins will operate on a not intended // bitwidth. -def BSwap : Builtin, Template<["unsigned short", "uint32_t", "uint64_t"], - ["16", "32", "64"]> { +def BSwap + : Builtin, + Template<["unsigned short", "uint32_t", "uint64_t"], ["16", "32", "64"]> { let Spellings = ["__builtin_bswap"]; let Attributes = [NoThrow, Const, Constexpr]; let Prototype = "T(T)"; @@ -829,13 +853,15 @@ def BuiltinCalloc : Builtin { def BuiltinConstantP : Builtin { let Spellings = ["__builtin_constant_p"]; - let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, Constexpr]; + let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, + Constexpr]; let Prototype = "int(...)"; } def BuiltinClassifyType : Builtin { let Spellings = ["__builtin_classify_type"]; - let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, Constexpr]; + let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, + Constexpr]; let Prototype = "int(...)"; } @@ -1139,7 +1165,8 @@ def StpncpyChk : Builtin { def SNPrintfChk : Builtin { let Spellings = ["__builtin___snprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, PrintfFormat<4>]; - let Prototype = "int(char* restrict, size_t, int, size_t, char const* restrict, ...)"; + let Prototype = + "int(char* restrict, size_t, int, size_t, char const* restrict, ...)"; } def SPrintfChk : Builtin { @@ -1151,13 +1178,15 @@ def SPrintfChk : Builtin { def VSNPrintfChk : Builtin { let Spellings = ["__builtin___vsnprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, VPrintfFormat<4>]; - let Prototype = "int(char* restrict, size_t, int, size_t, char const* restrict, __builtin_va_list)"; + let Prototype = "int(char* restrict, size_t, int, size_t, char const* " + "restrict, __builtin_va_list)"; } def VSPrintfChk : Builtin { let Spellings = ["__builtin___vsprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, VPrintfFormat<3>]; - let Prototype = "int(char* restrict, int, size_t, char const* restrict, __builtin_va_list)"; + let Prototype = "int(char* restrict, int, size_t, char const* restrict, " + "__builtin_va_list)"; } def FPrintfChk : Builtin { @@ -1175,7 +1204,8 @@ def PrintfChk : Builtin { def VFPrintfChk : Builtin { let Spellings = ["__builtin___vfprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, VPrintfFormat<2>]; - let Prototype = "int(FILE* restrict, int, char const* restrict, __builtin_va_list)"; + let Prototype = + "int(FILE* restrict, int, char const* restrict, __builtin_va_list)"; } def VPrintfChk : Builtin { @@ -2553,7 +2583,8 @@ def SyncFetchAndUMax : Builtin { let Prototype = "unsigned int(unsigned int volatile*, unsigned int)"; } -// ignored glibc builtin, see https://sourceware.org/bugzilla/show_bug.cgi?id=25399 +// ignored glibc builtin, see +// https://sourceware.org/bugzilla/show_bug.cgi?id=25399 def WarnMemsetZeroLen : Builtin { let Spellings = ["__warn_memset_zero_len"]; let Attributes = [NoThrow, Pure]; @@ -2603,9 +2634,10 @@ def BittestAndSet : MSLangBuiltin, MSInt32_64Template { let Prototype = "unsigned char(T*, T)"; } -def MSByteswap : MSLibBuiltin<"stdlib.h">, - Template<["unsigned short", "msuint32_t", "unsigned long long int"], - ["_ushort", "_ulong", "_uint64"]> { +def MSByteswap + : MSLibBuiltin<"stdlib.h">, + Template<["unsigned short", "msuint32_t", "unsigned long long int"], + ["_ushort", "_ulong", "_uint64"]> { let Spellings = ["_byteswap"]; let Attributes = [NoThrow, Const]; let Prototype = "T(T)"; @@ -3216,21 +3248,24 @@ def VPrintf : LibBuiltin<"stdio.h"> { def VfPrintf : LibBuiltin<"stdio.h"> { let Spellings = ["vfprintf"]; let Attributes = [NoThrow, VPrintfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = "int(FILE* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = + "int(FILE* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } def VsnPrintf : LibBuiltin<"stdio.h"> { let Spellings = ["vsnprintf"]; let Attributes = [NoThrow, VPrintfFormat<2>, NonNull<NonOptimizing, [2]>]; - let Prototype = "int(char* restrict, size_t, char const* restrict, __builtin_va_list)"; + let Prototype = + "int(char* restrict, size_t, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } def VsPrintf : LibBuiltin<"stdio.h"> { let Spellings = ["vsprintf"]; let Attributes = [NoThrow, VPrintfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = "int(char* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = + "int(char* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } @@ -3265,14 +3300,16 @@ def VScanf : LibBuiltin<"stdio.h"> { def VFScanf : LibBuiltin<"stdio.h"> { let Spellings = ["vfscanf"]; let Attributes = [VScanfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = "int(FILE* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = + "int(FILE* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } def VSScanf : LibBuiltin<"stdio.h"> { let Spellings = ["vsscanf"]; let Attributes = [VScanfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = "int(char const* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = + "int(char const* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } @@ -4492,11 +4529,9 @@ def AssumeSeparateStorage : Builtin { // Multiprecision Arithmetic Builtins. -class MPATemplate : Template< - ["unsigned char", "unsigned short", "unsigned int", - "unsigned long int", "unsigned long long int"], - ["b", "s", "", - "l", "ll"]>; +class MPATemplate : Template<["unsigned char", "unsigned short", "unsigned int", + "unsigned long int", "unsigned long long int"], + ["b", "s", "", "l", "ll"]>; def Addc : Builtin, MPATemplate { let Spellings = ["__builtin_addc"]; @@ -4531,9 +4566,9 @@ def MulOverflow : Builtin { let Prototype = "bool(...)"; } -class UOverflowTemplate : - Template<["unsigned int", "unsigned long int", "unsigned long long int"], - ["_overflow", "l_overflow", "ll_overflow"]>; +class UOverflowTemplate + : Template<["unsigned int", "unsigned long int", "unsigned long long int"], + ["_overflow", "l_overflow", "ll_overflow"]>; def UaddOverflow : Builtin, UOverflowTemplate { let Spellings = ["__builtin_uadd"]; @@ -4553,9 +4588,9 @@ def UmulOverflow : Builtin, UOverflowTemplate { let Prototype = "bool(T const, T const, T*)"; } -class SOverflowTemplate : - Template<["int", "long int", "long long int"], - ["_overflow", "l_overflow", "ll_overflow"]>; +class SOverflowTemplate + : Template<["int", "long int", "long long int"], ["_overflow", "l_overflow", + "ll_overflow"]>; def SaddOverflow : Builtin, SOverflowTemplate { let Spellings = ["__builtin_sadd"]; @@ -4815,7 +4850,8 @@ def PtrauthStringDiscriminator : Builtin { // AllocToken builtins. def InferAllocToken : Builtin { let Spellings = ["__builtin_infer_alloc_token"]; - let Attributes = [NoThrow, Const, Pure, CustomTypeChecking, Constexpr, UnevaluatedArguments]; + let Attributes = [NoThrow, Const, Pure, CustomTypeChecking, Constexpr, + UnevaluatedArguments]; let Prototype = "size_t(...)"; } @@ -4918,7 +4954,8 @@ def GetPipeMaxPackets : OCLPipeLangBuiltin { } // OpenCL v2.0 s6.13.17 - Enqueue kernel functions. -// Custom builtin check allows to perform special check of passed block arguments. +// Custom builtin check allows to perform special check of passed block +// arguments. def EnqueueKernel : OCL_DSELangBuiltin { let Spellings = ["enqueue_kernel"]; let Attributes = [CustomTypeChecking, NoThrow]; @@ -5006,7 +5043,7 @@ def GetDeviceSideMangledName : LangBuiltin<"CUDA_LANG"> { } // HLSL -def HLSLAddUint64: LangBuiltin<"HLSL_LANG"> { +def HLSLAddUint64 : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_adduint64"]; let Attributes = [NoThrow, Const]; let Prototype = "void(...)"; @@ -5093,13 +5130,15 @@ def HLSLResourceUninitializedHandle : LangBuiltin<"HLSL_LANG"> { def HLSLResourceHandleFromBinding : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_resource_handlefrombinding"]; let Attributes = [NoThrow]; - let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, int32_t, uint32_t, char const*)"; + let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, " + "int32_t, uint32_t, char const*)"; } def HLSLResourceHandleFromImplicitBinding : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_resource_handlefromimplicitbinding"]; let Attributes = [NoThrow]; - let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, int32_t, uint32_t, char const*)"; + let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, " + "int32_t, uint32_t, char const*)"; } def HLSLResourceCounterHandleFromImplicitBinding : LangBuiltin<"HLSL_LANG"> { @@ -5252,16 +5291,18 @@ def HLSLClamp : LangBuiltin<"HLSL_LANG"> { let Prototype = "void(...)"; } -def HLSLCrossFloat: LangBuiltin<"HLSL_LANG"> { +def HLSLCrossFloat : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_crossf32"]; let Attributes = [NoThrow, Const]; - let Prototype = "_ExtVector<3, float>(_ExtVector<3, float>, _ExtVector<3, float>)"; + let Prototype = + "_ExtVector<3, float>(_ExtVector<3, float>, _ExtVector<3, float>)"; } -def HLSLCrossHalf: LangBuiltin<"HLSL_LANG"> { +def HLSLCrossHalf : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_crossf16"]; let Attributes = [NoThrow, Const]; - let Prototype = "_ExtVector<3, __fp16>(_ExtVector<3, __fp16>, _ExtVector<3, __fp16>)"; + let Prototype = + "_ExtVector<3, __fp16>(_ExtVector<3, __fp16>, _ExtVector<3, __fp16>)"; } def HLSLDegrees : LangBuiltin<"HLSL_LANG"> { @@ -5372,7 +5413,7 @@ def HLSLSign : LangBuiltin<"HLSL_LANG"> { let Prototype = "void(...)"; } -def HLSLStep: LangBuiltin<"HLSL_LANG"> { +def HLSLStep : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_step"]; let Attributes = [NoThrow, Const]; let Prototype = "void(...)"; @@ -5390,19 +5431,25 @@ def HLSLBufferUpdateCounter : LangBuiltin<"HLSL_LANG"> { let Prototype = "uint32_t(__hlsl_resource_t, int)"; } -def HLSLSplitDouble: LangBuiltin<"HLSL_LANG"> { +def HLSLSplitDouble : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_elementwise_splitdouble"]; let Attributes = [NoThrow, Const]; let Prototype = "void(...)"; } -def HLSLClip: LangBuiltin<"HLSL_LANG"> { +def HLSLClip : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_elementwise_clip"]; let Attributes = [NoThrow, Const, CustomTypeChecking]; let Prototype = "void(...)"; } -def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> { +def HLSLGroupMemoryBarrier : LangBuiltin<"HLSL_LANG"> { + let Spellings = ["__builtin_hlsl_group_memory_barrier"]; + let Attributes = [NoThrow, Const]; + let Prototype = "void()"; +} + +def HLSLGroupMemoryBarrierWithGroupSync : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_group_memory_barrier_with_group_sync"]; let Attributes = [NoThrow, Const]; let Prototype = "void()"; diff --git a/clang/lib/CodeGen/CGHLSLBuiltins.cpp b/clang/lib/CodeGen/CGHLSLBuiltins.cpp index 177787d2a9630..ae7325384f9f8 100644 --- a/clang/lib/CodeGen/CGHLSLBuiltins.cpp +++ b/clang/lib/CodeGen/CGHLSLBuiltins.cpp @@ -1355,6 +1355,11 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID, assert(E->getArg(0)->getType()->hasFloatingRepresentation() && "clip operands types mismatch"); return handleHlslClip(E, this); + case Builtin::BI__builtin_hlsl_group_memory_barrier: { + Intrinsic::ID ID = CGM.getHLSLRuntime().getGroupMemoryBarrierIntrinsic(); + return EmitRuntimeCall( + Intrinsic::getOrInsertDeclaration(&CGM.getModule(), ID)); + } case Builtin::BI__builtin_hlsl_group_memory_barrier_with_group_sync: { Intrinsic::ID ID = CGM.getHLSLRuntime().getGroupMemoryBarrierWithGroupSyncIntrinsic(); diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h index 466c809fdef78..11d99de157ba8 100644 --- a/clang/lib/CodeGen/CGHLSLRuntime.h +++ b/clang/lib/CodeGen/CGHLSLRuntime.h @@ -184,6 +184,8 @@ class CGHLSLRuntime { GENERATE_HLSL_INTRINSIC_FUNCTION(NonUniformResourceIndex, resource_nonuniformindex) GENERATE_HLSL_INTRINSIC_FUNCTION(BufferUpdateCounter, resource_updatecounter) + GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrier, + group_memory_barrier) GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync, group_memory_barrier_with_group_sync) GENERATE_HLSL_INTRINSIC_FUNCTION(GetDimensionsX, resource_getdimensions_x) diff --git a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h index 7b6160091aece..f12b62df4733c 100644 --- a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h +++ b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h @@ -3411,6 +3411,17 @@ float3 radians(float3); _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_radians) float4 radians(float4); +//===----------------------------------------------------------------------===// +// GroupMemoryBarrierbuiltins +//===----------------------------------------------------------------------===// + +/// \fn void GroupMemoryBarrier(void) +/// \brief Blocks execution of all threads in a group until all group shared +/// accesses have been completed. + +_HLSL_BUILTIN_ALIAS(__builtin_hlsl_group_memory_barrier) +__attribute__((convergent)) void GroupMemoryBarrier(void); + //===----------------------------------------------------------------------===// // GroupMemoryBarrierWithGroupSync builtins //===----------------------------------------------------------------------===// diff --git a/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrier.hlsl b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrier.hlsl new file mode 100644 index 0000000000000..b52819973f677 --- /dev/null +++ b/clang/test/CodeGenHLSL/builtins/GroupMemoryBarrier.hlsl @@ -0,0 +1,20 @@ +// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ +// RUN: dxil-pc-shadermodel6.3-library %s \ +// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \ +// RUN: -DTARGET=dx -check-prefixes=CHECK,CHECK-DXIL +// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \ +// RUN: spirv-unknown-vulkan-compute %s \ +// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \ +// RUN: -DTARGET=spv -check-prefixes=CHECK,CHECK-SPIRV + +// CHECK-DXIL: define hidden void @ +// CHECK-SPIRV: define hidden spir_func void @ +void test_GroupMemoryBarrier() { +// CHECK-DXIL: call void @llvm.[[TARGET]].group.memory.barrier() +// CHECK-SPIRV: call spir_func void @llvm.[[TARGET]].group.memory.barrier() + GroupMemoryBarrier(); +} + +// CHECK: declare void @llvm.[[TARGET]].group.memory.barrier() #[[ATTRS:[0-9]+]] +// CHECK-NOT: attributes #[[ATTRS]] = {{.+}}memory(none){{.+}} +// CHECK: attributes #[[ATTRS]] = {{.+}}convergent{{.+}} diff --git a/clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier.hlsl b/clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier.hlsl new file mode 100644 index 0000000000000..5c5761c31eb90 --- /dev/null +++ b/clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier.hlsl @@ -0,0 +1,6 @@ +// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -emit-llvm-only -disable-llvm-passes -verify + +void test_too_many_arg() { + __builtin_hlsl_group_memory_barrier(0); + // expected-error@-1 {{too many arguments to function call, expected 0, have 1}} +} diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td index e2b2feb927318..3810fc340962d 100644 --- a/llvm/include/llvm/IR/IntrinsicsDirectX.td +++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td @@ -258,6 +258,9 @@ def int_dx_firstbituhigh : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, def int_dx_firstbitshigh : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_anyint_ty], [IntrNoMem]>; def int_dx_firstbitlow : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_anyint_ty], [IntrNoMem]>; +def int_dx_group_memory_barrier + : DefaultAttrsIntrinsic<[], [], [IntrConvergent]>; + def int_dx_group_memory_barrier_with_group_sync : DefaultAttrsIntrinsic<[], [], [IntrConvergent]>; diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td index 3fc18a254f672..e1328244ef7b0 100644 --- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td +++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td @@ -138,6 +138,7 @@ def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty] def int_spv_wave_prefix_product : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>], [IntrConvergent, IntrNoMem]>; def int_spv_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>; def int_spv_radians : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty], [IntrNoMem]>; + def int_spv_group_memory_barrier : DefaultAttrsIntrinsic<[], [], [IntrConvergent]>; def int_spv_group_memory_barrier_with_group_sync : ClangBuiltin<"__builtin_spirv_group_barrier">, DefaultAttrsIntrinsic<[], [], [IntrConvergent]>; def int_spv_discard : DefaultAttrsIntrinsic<[], [], []>; diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td index e64909b059d29..5ef1a7c130b4a 100644 --- a/llvm/lib/Target/DirectX/DXIL.td +++ b/llvm/lib/Target/DirectX/DXIL.td @@ -913,6 +913,8 @@ def GetDimensions : DXILOp<72, getDimensions> { def Barrier : DXILOp<80, barrier> { let Doc = "inserts a memory barrier in the shader"; let intrinsics = [ + IntrinSelect<int_dx_group_memory_barrier, + [IntrinArgI32<BarrierMode_GroupMemoryBarrier>]>, IntrinSelect<int_dx_group_memory_barrier_with_group_sync, [IntrinArgI32<BarrierMode_GroupMemoryBarrierWithGroupSync>]>, ]; diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp index 7b4c047593a3a..704c0dd6dc5d7 100644 --- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp @@ -336,6 +336,9 @@ class SPIRVInstructionSelector : public InstructionSelector { bool selectWaveOpInst(Register ResVReg, SPIRVTypeInst ResType, MachineInstr &I, unsigned Opcode) const; + bool selectBarrierInst(MachineInstr &I, unsigned Scope, + bool WithGroupSync) const; + bool selectWaveActiveCountBits(Register ResVReg, SPIRVTypeInst ResType, MachineInstr &I) const; @@ -1893,15 +1896,15 @@ bool SPIRVInstructionSelector::selectAtomicRMW(Register ResVReg, ValueReg = TmpReg; } - BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(NewOpcode)) - .addDef(ResVReg) - .addUse(GR.getSPIRVTypeID(ResType)) - .addUse(Ptr) - .addUse(ScopeReg) - .addUse(MemSemReg) - .addUse(ValueReg) - .constrainAllUses(TII, TRI, RBI); - return true; + BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(NewOpcode)) + .addDef(ResVReg) + .addUse(GR.getSPIRVTypeID(ResType)) + .addUse(Ptr) + .addUse(ScopeReg) + .addUse(MemSemReg) + .addUse(ValueReg) + .constrainAllUses(TII, TRI, RBI); + return true; } bool SPIRVInstructionSelector::selectUnmergeValues(MachineInstr &I) const { @@ -2551,12 +2554,12 @@ bool SPIRVInstructionSelector::selectIntegerDotExpansion( for (unsigned i = 1; i < GR.getScalarOrVectorComponentCount(VecType); i++) { Register Elt = MRI->createVirtualRegister(GR.getRegClass(ResType)); - BuildMI(BB, I, I.getDebugLoc(), TII.get(SPIRV::OpCompositeExtract)) - .addDef(Elt) - .addUse(GR.getSPIRVTypeID(ResType)) - .addUse(TmpVec) - .addImm(i) - .constrainAllUses(TII, TRI, RBI); + BuildMI(BB, I, I.getDebugLoc(), TII.get(SPIRV::OpCompositeExtract)) + .addDef(Elt) + .addUse(GR.getSPIRVTypeID(ResType)) + .addUse(TmpVec) + .addImm(i) + .constrainAllUses(TII, TRI, RBI); Register Sum = i < GR.getScalarOrVectorComponentCount(VecType) - 1 ? MRI->createVirtualRegister(GR.getRegClass(ResType)) @@ -2757,13 +2760,13 @@ bool SPIRVInstructionSelector::selectSign(Register ResVReg, ? MRI->createVirtualRegister(&SPIRV::IDRegClass) : ResVReg; - BuildMI(BB, I, DL, TII.get(SPIRV::OpExtInst)) - .addDef(SignReg) - .addUse(GR.getSPIRVTypeID(InputType)) - .addImm(static_cast<uint32_t>(SPIRV::InstructionSet::GLSL_std_450)) - .addImm(SignOpcode) - .addUse(InputRegister) - .constrainAllUses(TII, TRI, RBI); + BuildMI(BB, I, DL, TII.get(SPIRV::OpExtInst)) + .addDef(SignReg) + .addUse(GR.getSPIRVTypeID(InputType)) + .addImm(static_cast<uint32_t>(SPIRV::InstructionSet::GLSL_std_450)) + .addImm(SignOpcode) + .addUse(InputRegister) + .constrainAllUses(TII, TRI, RBI); if (NeedsConversion) { auto ConvertOpcode = IsFloatTy ? SPIRV::OpConvertFToS : SPIRV::OpSConvert; @@ -2798,6 +2801,23 @@ bool SPIRVInstructionSelector::selectWaveOpInst(Register ResVReg, return true; } +bool SPIRVInstructionSelector::selectBarrierInst(MachineInstr &I, + unsigned Scope, + bool WithGroupSync) const { + auto BarrierType = + WithGroupSync ? SPIRV::OpControlBarrier : SPIRV::OpMemoryBarrier; + Register MemSemReg = + buildI32Constant(SPIRV::MemorySemantics::SequentiallyConsistent, I); + Register ScopeReg = buildI32Constant(Scope, I); + MachineBasicBlock &BB = *I.getParent(); + BuildMI(BB, I, I.getDebugLoc(), TII.get(BarrierType)) + .addUse(ScopeReg) + .addUse(ScopeReg) + .addUse(MemSemReg) + .constrainAllUses(TII, TRI, RBI); + return true; +} + bool SPIRVInstructionSelector::selectWaveActiveCountBits( Register ResVReg, SPIRVTypeInst ResType, MachineInstr &I) const { @@ -3848,22 +3868,22 @@ bool SPIRVInstructionSelector::selectDerivativeInst( Register ConvertToVReg = MRI->createVirtualRegister(RegClass); Register DpdOpVReg = MRI->createVirtualRegister(RegClass); - BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(SPIRV::OpFConvert)) - .addDef(ConvertToVReg) - .addUse(GR.getSPIRVTypeID(F32ConvertTy)) - .addUse(SrcReg) - .constrainAllUses(TII, TRI, RBI); - BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(DPdOpCode)) - .addDef(DpdOpVReg) - .addUse(GR.getSPIRVTypeID(F32ConvertTy)) - .addUse(ConvertToVReg) - .constrainAllUses(TII, TRI, RBI); - BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(SPIRV::OpFConvert)) - .addDef(ResVReg) - .addUse(GR.getSPIRVTypeID(ResType)) - .addUse(DpdOpVReg) - .constrainAllUses(TII, TRI, RBI); - return true; + BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(SPIRV::OpFConvert)) + .addDef(ConvertToVReg) + .addUse(GR.getSPIRVTypeID(F32ConvertTy)) + .addUse(SrcReg) + .constrainAllUses(TII, TRI, RBI); + BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(DPdOpCode)) + .addDef(DpdOpVReg) + .addUse(GR.getSPIRVTypeID(F32ConvertTy)) + .addUse(ConvertToVReg) + .constrainAllUses(TII, TRI, RBI); + BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(SPIRV::OpFConvert)) + .addDef(ResVReg) + .addUse(GR.getSPIRVTypeID(ResType)) + .addUse(DpdOpVReg) + .constrainAllUses(TII, TRI, RBI); + return true; } bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg, @@ -4160,18 +4180,10 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg, return selectFirstBitHigh(ResVReg, ResType, I, /*IsSigned=*/true); case Intrinsic::spv_firstbitlow: // There is no CL equivlent of FindILsb return selectFirstBitLow(ResVReg, ResType, I); - case Intrinsic::spv_group_memory_barrier_with_group_sync: { - Register MemSemReg = - buildI32Constant(SPIRV::MemorySemantics::SequentiallyConsistent, I); - Register ScopeReg = buildI32Constant(SPIRV::Scope::Workgroup, I); - MachineBasicBlock &BB = *I.getParent(); - BuildMI(BB, I, I.getDebugLoc(), TII.get(SPIRV::OpControlBarrier)) - .addUse(ScopeReg) - .addUse(ScopeReg) - .addUse(MemSemReg) - .constrainAllUses(TII, TRI, RBI); - return true; - } + case Intrinsic::spv_group_memory_barrier: + return selectBarrierInst(I, SPIRV::Scope::Device, false); + case Intrinsic::spv_group_memory_barrier_with_group_sync: + return selectBarrierInst(I, SPIRV::Scope::Workgroup, true); case Intrinsic::spv_generic_cast_to_ptr_explicit: { Register PtrReg = I.getOperand(I.getNumExplicitDefs() + 1).getReg(); SPIRV::StorageClass::StorageClass ResSC = @@ -4333,8 +4345,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg, bool SPIRVInstructionSelector::selectHandleFromBinding(Register &ResVReg, SPIRVTypeInst ResType, MachineInstr &I) const { - // The images need to be loaded in the same basic block as their use. We defer - // loading the image to the intrinsic that uses it. + // The images need to be loaded in the same basic block as their use. We + // defer loading the image to the intrinsic that uses it. if (ResType->getOpcode() == SPIRV::OpTypeImage) return true; @@ -4381,9 +4393,9 @@ bool SPIRVInstructionSelector::selectUpdateCounter(Register &ResVReg, Register CounterHandleReg = Intr.getOperand(2).getReg(); Register IncrReg = Intr.getOperand(3).getReg(); - // The counter handle is a pointer to the counter variable (which is a struct - // containing an i32). We need to get a pointer to that i32 member to do the - // atomic operation. + // The counter handle is a pointer to the counter variable (which is a + // struct containing an i32). We need to get a pointer to that i32 member to + // do the atomic operation. #ifndef NDEBUG SPIRVTypeInst CounterVarType = GR.getSPIRVTypeForVReg(CounterHandleReg); SPIRVTypeInst CounterVarPointeeType = GR.getPointeeType(CounterVarType); @@ -4442,8 +4454,8 @@ bool SPIRVInstructionSelector::selectUpdateCounter(Register &ResVReg, } // In HLSL, IncrementCounter returns the value *before* the increment, while - // DecrementCounter returns the value *after* the decrement. Both are lowered - // to the same atomic intrinsic which returns the value *before* the + // DecrementCounter returns the value *after* the decrement. Both are + // lowered to the same atomic intrinsic which returns the value *before* the // operation. So for decrements (negative IncrVal), we must subtract the // increment value from the result to get the post-decrement value. BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(SPIRV::OpIAddS)) @@ -4462,8 +4474,8 @@ bool SPIRVInstructionSelector::selectReadImageIntrinsic(Register &ResVReg, // this will generate invalid code. A proper solution is to move // the OpLoad from selectHandleFromBinding here. However, to do // that we will need to change the return type of the intrinsic. - // We will do that when we can, but for now trying to move forward with other - // issues. + // We will do that when we can, but for now trying to move forward with + // other issues. Register ImageReg = I.getOperand(2).getReg(); auto *ImageDef = cast<GIntrinsic>(getVRegDef(*MRI, ImageReg)); Register NewImageReg = MRI->createVirtualRegister(MRI->getRegClass(ImageReg)); @@ -4682,8 +4694,8 @@ bool SPIRVInstructionSelector::selectGatherIntrinsic(Register &ResVReg, auto Dim = static_cast<SPIRV::Dim::Dim>(ImageType->getOperand(2).getImm()); if (Dim != SPIRV::Dim::DIM_2D && Dim != SPIRV::Dim::DIM_Cube && Dim != SPIRV::Dim::DIM_Rect) { - I.emitGenericError( - "Gather operations are only supported for 2D, Cube, and Rect images."); + I.emitGenericError("Gather operations are only supported for 2D, Cube, " + "and Rect images."); return false; } @@ -4806,10 +4818,10 @@ bool SPIRVInstructionSelector::selectResourceGetPointer(Register &ResVReg, Register ResourcePtr = I.getOperand(2).getReg(); SPIRVTypeInst RegType = GR.getSPIRVTypeForVReg(ResourcePtr, I.getMF()); if (RegType->getOpcode() == SPIRV::OpTypeImage) { - // For texel buffers, the index into the image is part of the OpImageRead or - // OpImageWrite instructions. So we will do nothing in this case. This - // intrinsic will be combined with the load or store when selecting the load - // or store. + // For texel buffers, the index into the image is part of the OpImageRead + // or OpImageWrite instructions. So we will do nothing in this case. This + // intrinsic will be combined with the load or store when selecting the + // load or store. return true; } @@ -4927,8 +4939,8 @@ bool SPIRVInstructionSelector::selectImageWriteIntrinsic( // this will generate invalid code. A proper solution is to move // the OpLoad from selectHandleFromBinding here. However, to do // that we will need to change the return type of the intrinsic. - // We will do that when we can, but for now trying to move forward with other - // issues. + // We will do that when we can, but for now trying to move forward with + // other issues. Register ImageReg = I.getOperand(1).getReg(); auto *ImageDef = cast<GIntrinsic>(getVRegDef(*MRI, ImageReg)); Register NewImageReg = MRI->createVirtualRegister(MRI->getRegClass(ImageReg)); @@ -5246,8 +5258,8 @@ bool SPIRVInstructionSelector::selectFirstBitHigh(Register ResVReg, return selectFirstBitSet64(ResVReg, ResType, I, OpReg, BitSetOpcode, /*SwapPrimarySide=*/false); default: - report_fatal_error( - "spv_firstbituhigh and spv_firstbitshigh only support 16,32,64 bits."); + report_fatal_error("spv_firstbituhigh and spv_firstbitshigh only support " + "16,32,64 bits."); } } @@ -5258,8 +5270,8 @@ bool SPIRVInstructionSelector::selectFirstBitLow(Register ResVReg, Register OpReg = I.getOperand(2).getReg(); SPIRVTypeInst OpType = GR.getSPIRVTypeForVReg(OpReg); // OpUConvert treats the operand bits as an unsigned i16 and zero extends it - // to an unsigned i32. As this leaves all the least significant bits unchanged - // so the first set bit from the LSB side doesn't change. + // to an unsigned i32. As this leaves all the least significant bits + // unchanged so the first set bit from the LSB side doesn't change. unsigned ExtendOpcode = SPIRV::OpUConvert; unsigned BitSetOpcode = GL::FindILsb; @@ -5340,11 +5352,11 @@ bool SPIRVInstructionSelector::selectBranchCond(MachineInstr &I) const { // G_BRCOND to create an OpBranchConditional. We should hit G_BR first, and // generate the OpBranchConditional in selectBranch above. // - // If an OpBranchConditional has been generated, we simply return, as the work - // is alread done. If there is no OpBranchConditional, LLVM must be relying on - // implicit fallthrough to the next basic block, so we need to create an - // OpBranchConditional with an explicit "false" argument pointing to the next - // basic block that LLVM would fall through to. + // If an OpBranchConditional has been generated, we simply return, as the + // work is alread done. If there is no OpBranchConditional, LLVM must be + // relying on implicit fallthrough to the next basic block, so we need to + // create an OpBranchConditional with an explicit "false" argument pointing + // to the next basic block that LLVM would fall through to. const MachineInstr *NextI = I.getNextNode(); // Check if this has already been successfully selected. if (NextI != nullptr && NextI->getOpcode() == SPIRV::OpBranchConditional) @@ -5458,8 +5470,8 @@ bool SPIRVInstructionSelector::selectGlobalValue( auto GlobalVar = cast<GlobalVariable>(GV); assert(GlobalVar->getName() != "llvm.global.annotations"); - // Skip empty declaration for GVs with initializers till we get the decl with - // passed initializer. + // Skip empty declaration for GVs with initializers till we get the decl + // with passed initializer. if (hasInitializer(GlobalVar) && !Init) return true; @@ -5475,9 +5487,9 @@ bool SPIRVInstructionSelector::selectGlobalValue( ResVReg, ResType, GlobalIdent, GV, StorageClass, Init, GlobalVar->isConstant(), LnkType, MIRBuilder, true); // TODO: For AMDGCN, we pipe externally_initialized through via - // HostAccessINTEL, with ReadWrite (3) access, which is we then handle during - // reverse translation. We should remove this once SPIR-V gains the ability to - // express the concept. + // HostAccessINTEL, with ReadWrite (3) access, which is we then handle + // during reverse translation. We should remove this once SPIR-V gains the + // ability to express the concept. if (GlobalVar->isExternallyInitialized() && STI.getTargetTriple().getVendor() == Triple::AMD) { constexpr unsigned ReadWriteINTEL = 3u; @@ -5746,8 +5758,8 @@ bool SPIRVInstructionSelector::loadHandleBeforePosition( buildPointerToResource(SPIRVTypeInst(VarType), SC, Set, Binding, ArraySize, IndexReg, Name, MIRBuilder); - // The handle for the buffer is the pointer to the resource. For an image, the - // handle is the image object. So images get an extra load. + // The handle for the buffer is the pointer to the resource. For an image, + // the handle is the image object. So images get an extra load. uint32_t LoadOpcode = IsStructuredBuffer ? SPIRV::OpCopyObject : SPIRV::OpLoad; GR.assignSPIRVTypeToVReg(ResType, HandleReg, *Pos.getMF()); diff --git a/llvm/test/CodeGen/DirectX/group_memory_barrier.ll b/llvm/test/CodeGen/DirectX/group_memory_barrier.ll new file mode 100644 index 0000000000000..2c865d9965b73 --- /dev/null +++ b/llvm/test/CodeGen/DirectX/group_memory_barrier.ll @@ -0,0 +1,8 @@ +; RUN: opt -S -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s + +define void @test_group_memory_barrier() { +entry: + ; CHECK: call void @dx.op.barrier(i32 80, i32 9) + call void @llvm.dx.group.memory.barrier() + ret void +} \ No newline at end of file diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll new file mode 100644 index 0000000000000..30697a2490c0f --- /dev/null +++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll @@ -0,0 +1,14 @@ +; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-vulkan %s -o - | FileCheck %s +; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-vulkan %s -o - -filetype=obj | spirv-val %} + +; CHECK: OpMemoryModel Logical GLSL450 + +define void @test_group_memory_barrier() { +entry: + ; CHECK: %[[#TY:]] = OpTypeInt 32 0 + ; CHECK-DAG: %[[#MEM_SEM:]] = OpConstant %[[#TY]] 16 + ; CHECK-DAG: %[[#EXEC_AND_MEM_SCOPE:]] = OpConstant %[[#TY]] 2 + ; CHECK: OpMemoryBarrier %[[#EXEC_AND_MEM_SCOPE]] %[[#EXEC_AND_MEM_SCOPE]] %[[#MEM_SEM]] + call void @llvm.spv.group.memory.barrier() + ret void +} >From 80c2bb4745bfb1b06dca9c03b05c0eb4c944c8dc Mon Sep 17 00:00:00 2001 From: KungFuDonkey <[email protected]> Date: Mon, 9 Mar 2026 10:41:04 +0100 Subject: [PATCH 2/5] Passing tests and reverted basic/builtins.tdd --- clang/include/clang/Basic/Builtins.td | 279 ++++++++---------- ...er.hlsl => GroupMemoryBarrier-errors.hlsl} | 0 .../Target/SPIRV/SPIRVInstructionSelector.cpp | 17 +- .../CodeGen/DirectX/group_memory_barrier.ll | 4 +- .../group_memory_barrier_with_group_sync.ll | 2 +- .../hlsl-intrinsics/group_memory_barrier.ll | 2 +- 6 files changed, 134 insertions(+), 170 deletions(-) rename clang/test/SemaHLSL/BuiltIns/{GroupMemoryBarrier.hlsl => GroupMemoryBarrier-errors.hlsl} (100%) diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td index c98236e4258d4..10b2e930792c7 100644 --- a/clang/include/clang/Basic/Builtins.td +++ b/clang/include/clang/Basic/Builtins.td @@ -8,64 +8,68 @@ include "clang/Basic/BuiltinsBase.td" -class FPMathTemplate - : Template<["float", "double", "long double"], ["f", "", "l"]>; +class FPMathTemplate : Template<["float", "double", "long double"], + ["f", "", "l"]>; -class FPMathWithF16Template - : Template<["float", "double", "long double", "__fp16"], ["f", "", "l", - "f16"]>; +class FPMathWithF16Template : + Template<["float", "double", "long double", "__fp16"], + ["f", "", "l", "f16"]>; -class FPMathWithF16F128Template - : Template<["float", "double", "long double", "__fp16", "__float128"], - ["f", "", "l", "f16", "f128"]>; +class FPMathWithF16F128Template : + Template<["float", "double", "long double", "__fp16", "__float128"], + ["f", "", "l", "f16", "f128"]>; -class FPMathWithF128Template - : Template<["float", "double", "long double", "__float128"], ["f", "", "l", - "f128"]>; +class FPMathWithF128Template : + Template<["float", "double", "long double", "__float128"], + ["f", "", "l", "f128"]>; -class F16F128MathTemplate : Template<["__fp16", "__float128"], ["f16", "f128"]>; +class F16F128MathTemplate : Template<["__fp16", "__float128"], + ["f16", "f128"]>; -class IntMathTemplate - : Template<["int", "long int", "long long int"], ["", "l", "ll"], - /*AsPrefix=*/1>; +class IntMathTemplate : Template<["int", "long int", "long long int"], + ["", "l", "ll"], /*AsPrefix=*/1>; -class MSInt8_16_32Template - : Template<["char", "short", "msint32_t"], ["8", "16", ""]>; +class MSInt8_16_32Template : Template<["char", "short", "msint32_t"], + ["8", "16", ""]>; class Int8_16_32_64Template - : Template<["char", "short", "int", "long long int"], ["8", "16", "32", - "64"]>; + : Template<["char", "short", "int", "long long int"], + ["8", "16", "32", "64"]>; class MSInt8_16_32_64Template - : Template<["char", "short", "msint32_t", "long long int"], ["8", "16", "", - "64"]>; + : Template<["char", "short", "msint32_t", "long long int"], + ["8", "16", "", "64"]>; -class MSInt16_32Template : Template<["short", "msint32_t"], ["16", ""]>; +class MSInt16_32Template : Template<["short", "msint32_t"], + ["16", ""]>; -class MSUInt16_32_64Template - : Template<["unsigned short", "unsigned int", "uint64_t"], ["16", "", - "64"]>; +class MSUInt16_32_64Template : + Template<["unsigned short", "unsigned int", "uint64_t"], + ["16", "", "64"]>; -class MSInt32_64Template : Template<["msint32_t", "int64_t"], ["", "64"]>; +class MSInt32_64Template : Template<["msint32_t", "int64_t"], + ["", "64"]>; -class FloatDoubleTemplate : Template<["float", "double"], ["f", ""]>; +class FloatDoubleTemplate : Template<["float", "double"], + ["f", ""]>; // FIXME: These assume that char -> i8, short -> i16, int -> i32, // long long -> i64. -class SyncBuiltinsTemplate - : Template<["char", "short", "int", "long long int", "__int128_t"], - ["1", "2", "4", "8", "16"]>; +class SyncBuiltinsTemplate : + Template<["char", "short", "int", "long long int", "__int128_t"], + ["1", "2", "4", "8", "16"]>; -class BitInt8_16_32_64BuiltinsTemplate - : Template<["unsigned char", "unsigned short", "uint32_t", "uint64_t"], - ["8", "16", "32", "64"]>; +class BitInt8_16_32_64BuiltinsTemplate : + Template<["unsigned char", "unsigned short", "uint32_t", "uint64_t"], + ["8", "16", "32", "64"]>; -class BitShort_Int_Long_LongLongTemplate - : Template<["short", "int", "long int", "long long int"], ["s", "", "l", - "ll"]>; +class BitShort_Int_Long_LongLongTemplate : + Template<["short", "int", "long int", "long long int"], + ["s", "", "l", "ll"]>; -class BitInt_Long_LongLongTemplate - : Template<["int", "long int", "long long int"], ["", "l", "ll"]>; +class BitInt_Long_LongLongTemplate : + Template<["int", "long int", "long long int"], + ["", "l", "ll"]>; // Most of the types used in the prototypes are types from C, C++ or ObjC. There // are a few builtin-specific types and qualifiers. @@ -135,64 +139,55 @@ def CeilF16F128 : Builtin, F16F128MathTemplate { def CosF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_cos"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def CoshF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_cosh"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def ErfF128 : Builtin { let Spellings = ["__builtin_erff128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def ErfcF128 : Builtin { let Spellings = ["__builtin_erfcf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def ExpF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_exp"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Exp2F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_exp2"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Exp10F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_exp10"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Expm1F128 : Builtin { let Spellings = ["__builtin_expm1f128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def FdimF128 : Builtin { let Spellings = ["__builtin_fdimf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } @@ -204,8 +199,7 @@ def FloorF16F128 : Builtin, F16F128MathTemplate { def FmaF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_fma"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T, T)"; } @@ -235,8 +229,7 @@ def FminimumNumF16F128 : Builtin, F16F128MathTemplate { def Atan2F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_atan2"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T)"; } @@ -266,8 +259,7 @@ def FabsF128 : Builtin { def FmodF16F128 : F16F128MathTemplate, Builtin { let Spellings = ["__builtin_fmod"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T)"; } @@ -284,7 +276,7 @@ def HugeVal : Builtin, FPMathWithF128Template { } def HugeValF16 : Builtin { - let Spellings = ["__builtin_huge_valf16"]; + let Spellings = ["__builtin_huge_valf16"]; let Attributes = [NoThrow, Const, Constexpr]; let Prototype = "_Float16()"; } @@ -303,8 +295,7 @@ def InfF16 : Builtin { def LdexpF16F128 : F16F128MathTemplate, Builtin { let Spellings = ["__builtin_ldexp"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, int)"; } @@ -328,10 +319,9 @@ def NanF128 : Builtin { let Prototype = "__float128(char const*)"; } -def Nans - : Builtin, - Template<["float", "double", "long double", "_Float16", "__float128"], - ["f", "", "l", "f16", "f128"]> { +def Nans : Builtin, + Template<["float", "double", "long double", "_Float16", "__float128"], + ["f", "", "l", "f16", "f128"]> { let Spellings = ["__builtin_nans"]; let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Pure, Constexpr]; let Prototype = "T(char const*)"; @@ -345,22 +335,19 @@ def PowI : Builtin, FPMathTemplate { def PowF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_pow"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T, T)"; } def HypotF128 : Builtin { let Spellings = ["__builtin_hypotf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } def ILogbF128 : Builtin { let Spellings = ["__builtin_ilogbf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "int(__float128)"; } @@ -372,64 +359,55 @@ def LgammaF128 : Builtin { def LLrintF128 : Builtin { let Spellings = ["__builtin_llrintf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "long long int(__float128)"; } def LLroundF128 : Builtin { let Spellings = ["__builtin_llroundf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "long long int(__float128)"; } def Log10F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_log10"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def Log1pF128 : Builtin { let Spellings = ["__builtin_log1pf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def Log2F16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_log2"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def LogbF128 : Builtin { let Spellings = ["__builtin_logbf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128)"; } def LogF16F128 : Builtin, F16F128MathTemplate { let Spellings = ["__builtin_log"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "T(T)"; } def LrintF128 : Builtin { let Spellings = ["__builtin_lrintf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "long int(__float128)"; } def LroundF128 : Builtin { let Spellings = ["__builtin_lroundf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "long int(__float128)"; } @@ -441,22 +419,19 @@ def NearbyintF128 : Builtin { def NextafterF128 : Builtin { let Spellings = ["__builtin_nextafterf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } def NexttowardF128 : Builtin { let Spellings = ["__builtin_nexttowardf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } def RemainderF128 : Builtin { let Spellings = ["__builtin_remainderf128"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, - ConstIgnoringErrnoAndExceptions]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions]; let Prototype = "__float128(__float128, __float128)"; } @@ -695,13 +670,15 @@ def Signbit : Builtin { def SignbitF : Builtin { let Spellings = ["__builtin_signbitf"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, Constexpr]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, + Constexpr]; let Prototype = "int(float)"; } def SignbitL : Builtin { let Spellings = ["__builtin_signbitl"]; - let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, Constexpr]; + let Attributes = [FunctionWithBuiltinPrefix, NoThrow, Const, + Constexpr]; let Prototype = "int(long double)"; } @@ -771,9 +748,8 @@ def Clrsb : Builtin, BitInt_Long_LongLongTemplate { // there exists native types on the target that are 32- and 64-bits wide, unless // these conditions are fulfilled these builtins will operate on a not intended // bitwidth. -def BSwap - : Builtin, - Template<["unsigned short", "uint32_t", "uint64_t"], ["16", "32", "64"]> { +def BSwap : Builtin, Template<["unsigned short", "uint32_t", "uint64_t"], + ["16", "32", "64"]> { let Spellings = ["__builtin_bswap"]; let Attributes = [NoThrow, Const, Constexpr]; let Prototype = "T(T)"; @@ -853,15 +829,13 @@ def BuiltinCalloc : Builtin { def BuiltinConstantP : Builtin { let Spellings = ["__builtin_constant_p"]; - let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, - Constexpr]; + let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, Constexpr]; let Prototype = "int(...)"; } def BuiltinClassifyType : Builtin { let Spellings = ["__builtin_classify_type"]; - let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, - Constexpr]; + let Attributes = [NoThrow, Const, CustomTypeChecking, UnevaluatedArguments, Constexpr]; let Prototype = "int(...)"; } @@ -1165,8 +1139,7 @@ def StpncpyChk : Builtin { def SNPrintfChk : Builtin { let Spellings = ["__builtin___snprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, PrintfFormat<4>]; - let Prototype = - "int(char* restrict, size_t, int, size_t, char const* restrict, ...)"; + let Prototype = "int(char* restrict, size_t, int, size_t, char const* restrict, ...)"; } def SPrintfChk : Builtin { @@ -1178,15 +1151,13 @@ def SPrintfChk : Builtin { def VSNPrintfChk : Builtin { let Spellings = ["__builtin___vsnprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, VPrintfFormat<4>]; - let Prototype = "int(char* restrict, size_t, int, size_t, char const* " - "restrict, __builtin_va_list)"; + let Prototype = "int(char* restrict, size_t, int, size_t, char const* restrict, __builtin_va_list)"; } def VSPrintfChk : Builtin { let Spellings = ["__builtin___vsprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, VPrintfFormat<3>]; - let Prototype = "int(char* restrict, int, size_t, char const* restrict, " - "__builtin_va_list)"; + let Prototype = "int(char* restrict, int, size_t, char const* restrict, __builtin_va_list)"; } def FPrintfChk : Builtin { @@ -1204,8 +1175,7 @@ def PrintfChk : Builtin { def VFPrintfChk : Builtin { let Spellings = ["__builtin___vfprintf_chk"]; let Attributes = [FunctionWithBuiltinPrefix, VPrintfFormat<2>]; - let Prototype = - "int(FILE* restrict, int, char const* restrict, __builtin_va_list)"; + let Prototype = "int(FILE* restrict, int, char const* restrict, __builtin_va_list)"; } def VPrintfChk : Builtin { @@ -2583,8 +2553,7 @@ def SyncFetchAndUMax : Builtin { let Prototype = "unsigned int(unsigned int volatile*, unsigned int)"; } -// ignored glibc builtin, see -// https://sourceware.org/bugzilla/show_bug.cgi?id=25399 +// ignored glibc builtin, see https://sourceware.org/bugzilla/show_bug.cgi?id=25399 def WarnMemsetZeroLen : Builtin { let Spellings = ["__warn_memset_zero_len"]; let Attributes = [NoThrow, Pure]; @@ -2634,10 +2603,9 @@ def BittestAndSet : MSLangBuiltin, MSInt32_64Template { let Prototype = "unsigned char(T*, T)"; } -def MSByteswap - : MSLibBuiltin<"stdlib.h">, - Template<["unsigned short", "msuint32_t", "unsigned long long int"], - ["_ushort", "_ulong", "_uint64"]> { +def MSByteswap : MSLibBuiltin<"stdlib.h">, + Template<["unsigned short", "msuint32_t", "unsigned long long int"], + ["_ushort", "_ulong", "_uint64"]> { let Spellings = ["_byteswap"]; let Attributes = [NoThrow, Const]; let Prototype = "T(T)"; @@ -3248,24 +3216,21 @@ def VPrintf : LibBuiltin<"stdio.h"> { def VfPrintf : LibBuiltin<"stdio.h"> { let Spellings = ["vfprintf"]; let Attributes = [NoThrow, VPrintfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = - "int(FILE* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = "int(FILE* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } def VsnPrintf : LibBuiltin<"stdio.h"> { let Spellings = ["vsnprintf"]; let Attributes = [NoThrow, VPrintfFormat<2>, NonNull<NonOptimizing, [2]>]; - let Prototype = - "int(char* restrict, size_t, char const* restrict, __builtin_va_list)"; + let Prototype = "int(char* restrict, size_t, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } def VsPrintf : LibBuiltin<"stdio.h"> { let Spellings = ["vsprintf"]; let Attributes = [NoThrow, VPrintfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = - "int(char* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = "int(char* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } @@ -3300,16 +3265,14 @@ def VScanf : LibBuiltin<"stdio.h"> { def VFScanf : LibBuiltin<"stdio.h"> { let Spellings = ["vfscanf"]; let Attributes = [VScanfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = - "int(FILE* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = "int(FILE* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } def VSScanf : LibBuiltin<"stdio.h"> { let Spellings = ["vsscanf"]; let Attributes = [VScanfFormat<1>, NonNull<NonOptimizing, [0, 1]>]; - let Prototype = - "int(char const* restrict, char const* restrict, __builtin_va_list)"; + let Prototype = "int(char const* restrict, char const* restrict, __builtin_va_list)"; let AddBuiltinPrefixedAlias = 1; } @@ -4529,9 +4492,11 @@ def AssumeSeparateStorage : Builtin { // Multiprecision Arithmetic Builtins. -class MPATemplate : Template<["unsigned char", "unsigned short", "unsigned int", - "unsigned long int", "unsigned long long int"], - ["b", "s", "", "l", "ll"]>; +class MPATemplate : Template< + ["unsigned char", "unsigned short", "unsigned int", + "unsigned long int", "unsigned long long int"], + ["b", "s", "", + "l", "ll"]>; def Addc : Builtin, MPATemplate { let Spellings = ["__builtin_addc"]; @@ -4566,9 +4531,9 @@ def MulOverflow : Builtin { let Prototype = "bool(...)"; } -class UOverflowTemplate - : Template<["unsigned int", "unsigned long int", "unsigned long long int"], - ["_overflow", "l_overflow", "ll_overflow"]>; +class UOverflowTemplate : + Template<["unsigned int", "unsigned long int", "unsigned long long int"], + ["_overflow", "l_overflow", "ll_overflow"]>; def UaddOverflow : Builtin, UOverflowTemplate { let Spellings = ["__builtin_uadd"]; @@ -4588,9 +4553,9 @@ def UmulOverflow : Builtin, UOverflowTemplate { let Prototype = "bool(T const, T const, T*)"; } -class SOverflowTemplate - : Template<["int", "long int", "long long int"], ["_overflow", "l_overflow", - "ll_overflow"]>; +class SOverflowTemplate : + Template<["int", "long int", "long long int"], + ["_overflow", "l_overflow", "ll_overflow"]>; def SaddOverflow : Builtin, SOverflowTemplate { let Spellings = ["__builtin_sadd"]; @@ -4850,8 +4815,7 @@ def PtrauthStringDiscriminator : Builtin { // AllocToken builtins. def InferAllocToken : Builtin { let Spellings = ["__builtin_infer_alloc_token"]; - let Attributes = [NoThrow, Const, Pure, CustomTypeChecking, Constexpr, - UnevaluatedArguments]; + let Attributes = [NoThrow, Const, Pure, CustomTypeChecking, Constexpr, UnevaluatedArguments]; let Prototype = "size_t(...)"; } @@ -4954,8 +4918,7 @@ def GetPipeMaxPackets : OCLPipeLangBuiltin { } // OpenCL v2.0 s6.13.17 - Enqueue kernel functions. -// Custom builtin check allows to perform special check of passed block -// arguments. +// Custom builtin check allows to perform special check of passed block arguments. def EnqueueKernel : OCL_DSELangBuiltin { let Spellings = ["enqueue_kernel"]; let Attributes = [CustomTypeChecking, NoThrow]; @@ -5043,7 +5006,7 @@ def GetDeviceSideMangledName : LangBuiltin<"CUDA_LANG"> { } // HLSL -def HLSLAddUint64 : LangBuiltin<"HLSL_LANG"> { +def HLSLAddUint64: LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_adduint64"]; let Attributes = [NoThrow, Const]; let Prototype = "void(...)"; @@ -5130,15 +5093,13 @@ def HLSLResourceUninitializedHandle : LangBuiltin<"HLSL_LANG"> { def HLSLResourceHandleFromBinding : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_resource_handlefrombinding"]; let Attributes = [NoThrow]; - let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, " - "int32_t, uint32_t, char const*)"; + let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, int32_t, uint32_t, char const*)"; } def HLSLResourceHandleFromImplicitBinding : LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_resource_handlefromimplicitbinding"]; let Attributes = [NoThrow]; - let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, " - "int32_t, uint32_t, char const*)"; + let Prototype = "__hlsl_resource_t(__hlsl_resource_t, uint32_t, uint32_t, int32_t, uint32_t, char const*)"; } def HLSLResourceCounterHandleFromImplicitBinding : LangBuiltin<"HLSL_LANG"> { @@ -5291,18 +5252,16 @@ def HLSLClamp : LangBuiltin<"HLSL_LANG"> { let Prototype = "void(...)"; } -def HLSLCrossFloat : LangBuiltin<"HLSL_LANG"> { +def HLSLCrossFloat: LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_crossf32"]; let Attributes = [NoThrow, Const]; - let Prototype = - "_ExtVector<3, float>(_ExtVector<3, float>, _ExtVector<3, float>)"; + let Prototype = "_ExtVector<3, float>(_ExtVector<3, float>, _ExtVector<3, float>)"; } -def HLSLCrossHalf : LangBuiltin<"HLSL_LANG"> { +def HLSLCrossHalf: LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_crossf16"]; let Attributes = [NoThrow, Const]; - let Prototype = - "_ExtVector<3, __fp16>(_ExtVector<3, __fp16>, _ExtVector<3, __fp16>)"; + let Prototype = "_ExtVector<3, __fp16>(_ExtVector<3, __fp16>, _ExtVector<3, __fp16>)"; } def HLSLDegrees : LangBuiltin<"HLSL_LANG"> { @@ -5413,7 +5372,7 @@ def HLSLSign : LangBuiltin<"HLSL_LANG"> { let Prototype = "void(...)"; } -def HLSLStep : LangBuiltin<"HLSL_LANG"> { +def HLSLStep: LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_step"]; let Attributes = [NoThrow, Const]; let Prototype = "void(...)"; @@ -5431,13 +5390,13 @@ def HLSLBufferUpdateCounter : LangBuiltin<"HLSL_LANG"> { let Prototype = "uint32_t(__hlsl_resource_t, int)"; } -def HLSLSplitDouble : LangBuiltin<"HLSL_LANG"> { +def HLSLSplitDouble: LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_elementwise_splitdouble"]; let Attributes = [NoThrow, Const]; let Prototype = "void(...)"; } -def HLSLClip : LangBuiltin<"HLSL_LANG"> { +def HLSLClip: LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_elementwise_clip"]; let Attributes = [NoThrow, Const, CustomTypeChecking]; let Prototype = "void(...)"; @@ -5449,7 +5408,7 @@ def HLSLGroupMemoryBarrier : LangBuiltin<"HLSL_LANG"> { let Prototype = "void()"; } -def HLSLGroupMemoryBarrierWithGroupSync : LangBuiltin<"HLSL_LANG"> { +def HLSLGroupMemoryBarrierWithGroupSync: LangBuiltin<"HLSL_LANG"> { let Spellings = ["__builtin_hlsl_group_memory_barrier_with_group_sync"]; let Attributes = [NoThrow, Const]; let Prototype = "void()"; diff --git a/clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier.hlsl b/clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier-errors.hlsl similarity index 100% rename from clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier.hlsl rename to clang/test/SemaHLSL/BuiltIns/GroupMemoryBarrier-errors.hlsl diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp index 704c0dd6dc5d7..b41e90480d3cb 100644 --- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp @@ -32,6 +32,7 @@ #include "llvm/IR/IntrinsicsSPIRV.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" +#include <fstream> #define DEBUG_TYPE "spirv-isel" @@ -2810,11 +2811,15 @@ bool SPIRVInstructionSelector::selectBarrierInst(MachineInstr &I, buildI32Constant(SPIRV::MemorySemantics::SequentiallyConsistent, I); Register ScopeReg = buildI32Constant(Scope, I); MachineBasicBlock &BB = *I.getParent(); - BuildMI(BB, I, I.getDebugLoc(), TII.get(BarrierType)) - .addUse(ScopeReg) - .addUse(ScopeReg) - .addUse(MemSemReg) - .constrainAllUses(TII, TRI, RBI); + auto MI = + BuildMI(BB, I, I.getDebugLoc(), TII.get(BarrierType)).addUse(ScopeReg); + + // OpControlBarrier needs to also set Execution Scope + if (WithGroupSync) { + MI.addUse(ScopeReg); + } + + MI.addUse(MemSemReg).constrainAllUses(TII, TRI, RBI); return true; } @@ -4181,7 +4186,7 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg, case Intrinsic::spv_firstbitlow: // There is no CL equivlent of FindILsb return selectFirstBitLow(ResVReg, ResType, I); case Intrinsic::spv_group_memory_barrier: - return selectBarrierInst(I, SPIRV::Scope::Device, false); + return selectBarrierInst(I, SPIRV::Scope::Workgroup, false); case Intrinsic::spv_group_memory_barrier_with_group_sync: return selectBarrierInst(I, SPIRV::Scope::Workgroup, true); case Intrinsic::spv_generic_cast_to_ptr_explicit: { diff --git a/llvm/test/CodeGen/DirectX/group_memory_barrier.ll b/llvm/test/CodeGen/DirectX/group_memory_barrier.ll index 2c865d9965b73..ce828168c0db1 100644 --- a/llvm/test/CodeGen/DirectX/group_memory_barrier.ll +++ b/llvm/test/CodeGen/DirectX/group_memory_barrier.ll @@ -2,7 +2,7 @@ define void @test_group_memory_barrier() { entry: - ; CHECK: call void @dx.op.barrier(i32 80, i32 9) + ; CHECK: call void @dx.op.barrier(i32 80, i32 8) call void @llvm.dx.group.memory.barrier() ret void -} \ No newline at end of file +} diff --git a/llvm/test/CodeGen/DirectX/group_memory_barrier_with_group_sync.ll b/llvm/test/CodeGen/DirectX/group_memory_barrier_with_group_sync.ll index baf93d4e177f0..ce3a20d331d0b 100644 --- a/llvm/test/CodeGen/DirectX/group_memory_barrier_with_group_sync.ll +++ b/llvm/test/CodeGen/DirectX/group_memory_barrier_with_group_sync.ll @@ -5,4 +5,4 @@ entry: ; CHECK: call void @dx.op.barrier(i32 80, i32 9) call void @llvm.dx.group.memory.barrier.with.group.sync() ret void -} \ No newline at end of file +} diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll index 30697a2490c0f..8884c738ed05c 100644 --- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll +++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/group_memory_barrier.ll @@ -8,7 +8,7 @@ entry: ; CHECK: %[[#TY:]] = OpTypeInt 32 0 ; CHECK-DAG: %[[#MEM_SEM:]] = OpConstant %[[#TY]] 16 ; CHECK-DAG: %[[#EXEC_AND_MEM_SCOPE:]] = OpConstant %[[#TY]] 2 - ; CHECK: OpMemoryBarrier %[[#EXEC_AND_MEM_SCOPE]] %[[#EXEC_AND_MEM_SCOPE]] %[[#MEM_SEM]] + ; CHECK: OpMemoryBarrier %[[#EXEC_AND_MEM_SCOPE]] %[[#MEM_SEM]] call void @llvm.spv.group.memory.barrier() ret void } >From 1f9409e3d539cf9219b3b2f816a675df6bc819d4 Mon Sep 17 00:00:00 2001 From: KungFuDonkey <[email protected]> Date: Mon, 9 Mar 2026 10:45:47 +0100 Subject: [PATCH 3/5] clang-format --- clang/lib/CodeGen/CGHLSLRuntime.h | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h index 11d99de157ba8..548ed776d12bd 100644 --- a/clang/lib/CodeGen/CGHLSLRuntime.h +++ b/clang/lib/CodeGen/CGHLSLRuntime.h @@ -184,8 +184,7 @@ class CGHLSLRuntime { GENERATE_HLSL_INTRINSIC_FUNCTION(NonUniformResourceIndex, resource_nonuniformindex) GENERATE_HLSL_INTRINSIC_FUNCTION(BufferUpdateCounter, resource_updatecounter) - GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrier, - group_memory_barrier) + GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrier, group_memory_barrier) GENERATE_HLSL_INTRINSIC_FUNCTION(GroupMemoryBarrierWithGroupSync, group_memory_barrier_with_group_sync) GENERATE_HLSL_INTRINSIC_FUNCTION(GetDimensionsX, resource_getdimensions_x) >From d36f250525ee9e547d4ae758e801f342c0bffe1c Mon Sep 17 00:00:00 2001 From: KungFuDonkey <[email protected]> Date: Mon, 9 Mar 2026 10:47:09 +0100 Subject: [PATCH 4/5] clang-format --- llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp | 1 - 1 file changed, 1 deletion(-) diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp index b41e90480d3cb..095cc675683c5 100644 --- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp @@ -32,7 +32,6 @@ #include "llvm/IR/IntrinsicsSPIRV.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" -#include <fstream> #define DEBUG_TYPE "spirv-isel" >From 0f3120cec0b541c95d0e11d93e4917184c25024f Mon Sep 17 00:00:00 2001 From: KungFuDonkey <[email protected]> Date: Mon, 9 Mar 2026 11:16:44 +0100 Subject: [PATCH 5/5] reformat SPIRVInstructionSelector.cpp --- .../Target/SPIRV/SPIRVInstructionSelector.cpp | 66 +++++++++---------- 1 file changed, 33 insertions(+), 33 deletions(-) diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp index 095cc675683c5..c34e45f8cdf57 100644 --- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp @@ -4349,8 +4349,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg, bool SPIRVInstructionSelector::selectHandleFromBinding(Register &ResVReg, SPIRVTypeInst ResType, MachineInstr &I) const { - // The images need to be loaded in the same basic block as their use. We - // defer loading the image to the intrinsic that uses it. + // The images need to be loaded in the same basic block as their use. We defer + // loading the image to the intrinsic that uses it. if (ResType->getOpcode() == SPIRV::OpTypeImage) return true; @@ -4397,9 +4397,9 @@ bool SPIRVInstructionSelector::selectUpdateCounter(Register &ResVReg, Register CounterHandleReg = Intr.getOperand(2).getReg(); Register IncrReg = Intr.getOperand(3).getReg(); - // The counter handle is a pointer to the counter variable (which is a - // struct containing an i32). We need to get a pointer to that i32 member to - // do the atomic operation. + // The counter handle is a pointer to the counter variable (which is a struct + // containing an i32). We need to get a pointer to that i32 member to do the + // atomic operation. #ifndef NDEBUG SPIRVTypeInst CounterVarType = GR.getSPIRVTypeForVReg(CounterHandleReg); SPIRVTypeInst CounterVarPointeeType = GR.getPointeeType(CounterVarType); @@ -4458,8 +4458,8 @@ bool SPIRVInstructionSelector::selectUpdateCounter(Register &ResVReg, } // In HLSL, IncrementCounter returns the value *before* the increment, while - // DecrementCounter returns the value *after* the decrement. Both are - // lowered to the same atomic intrinsic which returns the value *before* the + // DecrementCounter returns the value *after* the decrement. Both are lowered + // to the same atomic intrinsic which returns the value *before* the // operation. So for decrements (negative IncrVal), we must subtract the // increment value from the result to get the post-decrement value. BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(SPIRV::OpIAddS)) @@ -4478,8 +4478,8 @@ bool SPIRVInstructionSelector::selectReadImageIntrinsic(Register &ResVReg, // this will generate invalid code. A proper solution is to move // the OpLoad from selectHandleFromBinding here. However, to do // that we will need to change the return type of the intrinsic. - // We will do that when we can, but for now trying to move forward with - // other issues. + // We will do that when we can, but for now trying to move forward with other + // issues. Register ImageReg = I.getOperand(2).getReg(); auto *ImageDef = cast<GIntrinsic>(getVRegDef(*MRI, ImageReg)); Register NewImageReg = MRI->createVirtualRegister(MRI->getRegClass(ImageReg)); @@ -4698,8 +4698,8 @@ bool SPIRVInstructionSelector::selectGatherIntrinsic(Register &ResVReg, auto Dim = static_cast<SPIRV::Dim::Dim>(ImageType->getOperand(2).getImm()); if (Dim != SPIRV::Dim::DIM_2D && Dim != SPIRV::Dim::DIM_Cube && Dim != SPIRV::Dim::DIM_Rect) { - I.emitGenericError("Gather operations are only supported for 2D, Cube, " - "and Rect images."); + I.emitGenericError( + "Gather operations are only supported for 2D, Cube, and Rect images."); return false; } @@ -4822,10 +4822,10 @@ bool SPIRVInstructionSelector::selectResourceGetPointer(Register &ResVReg, Register ResourcePtr = I.getOperand(2).getReg(); SPIRVTypeInst RegType = GR.getSPIRVTypeForVReg(ResourcePtr, I.getMF()); if (RegType->getOpcode() == SPIRV::OpTypeImage) { - // For texel buffers, the index into the image is part of the OpImageRead - // or OpImageWrite instructions. So we will do nothing in this case. This - // intrinsic will be combined with the load or store when selecting the - // load or store. + // For texel buffers, the index into the image is part of the OpImageRead or + // OpImageWrite instructions. So we will do nothing in this case. This + // intrinsic will be combined with the load or store when selecting the load + // or store. return true; } @@ -4943,8 +4943,8 @@ bool SPIRVInstructionSelector::selectImageWriteIntrinsic( // this will generate invalid code. A proper solution is to move // the OpLoad from selectHandleFromBinding here. However, to do // that we will need to change the return type of the intrinsic. - // We will do that when we can, but for now trying to move forward with - // other issues. + // We will do that when we can, but for now trying to move forward with other + // issues. Register ImageReg = I.getOperand(1).getReg(); auto *ImageDef = cast<GIntrinsic>(getVRegDef(*MRI, ImageReg)); Register NewImageReg = MRI->createVirtualRegister(MRI->getRegClass(ImageReg)); @@ -5262,8 +5262,8 @@ bool SPIRVInstructionSelector::selectFirstBitHigh(Register ResVReg, return selectFirstBitSet64(ResVReg, ResType, I, OpReg, BitSetOpcode, /*SwapPrimarySide=*/false); default: - report_fatal_error("spv_firstbituhigh and spv_firstbitshigh only support " - "16,32,64 bits."); + report_fatal_error( + "spv_firstbituhigh and spv_firstbitshigh only support 16,32,64 bits."); } } @@ -5274,8 +5274,8 @@ bool SPIRVInstructionSelector::selectFirstBitLow(Register ResVReg, Register OpReg = I.getOperand(2).getReg(); SPIRVTypeInst OpType = GR.getSPIRVTypeForVReg(OpReg); // OpUConvert treats the operand bits as an unsigned i16 and zero extends it - // to an unsigned i32. As this leaves all the least significant bits - // unchanged so the first set bit from the LSB side doesn't change. + // to an unsigned i32. As this leaves all the least significant bits unchanged + // so the first set bit from the LSB side doesn't change. unsigned ExtendOpcode = SPIRV::OpUConvert; unsigned BitSetOpcode = GL::FindILsb; @@ -5356,11 +5356,11 @@ bool SPIRVInstructionSelector::selectBranchCond(MachineInstr &I) const { // G_BRCOND to create an OpBranchConditional. We should hit G_BR first, and // generate the OpBranchConditional in selectBranch above. // - // If an OpBranchConditional has been generated, we simply return, as the - // work is alread done. If there is no OpBranchConditional, LLVM must be - // relying on implicit fallthrough to the next basic block, so we need to - // create an OpBranchConditional with an explicit "false" argument pointing - // to the next basic block that LLVM would fall through to. + // If an OpBranchConditional has been generated, we simply return, as the work + // is alread done. If there is no OpBranchConditional, LLVM must be relying on + // implicit fallthrough to the next basic block, so we need to create an + // OpBranchConditional with an explicit "false" argument pointing to the next + // basic block that LLVM would fall through to. const MachineInstr *NextI = I.getNextNode(); // Check if this has already been successfully selected. if (NextI != nullptr && NextI->getOpcode() == SPIRV::OpBranchConditional) @@ -5474,8 +5474,8 @@ bool SPIRVInstructionSelector::selectGlobalValue( auto GlobalVar = cast<GlobalVariable>(GV); assert(GlobalVar->getName() != "llvm.global.annotations"); - // Skip empty declaration for GVs with initializers till we get the decl - // with passed initializer. + // Skip empty declaration for GVs with initializers till we get the decl with + // passed initializer. if (hasInitializer(GlobalVar) && !Init) return true; @@ -5491,9 +5491,9 @@ bool SPIRVInstructionSelector::selectGlobalValue( ResVReg, ResType, GlobalIdent, GV, StorageClass, Init, GlobalVar->isConstant(), LnkType, MIRBuilder, true); // TODO: For AMDGCN, we pipe externally_initialized through via - // HostAccessINTEL, with ReadWrite (3) access, which is we then handle - // during reverse translation. We should remove this once SPIR-V gains the - // ability to express the concept. + // HostAccessINTEL, with ReadWrite (3) access, which is we then handle during + // reverse translation. We should remove this once SPIR-V gains the ability to + // express the concept. if (GlobalVar->isExternallyInitialized() && STI.getTargetTriple().getVendor() == Triple::AMD) { constexpr unsigned ReadWriteINTEL = 3u; @@ -5762,8 +5762,8 @@ bool SPIRVInstructionSelector::loadHandleBeforePosition( buildPointerToResource(SPIRVTypeInst(VarType), SC, Set, Binding, ArraySize, IndexReg, Name, MIRBuilder); - // The handle for the buffer is the pointer to the resource. For an image, - // the handle is the image object. So images get an extra load. + // The handle for the buffer is the pointer to the resource. For an image, the + // handle is the image object. So images get an extra load. uint32_t LoadOpcode = IsStructuredBuffer ? SPIRV::OpCopyObject : SPIRV::OpLoad; GR.assignSPIRVTypeToVReg(ResType, HandleReg, *Pos.getMF()); _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
