Skip to content

Commit 93caee1

Browse files
mga-scmga-sc
authored
[RISCV][SLEEF]: Support SLEEF vector library for RISC-V target. (#114014)
SLEEF math vector library now supports RISC-V target. Commit: shibatch/sleef#477 This patch enables the use of auto-vectorization with subsequent replacement by the corresponding SLEEF function.
1 parent 9acd8e3 commit 93caee1

File tree

9 files changed

+3258
-22
lines changed

9 files changed

+3258
-22
lines changed

‎clang/lib/Driver/ToolChains/Clang.cpp

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -5807,7 +5807,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
58075807
<< Name << Triple.getArchName();
58085808
} else if (Name == "SLEEF" || Name == "ArmPL") {
58095809
if (Triple.getArch() != llvm::Triple::aarch64 &&
5810-
Triple.getArch() != llvm::Triple::aarch64_be)
5810+
Triple.getArch() != llvm::Triple::aarch64_be &&
5811+
Triple.getArch() != llvm::Triple::riscv64)
58115812
D.Diag(diag::err_drv_unsupported_opt_for_target)
58125813
<< Name << Triple.getArchName();
58135814
}

‎clang/test/Driver/fveclib.c

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,7 @@
44
// RUN: %clang -### -c -fveclib=MASSV %s 2>&1 | FileCheck --check-prefix=CHECK-MASSV %s
55
// RUN: %clang -### -c -fveclib=Darwin_libsystem_m %s 2>&1 | FileCheck --check-prefix=CHECK-DARWIN_LIBSYSTEM_M %s
66
// RUN: %clang -### -c --target=aarch64 -fveclib=SLEEF %s 2>&1 | FileCheck --check-prefix=CHECK-SLEEF %s
7+
// RUN: %clang -### -c --target=riscv64-unknown-linux-gnu -fveclib=SLEEF -march=rv64gcv %s 2>&1 | FileCheck -check-prefix CHECK-SLEEF-RISCV %s
78
// RUN: %clang -### -c --target=aarch64 -fveclib=ArmPL %s 2>&1 | FileCheck --check-prefix=CHECK-ARMPL %s
89
// RUN: not %clang -c -fveclib=something %s 2>&1 | FileCheck --check-prefix=CHECK-INVALID %s
910

@@ -13,6 +14,7 @@
1314
// CHECK-MASSV: "-fveclib=MASSV"
1415
// CHECK-DARWIN_LIBSYSTEM_M: "-fveclib=Darwin_libsystem_m"
1516
// CHECK-SLEEF: "-fveclib=SLEEF"
17+
// CHECK-SLEEF-RISCV: "-fveclib=SLEEF"
1618
// CHECK-ARMPL: "-fveclib=ArmPL"
1719

1820
// CHECK-INVALID: error: invalid value 'something' in '-fveclib=something'
@@ -47,6 +49,9 @@
4749
// RUN: %clang -### --target=aarch64-linux-gnu -fveclib=SLEEF -flto %s 2>&1 | FileCheck --check-prefix=CHECK-LTO-SLEEF %s
4850
// CHECK-LTO-SLEEF: "-plugin-opt=-vector-library=sleefgnuabi"
4951

52+
// RUN: %clang -### --target=riscv64-unknown-linux-gnu -fveclib=SLEEF -flto -march=rv64gcv %s 2>&1 | FileCheck -check-prefix CHECK-LTO-SLEEF-RISCV %s
53+
// CHECK-LTO-SLEEF-RISCV: "-plugin-opt=-vector-library=sleefgnuabi"
54+
5055
// RUN: %clang -### --target=aarch64-linux-gnu -fveclib=ArmPL -flto %s 2>&1 | FileCheck --check-prefix=CHECK-LTO-ARMPL %s
5156
// CHECK-LTO-ARMPL: "-plugin-opt=-vector-library=ArmPL"
5257

@@ -85,6 +90,11 @@
8590
// CHECK-REENABLE-ERRNO-SLEEF: "-fveclib=SLEEF"
8691
// CHECK-REENABLE-ERRNO-SLEEF-SAME: "-fmath-errno"
8792

93+
// RUN: %clang -### --target=riscv64-unknown-linux-gnu -fveclib=SLEEF -fmath-errno -march=rv64gcv %s 2>&1 | FileCheck --check-prefix=CHECK-REENABLE-ERRNO-SLEEF-RISCV %s
94+
// CHECK-REENABLE-ERRNO-SLEEF-RISCV: math errno enabled by '-fmath-errno' after it was implicitly disabled by '-fveclib=SLEEF', this may limit the utilization of the vector library [-Wmath-errno-enabled-with-veclib]
95+
// CHECK-REENABLE-ERRNO-SLEEF-RISCV: "-fveclib=SLEEF"
96+
// CHECK-REENABLE-ERRNO-SLEEF-RISCV-SAME: "-fmath-errno"
97+
8898
// RUN: %clang -### --target=aarch64-linux-gnu -fveclib=ArmPL -fno-fast-math %s 2>&1 | FileCheck --check-prefix=CHECK-REENABLE-ERRNO-NFM %s
8999
// CHECK-REENABLE-ERRNO-NFM: math errno enabled by '-fno-fast-math' after it was implicitly disabled by '-fveclib=ArmPL', this may limit the utilization of the vector library [-Wmath-errno-enabled-with-veclib]
90100
// CHECK-REENABLE-ERRNO-NFM: "-fveclib=ArmPL"

‎llvm/include/llvm/Analysis/VecFuncs.def

Lines changed: 169 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -896,6 +896,175 @@ TLI_DEFINE_VECFUNC("llvm.tanh.f32", "_ZGVsMxv_tanhf", SCALABLE(4), MASKED, "_ZGV
896896
TLI_DEFINE_VECFUNC("tgamma", "_ZGVsMxv_tgamma", SCALABLE(2), MASKED, "_ZGVsMxv")
897897
TLI_DEFINE_VECFUNC("tgammaf", "_ZGVsMxv_tgammaf", SCALABLE(4), MASKED, "_ZGVsMxv")
898898

899+
#elif defined(TLI_DEFINE_SLEEFGNUABI_SCALABLE_VECFUNCS_RISCV)
900+
901+
TLI_DEFINE_VECFUNC("acos", "Sleef_acosdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
902+
TLI_DEFINE_VECFUNC("llvm.acos.f64", "Sleef_acosdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
903+
TLI_DEFINE_VECFUNC("acosf", "Sleef_acosfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
904+
TLI_DEFINE_VECFUNC("llvm.acos.f32", "Sleef_acosfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
905+
906+
TLI_DEFINE_VECFUNC("acosh", "Sleef_acoshdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
907+
TLI_DEFINE_VECFUNC("acoshf", "Sleef_acoshfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
908+
909+
TLI_DEFINE_VECFUNC("asin", "Sleef_asindx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
910+
TLI_DEFINE_VECFUNC("llvm.asin.f64", "Sleef_asindx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
911+
TLI_DEFINE_VECFUNC("asinf", "Sleef_asinfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
912+
TLI_DEFINE_VECFUNC("llvm.asin.f32", "Sleef_asinfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
913+
914+
TLI_DEFINE_VECFUNC("asinh", "Sleef_asinhdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
915+
TLI_DEFINE_VECFUNC("asinhf", "Sleef_asinhfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
916+
917+
TLI_DEFINE_VECFUNC("atan", "Sleef_atandx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
918+
TLI_DEFINE_VECFUNC("llvm.atan.f64", "Sleef_atandx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
919+
TLI_DEFINE_VECFUNC("atanf", "Sleef_atanfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
920+
TLI_DEFINE_VECFUNC("llvm.atan.f32", "Sleef_atanfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
921+
922+
TLI_DEFINE_VECFUNC("atan2", "Sleef_atan2dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
923+
TLI_DEFINE_VECFUNC("llvm.atan2.f64", "Sleef_atan2dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
924+
TLI_DEFINE_VECFUNC("atan2f", "Sleef_atan2fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
925+
TLI_DEFINE_VECFUNC("llvm.atan2.f32", "Sleef_atan2fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
926+
927+
TLI_DEFINE_VECFUNC("atanh", "Sleef_atanhdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
928+
TLI_DEFINE_VECFUNC("atanhf", "Sleef_atanhfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
929+
930+
TLI_DEFINE_VECFUNC("cbrt", "Sleef_cbrtdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
931+
TLI_DEFINE_VECFUNC("cbrtf", "Sleef_cbrtfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
932+
933+
TLI_DEFINE_VECFUNC("copysign", "Sleef_copysigndx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
934+
TLI_DEFINE_VECFUNC("copysignf", "Sleef_copysignfx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
935+
936+
TLI_DEFINE_VECFUNC("cos", "Sleef_cosdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
937+
TLI_DEFINE_VECFUNC("llvm.cos.f64", "Sleef_cosdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
938+
TLI_DEFINE_VECFUNC("cosf", "Sleef_cosfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
939+
TLI_DEFINE_VECFUNC("llvm.cos.f32", "Sleef_cosfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
940+
941+
TLI_DEFINE_VECFUNC("cosh", "Sleef_coshdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
942+
TLI_DEFINE_VECFUNC("llvm.cosh.f64", "Sleef_coshdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
943+
TLI_DEFINE_VECFUNC("coshf", "Sleef_coshfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
944+
TLI_DEFINE_VECFUNC("llvm.cosh.f32", "Sleef_coshfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
945+
946+
TLI_DEFINE_VECFUNC("cospi", "Sleef_cospidx_u05rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
947+
TLI_DEFINE_VECFUNC("cospif", "Sleef_cospifx_u05rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
948+
949+
TLI_DEFINE_VECFUNC("erf", "Sleef_erfdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
950+
TLI_DEFINE_VECFUNC("erff", "Sleef_erffx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
951+
952+
TLI_DEFINE_VECFUNC("erfc", "Sleef_erfcdx_u15rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
953+
TLI_DEFINE_VECFUNC("erfcf", "Sleef_erfcfx_u15rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
954+
955+
TLI_DEFINE_VECFUNC("exp", "Sleef_expdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
956+
TLI_DEFINE_VECFUNC("llvm.exp.f64", "Sleef_expdx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
957+
TLI_DEFINE_VECFUNC("expf", "Sleef_expfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
958+
TLI_DEFINE_VECFUNC("llvm.exp.f32", "Sleef_expfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
959+
960+
TLI_DEFINE_VECFUNC("exp10", "Sleef_exp10dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
961+
TLI_DEFINE_VECFUNC("llvm.exp10.f64", "Sleef_exp10dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
962+
TLI_DEFINE_VECFUNC("exp10f", "Sleef_exp10fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
963+
TLI_DEFINE_VECFUNC("llvm.exp10.f32", "Sleef_exp10fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
964+
965+
TLI_DEFINE_VECFUNC("exp2", "Sleef_exp2dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
966+
TLI_DEFINE_VECFUNC("llvm.exp2.f64", "Sleef_exp2dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
967+
TLI_DEFINE_VECFUNC("exp2f", "Sleef_exp2fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
968+
TLI_DEFINE_VECFUNC("llvm.exp2.f32", "Sleef_exp2fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
969+
970+
TLI_DEFINE_VECFUNC("expm1", "Sleef_expm1dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
971+
TLI_DEFINE_VECFUNC("expm1f", "Sleef_expm1fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
972+
973+
TLI_DEFINE_VECFUNC("fdim", "Sleef_fdimdx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
974+
TLI_DEFINE_VECFUNC("fdimf", "Sleef_fdimfx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
975+
976+
TLI_DEFINE_VECFUNC("fma", "Sleef_fmadx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvvv")
977+
TLI_DEFINE_VECFUNC("llvm.fma.f64", "Sleef_fmadx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvvv")
978+
TLI_DEFINE_VECFUNC("fmaf", "Sleef_fmafx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvvv")
979+
TLI_DEFINE_VECFUNC("llvm.fma.f32", "Sleef_fmafx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvvv")
980+
981+
TLI_DEFINE_VECFUNC("fmax", "Sleef_fmaxdx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
982+
TLI_DEFINE_VECFUNC("fmaxf", "Sleef_fmaxfx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
983+
984+
TLI_DEFINE_VECFUNC("fmin", "Sleef_fmindx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
985+
TLI_DEFINE_VECFUNC("fminf", "Sleef_fminfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
986+
987+
TLI_DEFINE_VECFUNC("fmod", "Sleef_fmoddx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
988+
TLI_DEFINE_VECFUNC("fmodf", "Sleef_fmodfx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
989+
990+
TLI_DEFINE_VECFUNC("hypot", "Sleef_hypotdx_u05rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
991+
TLI_DEFINE_VECFUNC("hypotf", "Sleef_hypotfx_u05rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
992+
993+
TLI_DEFINE_VECFUNC("ilogb", "Sleef_ilogbdx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
994+
TLI_DEFINE_VECFUNC("ilogbf", "Sleef_ilogbfx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
995+
996+
TLI_DEFINE_VECFUNC("ldexp", "Sleef_ldexpdx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
997+
TLI_DEFINE_VECFUNC("ldexpf", "Sleef_ldexpfx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
998+
999+
TLI_DEFINE_VECFUNC("lgamma", "Sleef_lgammadx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1000+
TLI_DEFINE_VECFUNC("lgammaf", "Sleef_lgammafx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1001+
1002+
TLI_DEFINE_VECFUNC("log", "Sleef_logdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVsNxv")
1003+
TLI_DEFINE_VECFUNC("llvm.log.f64", "Sleef_logdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1004+
TLI_DEFINE_VECFUNC("logf", "Sleef_logfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1005+
TLI_DEFINE_VECFUNC("llvm.log.f32", "Sleef_logfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1006+
1007+
TLI_DEFINE_VECFUNC("log10", "Sleef_log10dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1008+
TLI_DEFINE_VECFUNC("llvm.log10.f64", "Sleef_log10dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1009+
TLI_DEFINE_VECFUNC("log10f", "Sleef_log10fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1010+
TLI_DEFINE_VECFUNC("llvm.log10.f32", "Sleef_log10fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1011+
1012+
TLI_DEFINE_VECFUNC("log1p", "Sleef_log1pdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1013+
TLI_DEFINE_VECFUNC("log1pf", "Sleef_log1pfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1014+
1015+
TLI_DEFINE_VECFUNC("log2", "Sleef_log2dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1016+
TLI_DEFINE_VECFUNC("llvm.log2.f64", "Sleef_log2dx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1017+
TLI_DEFINE_VECFUNC("log2f", "Sleef_log2fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1018+
TLI_DEFINE_VECFUNC("llvm.log2.f32", "Sleef_log2fx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1019+
1020+
TLI_DEFINE_VECFUNC("modf", "Sleef_modfdx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvl8")
1021+
TLI_DEFINE_VECFUNC("modff", "Sleef_modffx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvl4")
1022+
1023+
TLI_DEFINE_VECFUNC("nextafter", "Sleef_nextafterdx_rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
1024+
TLI_DEFINE_VECFUNC("nextafterf", "Sleef_nextafterfx_rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
1025+
1026+
TLI_DEFINE_VECFUNC("pow", "Sleef_powdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
1027+
TLI_DEFINE_VECFUNC("llvm.pow.f64", "Sleef_powdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvv")
1028+
TLI_DEFINE_VECFUNC("powf", "Sleef_powfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
1029+
TLI_DEFINE_VECFUNC("llvm.pow.f32", "Sleef_powfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvv")
1030+
1031+
TLI_DEFINE_VECFUNC("sin", "Sleef_sindx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1032+
TLI_DEFINE_VECFUNC("llvm.sin.f64", "Sleef_sindx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1033+
TLI_DEFINE_VECFUNC("sinf", "Sleef_sinfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1034+
TLI_DEFINE_VECFUNC("llvm.sin.f32", "Sleef_sinfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1035+
1036+
TLI_DEFINE_VECFUNC("sincos", "Sleef_sincosdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvl8l8")
1037+
TLI_DEFINE_VECFUNC("sincosf", "Sleef_sincosfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvl4l4")
1038+
1039+
TLI_DEFINE_VECFUNC("sincospi", "Sleef_sincospidx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxvl8l8")
1040+
TLI_DEFINE_VECFUNC("sincospif", "Sleef_sincospifx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxvl4l4")
1041+
1042+
TLI_DEFINE_VECFUNC("sinh", "Sleef_sinhdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1043+
TLI_DEFINE_VECFUNC("llvm.sinh.f64", "Sleef_sinhdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1044+
TLI_DEFINE_VECFUNC("sinhf", "Sleef_sinhfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1045+
TLI_DEFINE_VECFUNC("llvm.sinh.f32", "Sleef_sinhfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1046+
1047+
TLI_DEFINE_VECFUNC("sinpi", "Sleef_sinpidx_u05rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1048+
TLI_DEFINE_VECFUNC("sinpif", "Sleef_sinpifx_u05rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1049+
1050+
TLI_DEFINE_VECFUNC("sqrt", "Sleef_sqrtdx_u05rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1051+
TLI_DEFINE_VECFUNC("llvm.sqrt.f64", "Sleef_sqrtdx_u05rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1052+
TLI_DEFINE_VECFUNC("sqrtf", "Sleef_sqrtfx_u05rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1053+
TLI_DEFINE_VECFUNC("llvm.sqrt.f32", "Sleef_sqrtfx_u05rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1054+
1055+
TLI_DEFINE_VECFUNC("tan", "Sleef_tandx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1056+
TLI_DEFINE_VECFUNC("llvm.tan.f64", "Sleef_tandx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1057+
TLI_DEFINE_VECFUNC("tanf", "Sleef_tanfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1058+
TLI_DEFINE_VECFUNC("llvm.tan.f32", "Sleef_tanfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1059+
1060+
TLI_DEFINE_VECFUNC("tanh", "Sleef_tanhdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1061+
TLI_DEFINE_VECFUNC("llvm.tanh.f64", "Sleef_tanhdx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1062+
TLI_DEFINE_VECFUNC("tanhf", "Sleef_tanhfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1063+
TLI_DEFINE_VECFUNC("llvm.tanh.f32", "Sleef_tanhfx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1064+
1065+
TLI_DEFINE_VECFUNC("tgamma", "Sleef_tgammadx_u10rvvm2", SCALABLE(2), NOMASK, "_ZGVrNxv")
1066+
TLI_DEFINE_VECFUNC("tgammaf", "Sleef_tgammafx_u10rvvm2", SCALABLE(4), NOMASK, "_ZGVrNxv")
1067+
8991068
#elif defined(TLI_DEFINE_ARMPL_VECFUNCS)
9001069

9011070
TLI_DEFINE_VECFUNC("acos", "armpl_vacosq_f64", FIXED(2), NOMASK, "_ZGV_LLVM_N2v")

‎llvm/include/llvm/IR/VFABIDemangler.h

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -44,6 +44,7 @@ enum class VFParamKind {
4444
enum class VFISAKind {
4545
AdvancedSIMD, // AArch64 Advanced SIMD (NEON)
4646
SVE, // AArch64 Scalable Vector Extension
47+
RVV, // RISC-V Vector Extension
4748
SSE, // x86 SSE
4849
AVX, // x86 AVX
4950
AVX2, // x86 AVX2

‎llvm/lib/Analysis/TargetLibraryInfo.cpp

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1325,6 +1325,14 @@ static const VecDesc VecFuncs_SLEEFGNUABI_VFScalable[] = {
13251325
#undef TLI_DEFINE_SLEEFGNUABI_SCALABLE_VECFUNCS
13261326
};
13271327

1328+
static const VecDesc VecFuncs_SLEEFGNUABI_VFScalableRISCV[] = {
1329+
#define TLI_DEFINE_SLEEFGNUABI_SCALABLE_VECFUNCS_RISCV
1330+
#define TLI_DEFINE_VECFUNC(SCAL, VEC, VF, MASK, VABI_PREFIX) \
1331+
{SCAL, VEC, VF, MASK, VABI_PREFIX},
1332+
#include "llvm/Analysis/VecFuncs.def"
1333+
#undef TLI_DEFINE_SLEEFGNUABI_SCALABLE_VECFUNCS_RISCV
1334+
};
1335+
13281336
static const VecDesc VecFuncs_ArmPL[] = {
13291337
#define TLI_DEFINE_ARMPL_VECFUNCS
13301338
#define TLI_DEFINE_VECFUNC(SCAL, VEC, VF, MASK, VABI_PREFIX) \
@@ -1374,6 +1382,9 @@ void TargetLibraryInfoImpl::addVectorizableFunctionsFromVecLib(
13741382
addVectorizableFunctions(VecFuncs_SLEEFGNUABI_VF4);
13751383
addVectorizableFunctions(VecFuncs_SLEEFGNUABI_VFScalable);
13761384
break;
1385+
case llvm::Triple::riscv64:
1386+
addVectorizableFunctions(VecFuncs_SLEEFGNUABI_VFScalableRISCV);
1387+
break;
13771388
}
13781389
break;
13791390
}

‎llvm/lib/IR/VFABIDemangler.cpp

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -42,6 +42,7 @@ static ParseRet tryParseISA(StringRef &MangledName, VFISAKind &ISA) {
4242
ISA = StringSwitch<VFISAKind>(MangledName.take_front(1))
4343
.Case("n", VFISAKind::AdvancedSIMD)
4444
.Case("s", VFISAKind::SVE)
45+
.Case("r", VFISAKind::RVV)
4546
.Case("b", VFISAKind::SSE)
4647
.Case("c", VFISAKind::AVX)
4748
.Case("d", VFISAKind::AVX2)
@@ -79,9 +80,9 @@ static ParseRet tryParseVLEN(StringRef &ParseString, VFISAKind ISA,
7980
std::pair<unsigned, bool> &ParsedVF) {
8081
if (ParseString.consume_front("x")) {
8182
// SVE is the only scalable ISA currently supported.
82-
if (ISA != VFISAKind::SVE) {
83+
if (ISA != VFISAKind::SVE && ISA != VFISAKind::RVV) {
8384
LLVM_DEBUG(dbgs() << "Vector function variant declared with scalable VF "
84-
<< "but ISA is not SVE\n");
85+
<< "but ISA supported for SVE and RVV only\n");
8586
return ParseRet::Error;
8687
}
8788
// We can't determine the VF of a scalable vector by looking at the vlen
@@ -301,9 +302,8 @@ static ParseRet tryParseAlign(StringRef &ParseString, Align &Alignment) {
301302
// the number of elements of the given type which would fit in such a vector.
302303
static std::optional<ElementCount> getElementCountForTy(const VFISAKind ISA,
303304
const Type *Ty) {
304-
// Only AArch64 SVE is supported at present.
305-
assert(ISA == VFISAKind::SVE &&
306-
"Scalable VF decoding only implemented for SVE\n");
305+
assert((ISA == VFISAKind::SVE || ISA == VFISAKind::RVV) &&
306+
"Scalable VF decoding only implemented for SVE and RVV\n");
307307

308308
if (Ty->isIntegerTy(64) || Ty->isDoubleTy() || Ty->isPointerTy())
309309
return ElementCount::getScalable(2);

0 commit comments

Comments
 (0)