leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

CUTLASS 2.10 (#615)

Browse Source 
Co-authored-by: Aniket Shivam <ashivam@nvidia.com>
This commit is contained in:
ANIKET SHIVAM
2022-09-03 15:48:46 -07:00
committed by GitHub
parent ca23ff7924
commit b72cbf957d
289 changed files with 43708 additions and 2513 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
include/cutlass/numeric_conversion.h
View File

@@ -185,7 +185,8 @@ struct NumericConverter<int8_t, float, FloatRoundStyle::round_to_nearest> {
CUTLASS_DEVICE
static result_type convert(source_type const & s) {
int32_t intermediate = __float2int_rn(s);
int32_t intermediate;
asm volatile("cvt.rni.sat.s8.f32 %0, %1;" : "=r"(intermediate) : "f"(s));
return static_cast<result_type>(intermediate);
}
@@ -206,7 +207,8 @@ struct NumericConverter<int8_t, float, FloatRoundStyle::round_toward_zero> {
CUTLASS_DEVICE
static result_type convert(source_type const & s) {
int32_t intermediate = __float2int_rz(s);
int32_t intermediate;
asm volatile("cvt.rzi.sat.s8.f32 %0, %1;" : "=r"(intermediate) : "f"(s));
return static_cast<result_type>(intermediate);
}
@@ -228,7 +230,14 @@ struct NumericConverter<int8_t, float, FloatRoundStyle::round_to_nearest> {
static result_type convert(source_type const & s) {
std::fesetround(FE_TONEAREST);
int32_t intermediate = (result_type)std::nearbyint(s);
int32_t intermediate = (int32_t)std::nearbyint(s);
// Low-end saturation
intermediate = std::max(intermediate, (int32_t)std::numeric_limits<int8_t>::lowest());
// High-end saturation
intermediate = std::min(intermediate, (int32_t)std::numeric_limits<int8_t>::max());
return static_cast<result_type>(intermediate);
}
@@ -246,7 +255,14 @@ struct NumericConverter<int8_t, float, FloatRoundStyle::round_toward_zero> {
static result_type convert(source_type const & s) {
std::fesetround(FE_TOWARDZERO);
int32_t intermediate = (result_type)std::nearbyint(s);
int32_t intermediate = (int32_t)std::nearbyint(s);
// Low-end saturation
intermediate = std::max(intermediate, (int32_t)std::numeric_limits<int8_t>::lowest());
// High-end saturation
intermediate = std::min(intermediate, (int32_t)std::numeric_limits<int8_t>::max());
return static_cast<result_type>(intermediate);
}