Skip to content

Commit 3264a50

Browse files
Wolfram70Wolfram70
authored
[clang][NVPTX] Add builtins and intrinsics for conversions of new FP types (#134345)
This change: - Adds NVVM intrinsics and clang builtins for the cvt instruction variants of types (FP6) `.e2m3x2`, `.e3m2x2`, and (FP8) `.ue8m0x2` introduced in PTX 8.6 for `sm_100a`, `sm_101a`, and `sm_120a`. - Adds clang builtins for cvt instruction variant of type tf32. Tests are added in `NVPTX/convert-sm100a.ll` and `clang/test/CodeGen/builtins-nvptx.c` and verified through ptxas 12.8.0. PTX Spec Reference: https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt
1 parent 04b87e1 commit 3264a50

File tree

6 files changed

+589
-4
lines changed

6 files changed

+589
-4
lines changed

‎clang/include/clang/Basic/BuiltinsNVPTX.td‎

Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -580,6 +580,15 @@ def __nvvm_f2bf16_rz : NVPTXBuiltinSMAndPTX<"__bf16(float)", SM_80, PTX70>;
580580
def __nvvm_f2bf16_rz_relu : NVPTXBuiltinSMAndPTX<"__bf16(float)", SM_80, PTX70>;
581581

582582
def __nvvm_f2tf32_rna : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_80, PTX70>;
583+
def __nvvm_f2tf32_rna_satfinite : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_89, PTX81>;
584+
def __nvvm_f2tf32_rn : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_90, PTX78>;
585+
def __nvvm_f2tf32_rn_relu : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_90, PTX78>;
586+
def __nvvm_f2tf32_rn_satfinite : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_100, PTX86>;
587+
def __nvvm_f2tf32_rn_relu_satfinite : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_100, PTX86>;
588+
def __nvvm_f2tf32_rz : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_90, PTX78>;
589+
def __nvvm_f2tf32_rz_relu : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_90, PTX78>;
590+
def __nvvm_f2tf32_rz_satfinite : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_100, PTX86>;
591+
def __nvvm_f2tf32_rz_relu_satfinite : NVPTXBuiltinSMAndPTX<"int32_t(float)", SM_100, PTX86>;
583592

584593
def __nvvm_ff_to_e4m3x2_rn : NVPTXBuiltinSMAndPTX<"short(float, float)", SM_89, PTX81>;
585594
def __nvvm_ff_to_e4m3x2_rn_relu : NVPTXBuiltinSMAndPTX<"short(float, float)", SM_89, PTX81>;
@@ -596,6 +605,28 @@ def __nvvm_e4m3x2_to_f16x2_rn_relu : NVPTXBuiltinSMAndPTX<"_Vector<2, __fp16>(sh
596605
def __nvvm_e5m2x2_to_f16x2_rn : NVPTXBuiltinSMAndPTX<"_Vector<2, __fp16>(short)", SM_89, PTX81>;
597606
def __nvvm_e5m2x2_to_f16x2_rn_relu : NVPTXBuiltinSMAndPTX<"_Vector<2, __fp16>(short)", SM_89, PTX81>;
598607

608+
def __nvvm_ff_to_e2m3x2_rn_satfinite : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
609+
def __nvvm_ff_to_e2m3x2_rn_relu_satfinite : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
610+
def __nvvm_ff_to_e3m2x2_rn_satfinite : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
611+
def __nvvm_ff_to_e3m2x2_rn_relu_satfinite : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
612+
613+
def __nvvm_e2m3x2_to_f16x2_rn : NVPTXBuiltinSMAndPTX<"_Vector<2, __fp16>(short)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
614+
def __nvvm_e2m3x2_to_f16x2_rn_relu : NVPTXBuiltinSMAndPTX<"_Vector<2, __fp16>(short)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
615+
def __nvvm_e3m2x2_to_f16x2_rn : NVPTXBuiltinSMAndPTX<"_Vector<2, __fp16>(short)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
616+
def __nvvm_e3m2x2_to_f16x2_rn_relu : NVPTXBuiltinSMAndPTX<"_Vector<2, __fp16>(short)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
617+
618+
def __nvvm_ff_to_ue8m0x2_rz : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
619+
def __nvvm_ff_to_ue8m0x2_rz_satfinite : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
620+
def __nvvm_ff_to_ue8m0x2_rp : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
621+
def __nvvm_ff_to_ue8m0x2_rp_satfinite : NVPTXBuiltinSMAndPTX<"short(float, float)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
622+
623+
def __nvvm_bf16x2_to_ue8m0x2_rz : NVPTXBuiltinSMAndPTX<"short(_Vector<2, __bf16>)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
624+
def __nvvm_bf16x2_to_ue8m0x2_rz_satfinite : NVPTXBuiltinSMAndPTX<"short(_Vector<2, __bf16>)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
625+
def __nvvm_bf16x2_to_ue8m0x2_rp : NVPTXBuiltinSMAndPTX<"short(_Vector<2, __bf16>)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
626+
def __nvvm_bf16x2_to_ue8m0x2_rp_satfinite : NVPTXBuiltinSMAndPTX<"short(_Vector<2, __bf16>)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
627+
628+
def __nvvm_ue8m0x2_to_bf16x2 : NVPTXBuiltinSMAndPTX<"_Vector<2, __bf16>(short)", SM<"100a", [SM_101a, SM_120a]>, PTX86>;
629+
599630
// FNS
600631
let Attributes = [NoThrow] in {
601632
def __nvvm_fns : NVPTXBuiltinPTX<"unsigned int(unsigned int, unsigned int, int)", PTX60>;

‎clang/test/CodeGen/builtins-nvptx.c‎

Lines changed: 146 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -25,14 +25,29 @@
2525
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_86 -target-feature +ptx72 \
2626
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
2727
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX72_SM86 -check-prefix=LP64 %s
28-
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_89 -target-feature +ptx81 \
28+
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_89 -target-feature +ptx81 -DPTX=81\
2929
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
3030
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX81_SM89 %s
31+
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_90 -target-feature +ptx78 -DPTX=78 \
32+
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
33+
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX78_SM90 %s
34+
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_100 -target-feature +ptx86 -DPTX=86 \
35+
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
36+
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX86_SM100 %s
37+
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_100a -target-feature +ptx86 -DPTX=86 \
38+
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
39+
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX86_SM100a %s
40+
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_101a -target-feature +ptx86 -DPTX=86 \
41+
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
42+
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX86_SM101a %s
43+
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_120a -target-feature +ptx86 -DPTX=86 \
44+
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
45+
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX86_SM120a %s
3146
// ### The last run to check with the highest SM and PTX version available
3247
// ### to make sure target builtins are still accepted.
33-
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_100a -target-feature +ptx87 \
48+
// RUN: %clang_cc1 -ffp-contract=off -triple nvptx64-unknown-unknown -target-cpu sm_120a -target-feature +ptx87 -DPTX=87 \
3449
// RUN: -disable-llvm-optzns -fcuda-is-device -emit-llvm -o - -x cuda %s \
35-
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX81_SM89 %s
50+
// RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK_PTX86_SM120a %s
3651

3752
#define __device__ __attribute__((device))
3853
#define __global__ __attribute__((global))
@@ -995,7 +1010,7 @@ __device__ void nvvm_cvt_sm80() {
9951010

9961011
// CHECK-LABEL: nvvm_cvt_sm89
9971012
__device__ void nvvm_cvt_sm89() {
998-
#if __CUDA_ARCH__ >= 890
1013+
#if (PTX >= 81) && (__CUDA_ARCH__ >= 890)
9991014
// CHECK_PTX81_SM89: call i16 @llvm.nvvm.ff.to.e4m3x2.rn(float 1.000000e+00, float 1.000000e+00)
10001015
__nvvm_ff_to_e4m3x2_rn(1.0f, 1.0f);
10011016
// CHECK_PTX81_SM89: call i16 @llvm.nvvm.ff.to.e4m3x2.rn.relu(float 1.000000e+00, float 1.000000e+00)
@@ -1022,6 +1037,133 @@ __device__ void nvvm_cvt_sm89() {
10221037
__nvvm_e5m2x2_to_f16x2_rn(0x4c4c);
10231038
// CHECK_PTX81_SM89: call <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn.relu(i16 19532)
10241039
__nvvm_e5m2x2_to_f16x2_rn_relu(0x4c4c);
1040+
1041+
// CHECK_PTX81_SM89: call i32 @llvm.nvvm.f2tf32.rna.satfinite(float 1.000000e+00)
1042+
__nvvm_f2tf32_rna_satfinite(1.0f);
1043+
#endif
1044+
// CHECK: ret void
1045+
}
1046+
1047+
// CHECK-LABEL: nvvm_cvt_sm90
1048+
__device__ void nvvm_cvt_sm90() {
1049+
#if (PTX >= 78) && (__CUDA_ARCH__ >= 900)
1050+
// CHECK_PTX78_SM90: call i32 @llvm.nvvm.f2tf32.rn(float 1.000000e+00)
1051+
__nvvm_f2tf32_rn(1.0f);
1052+
// CHECK_PTX78_SM90: call i32 @llvm.nvvm.f2tf32.rn.relu(float 1.000000e+00)
1053+
__nvvm_f2tf32_rn_relu(1.0f);
1054+
// CHECK_PTX78_SM90: call i32 @llvm.nvvm.f2tf32.rz(float 1.000000e+00)
1055+
__nvvm_f2tf32_rz(1.0f);
1056+
// CHECK_PTX78_SM90: call i32 @llvm.nvvm.f2tf32.rz.relu(float 1.000000e+00)
1057+
__nvvm_f2tf32_rz_relu(1.0f);
1058+
#endif
1059+
// CHECK: ret void
1060+
}
1061+
1062+
// CHECK-LABEL: nvvm_cvt_sm100
1063+
__device__ void nvvm_cvt_sm100() {
1064+
#if (PTX >= 86) && (__CUDA_ARCH__ >= 1000)
1065+
// CHECK_PTX86_SM100: call i32 @llvm.nvvm.f2tf32.rn.satfinite(float 1.000000e+00)
1066+
__nvvm_f2tf32_rn_satfinite(1.0f);
1067+
// CHECK_PTX86_SM100: call i32 @llvm.nvvm.f2tf32.rn.relu.satfinite(float 1.000000e+00)
1068+
__nvvm_f2tf32_rn_relu_satfinite(1.0f);
1069+
// CHECK_PTX86_SM100: call i32 @llvm.nvvm.f2tf32.rz.satfinite(float 1.000000e+00)
1070+
__nvvm_f2tf32_rz_satfinite(1.0f);
1071+
// CHECK_PTX86_SM100: call i32 @llvm.nvvm.f2tf32.rz.relu.satfinite(float 1.000000e+00)
1072+
__nvvm_f2tf32_rz_relu_satfinite(1.0f);
1073+
#endif
1074+
// CHECK: ret void
1075+
}
1076+
1077+
// CHECK-LABEL: nvvm_cvt_sm100a_sm101a_sm120a
1078+
__device__ void nvvm_cvt_sm100a_sm101a_sm120a() {
1079+
#if (PTX >= 86) && \
1080+
(__CUDA_ARCH_FEAT_SM100_ALL || __CUDA_ARCH_FEAT_SM101_ALL || \
1081+
__CUDA_ARCH_FEAT_SM120_ALL)
1082+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.e2m3x2.rn.satfinite(float 1.000000e+00, float 1.000000e+00)
1083+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.e2m3x2.rn.satfinite(float 1.000000e+00, float 1.000000e+00)
1084+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.e2m3x2.rn.satfinite(float 1.000000e+00, float 1.000000e+00)
1085+
__nvvm_ff_to_e2m3x2_rn_satfinite(1.0f, 1.0f);
1086+
1087+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.e2m3x2.rn.relu.satfinite(float 1.000000e+00, float 1.000000e+00)
1088+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.e2m3x2.rn.relu.satfinite(float 1.000000e+00, float 1.000000e+00)
1089+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.e2m3x2.rn.relu.satfinite(float 1.000000e+00, float 1.000000e+00)
1090+
__nvvm_ff_to_e2m3x2_rn_relu_satfinite(1.0f, 1.0f);
1091+
1092+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.e3m2x2.rn.satfinite(float 1.000000e+00, float 1.000000e+00)
1093+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.e3m2x2.rn.satfinite(float 1.000000e+00, float 1.000000e+00)
1094+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.e3m2x2.rn.satfinite(float 1.000000e+00, float 1.000000e+00)
1095+
__nvvm_ff_to_e3m2x2_rn_satfinite(1.0f, 1.0f);
1096+
1097+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.e3m2x2.rn.relu.satfinite(float 1.000000e+00, float 1.000000e+00)
1098+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.e3m2x2.rn.relu.satfinite(float 1.000000e+00, float 1.000000e+00)
1099+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.e3m2x2.rn.relu.satfinite(float 1.000000e+00, float 1.000000e+00)
1100+
__nvvm_ff_to_e3m2x2_rn_relu_satfinite(1.0f, 1.0f);
1101+
1102+
// CHECK_PTX86_SM100a: call <2 x half> @llvm.nvvm.e2m3x2.to.f16x2.rn(i16 19532)
1103+
// CHECK_PTX86_SM101a: call <2 x half> @llvm.nvvm.e2m3x2.to.f16x2.rn(i16 19532)
1104+
// CHECK_PTX86_SM120a: call <2 x half> @llvm.nvvm.e2m3x2.to.f16x2.rn(i16 19532)
1105+
__nvvm_e2m3x2_to_f16x2_rn(0x4C4C);
1106+
1107+
// CHECK_PTX86_SM100a: call <2 x half> @llvm.nvvm.e2m3x2.to.f16x2.rn.relu(i16 18504)
1108+
// CHECK_PTX86_SM101a: call <2 x half> @llvm.nvvm.e2m3x2.to.f16x2.rn.relu(i16 18504)
1109+
// CHECK_PTX86_SM120a: call <2 x half> @llvm.nvvm.e2m3x2.to.f16x2.rn.relu(i16 18504)
1110+
__nvvm_e2m3x2_to_f16x2_rn_relu(0x4848);
1111+
1112+
// CHECK_PTX86_SM100a: call <2 x half> @llvm.nvvm.e3m2x2.to.f16x2.rn(i16 18504)
1113+
// CHECK_PTX86_SM101a: call <2 x half> @llvm.nvvm.e3m2x2.to.f16x2.rn(i16 18504)
1114+
// CHECK_PTX86_SM120a: call <2 x half> @llvm.nvvm.e3m2x2.to.f16x2.rn(i16 18504)
1115+
__nvvm_e3m2x2_to_f16x2_rn(0x4848);
1116+
1117+
// CHECK_PTX86_SM100a: call <2 x half> @llvm.nvvm.e3m2x2.to.f16x2.rn.relu(i16 19532)
1118+
// CHECK_PTX86_SM101a: call <2 x half> @llvm.nvvm.e3m2x2.to.f16x2.rn.relu(i16 19532)
1119+
// CHECK_PTX86_SM120a: call <2 x half> @llvm.nvvm.e3m2x2.to.f16x2.rn.relu(i16 19532)
1120+
__nvvm_e3m2x2_to_f16x2_rn_relu(0x4C4C);
1121+
1122+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rz(float 1.000000e+00, float 1.000000e+00)
1123+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rz(float 1.000000e+00, float 1.000000e+00)
1124+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rz(float 1.000000e+00, float 1.000000e+00)
1125+
__nvvm_ff_to_ue8m0x2_rz(1.0f, 1.0f);
1126+
1127+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rz.satfinite(float 1.000000e+00, float 1.000000e+00)
1128+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rz.satfinite(float 1.000000e+00, float 1.000000e+00)
1129+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rz.satfinite(float 1.000000e+00, float 1.000000e+00)
1130+
__nvvm_ff_to_ue8m0x2_rz_satfinite(1.0f, 1.0f);
1131+
1132+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rp(float 1.000000e+00, float 1.000000e+00)
1133+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rp(float 1.000000e+00, float 1.000000e+00)
1134+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rp(float 1.000000e+00, float 1.000000e+00)
1135+
__nvvm_ff_to_ue8m0x2_rp(1.0f, 1.0f);
1136+
1137+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rp.satfinite(float 1.000000e+00, float 1.000000e+00)
1138+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rp.satfinite(float 1.000000e+00, float 1.000000e+00)
1139+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.ff.to.ue8m0x2.rp.satfinite(float 1.000000e+00, float 1.000000e+00)
1140+
__nvvm_ff_to_ue8m0x2_rp_satfinite(1.0f, 1.0f);
1141+
1142+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz(<2 x bfloat> splat (bfloat 0xR3DCD)
1143+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz(<2 x bfloat> splat (bfloat 0xR3DCD)
1144+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz(<2 x bfloat> splat (bfloat 0xR3DCD)
1145+
__nvvm_bf16x2_to_ue8m0x2_rz({(__bf16)0.1f, (__bf16)0.1f});
1146+
1147+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz.satfinite(<2 x bfloat> splat (bfloat 0xR3DCD)
1148+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz.satfinite(<2 x bfloat> splat (bfloat 0xR3DCD)
1149+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz.satfinite(<2 x bfloat> splat (bfloat 0xR3DCD)
1150+
__nvvm_bf16x2_to_ue8m0x2_rz_satfinite({(__bf16)0.1f, (__bf16)0.1f});
1151+
1152+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp(<2 x bfloat> splat (bfloat 0xR3DCD)
1153+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp(<2 x bfloat> splat (bfloat 0xR3DCD)
1154+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp(<2 x bfloat> splat (bfloat 0xR3DCD)
1155+
__nvvm_bf16x2_to_ue8m0x2_rp({(__bf16)0.1f, (__bf16)0.1f});
1156+
1157+
// CHECK_PTX86_SM100a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp.satfinite(<2 x bfloat> splat (bfloat 0xR3DCD)
1158+
// CHECK_PTX86_SM101a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp.satfinite(<2 x bfloat> splat (bfloat 0xR3DCD)
1159+
// CHECK_PTX86_SM120a: call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp.satfinite(<2 x bfloat> splat (bfloat 0xR3DCD)
1160+
__nvvm_bf16x2_to_ue8m0x2_rp_satfinite({(__bf16)0.1f, (__bf16)0.1f});
1161+
1162+
// CHECK_PTX86_SM100a: call <2 x bfloat> @llvm.nvvm.ue8m0x2.to.bf16x2(i16 19532)
1163+
// CHECK_PTX86_SM101a: call <2 x bfloat> @llvm.nvvm.ue8m0x2.to.bf16x2(i16 19532)
1164+
// CHECK_PTX86_SM120a: call <2 x bfloat> @llvm.nvvm.ue8m0x2.to.bf16x2(i16 19532)
1165+
__nvvm_ue8m0x2_to_bf16x2(0x4C4C);
1166+
10251167
#endif
10261168
// CHECK: ret void
10271169
}

‎llvm/include/llvm/IR/IntrinsicsNVVM.td‎

Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1628,6 +1628,37 @@ let TargetPrefix = "nvvm" in {
16281628
Intrinsic<[llvm_v2f16_ty], [llvm_i16_ty], [IntrNoMem, IntrNoCallback]>;
16291629
def int_nvvm_e5m2x2_to_f16x2_rn_relu : ClangBuiltin<"__nvvm_e5m2x2_to_f16x2_rn_relu">,
16301630
Intrinsic<[llvm_v2f16_ty], [llvm_i16_ty], [IntrNoMem, IntrNoCallback]>;
1631+
1632+
class CVT_FF_TO_I16<string type, string suffix> : ClangBuiltin<!strconcat("__nvvm_ff_to_", type, suffix)>,
1633+
DefaultAttrsIntrinsic<[llvm_i16_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem, IntrNoCallback]>;
1634+
1635+
class CVT_I16_TO_F16X2<string type, string suffix>
1636+
: ClangBuiltin<!strconcat("__nvvm_", type, "_to_f16x2", suffix)>,
1637+
DefaultAttrsIntrinsic<[llvm_v2f16_ty], [llvm_i16_ty], [IntrNoMem, IntrNoCallback]>;
1638+
1639+
class CVT_BF16X2_TO_I16<string type, string suffix> : ClangBuiltin<!strconcat("__nvvm_bf16x2_to_", type, suffix)>,
1640+
DefaultAttrsIntrinsic<[llvm_i16_ty], [llvm_v2bf16_ty], [IntrNoMem, IntrNoCallback]>;
1641+
1642+
// FP6 conversions.
1643+
foreach type = ["e2m3x2", "e3m2x2"] in {
1644+
foreach relu = ["", "_relu"] in {
1645+
defvar suffix = !strconcat("_rn", relu);
1646+
def int_nvvm_ff_to_ # type # suffix # _satfinite : CVT_FF_TO_I16<type, !strconcat(suffix, "_satfinite")>;
1647+
def int_nvvm_ # type # _to_f16x2 # suffix : CVT_I16_TO_F16X2<type, suffix>;
1648+
}
1649+
}
1650+
1651+
// UE8M0x2 conversions.
1652+
foreach rmode = ["_rz", "_rp"] in {
1653+
foreach satmode = ["", "_satfinite"] in {
1654+
defvar suffix = !strconcat(rmode, satmode);
1655+
def int_nvvm_ff_to_ue8m0x2 # suffix : CVT_FF_TO_I16<"ue8m0x2", suffix>;
1656+
def int_nvvm_bf16x2_to_ue8m0x2 # suffix : CVT_BF16X2_TO_I16<"ue8m0x2", suffix>;
1657+
}
1658+
}
1659+
1660+
def int_nvvm_ue8m0x2_to_bf16x2 : ClangBuiltin<"__nvvm_ue8m0x2_to_bf16x2">,
1661+
Intrinsic<[llvm_v2bf16_ty], [llvm_i16_ty], [IntrNoMem, IntrNoCallback]>;
16311662

16321663
// FNS
16331664

‎llvm/lib/Target/NVPTX/NVPTXInstrInfo.td‎

Lines changed: 35 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -703,6 +703,41 @@ let hasSideEffects = false in {
703703
defm CVT_to_tf32_rz_satf : CVT_TO_TF32<"rz.satfinite", [hasPTX<86>, hasSM<100>]>;
704704
defm CVT_to_tf32_rn_relu_satf : CVT_TO_TF32<"rn.relu.satfinite", [hasPTX<86>, hasSM<100>]>;
705705
defm CVT_to_tf32_rz_relu_satf : CVT_TO_TF32<"rz.relu.satfinite", [hasPTX<86>, hasSM<100>]>;
706+
707+
// FP6 conversions.
708+
foreach type = ["e2m3x2", "e3m2x2"] in {
709+
def CVT_ # type # _f32_sf : NVPTXInst<(outs Int16Regs:$dst),
710+
(ins Float32Regs:$src1,
711+
Float32Regs:$src2, CvtMode:$mode),
712+
"cvt${mode:base}.satfinite${mode:relu}."
713+
# type # ".f32 \t$dst, $src1, $src2;", []>;
714+
def CVT_f16x2_ # type : NVPTXInst<(outs Int32Regs:$dst),
715+
(ins Int16Regs:$src, CvtMode:$mode),
716+
"cvt${mode:base}${mode:relu}.f16x2."
717+
# type # " \t$dst, $src;", []>;
718+
}
719+
720+
// UE8M0x2 conversions.
721+
class CVT_f32_to_ue8m0x2<string sat = ""> :
722+
NVPTXInst<(outs Int16Regs:$dst),
723+
(ins Float32Regs:$src1, Float32Regs:$src2, CvtMode:$mode),
724+
"cvt${mode:base}" # sat # ".ue8m0x2.f32 \t$dst, $src1, $src2;", []>;
725+
726+
class CVT_bf16x2_to_ue8m0x2<string sat = ""> :
727+
NVPTXInst<(outs Int16Regs:$dst),
728+
(ins Int32Regs:$src, CvtMode:$mode),
729+
"cvt${mode:base}" # sat # ".ue8m0x2.bf16x2 \t$dst, $src;", []>;
730+
731+
def CVT_ue8m0x2_f32 : CVT_f32_to_ue8m0x2;
732+
def CVT_ue8m0x2_f32_sf : CVT_f32_to_ue8m0x2<".satfinite">;
733+
def CVT_ue8m0x2_bf16x2 : CVT_bf16x2_to_ue8m0x2;
734+
def CVT_ue8m0x2_bf16x2_sf : CVT_bf16x2_to_ue8m0x2<".satfinite">;
735+
736+
def CVT_bf16x2_ue8m0x2 :
737+
NVPTXInst<(outs Int32Regs:$dst),
738+
(ins Int16Regs:$src),
739+
"cvt.rn.bf16x2.ue8m0x2 \t$dst, $src;", []>;
740+
706741
}
707742

708743
def fpround_oneuse : OneUse1<fpround>;

0 commit comments

Comments
 (0)