leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 24 Packages Projects Releases Wiki Activity

Updates for 3.0 (#857)

Browse Source 
Co-authored-by: Aniket Shivam <ashivam@nvidia.com>
This commit is contained in:
ANIKET SHIVAM
2023-03-09 15:27:40 -05:00
committed by GitHub
co-authored by Aniket Shivam
parent a68e2f95f0
commit c4f6b8c6bc
7 changed files with 131 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/float8.h
+6 -6
View File
@@ -399,7 +399,7 @@ struct alignas(1) float_e4m3_t : float8_base<FloatEncoding::E4M3> {
return *reinterpret_cast<float_e4m3_t *>(&tmp);
#else
return bitcast(Base::convert_float_to_fp8(float(flt)));
return bitcast(Base::convert_float_to_fp8(__half2float(flt)));
#endif
}
@@ -413,7 +413,7 @@ struct alignas(1) float_e4m3_t : float8_base<FloatEncoding::E4M3> {
return reinterpret_cast<half2 const &>(packed).x;
#else
return half(Base::convert_fp8_to_float(x.storage));
return __float2half(Base::convert_fp8_to_float(x.storage));
#endif
}
@@ -425,7 +425,7 @@ struct alignas(1) float_e4m3_t : float8_base<FloatEncoding::E4M3> {
uint32_t packed;
asm volatile("cvt.rn.f16x2.e4m3x2 %0, %1;\n" : "=r"(packed) : "h"(bits));
return float(reinterpret_cast<half2 const &>(packed).x);
return __half2float(reinterpret_cast<half2 const &>(packed).x);
#else
return Base::convert_fp8_to_float(x.storage);
#endif
@@ -609,7 +609,7 @@ struct alignas(1) float_e5m2_t : float8_base<FloatEncoding::E5M2> {
return *reinterpret_cast<float_e5m2_t *>(&tmp);
#else
return bitcast(Base::convert_float_to_fp8(float(flt)));
return bitcast(Base::convert_float_to_fp8(__half2float(flt)));
#endif
}
@@ -623,7 +623,7 @@ struct alignas(1) float_e5m2_t : float8_base<FloatEncoding::E5M2> {
return reinterpret_cast<half2 const &>(packed).x;
#else
return half(Base::convert_fp8_to_float(x.storage));
return __float2half(Base::convert_fp8_to_float(x.storage));
#endif
}
@@ -635,7 +635,7 @@ struct alignas(1) float_e5m2_t : float8_base<FloatEncoding::E5M2> {
uint32_t packed;
asm volatile("cvt.rn.f16x2.e5m2x2 %0, %1;\n" : "=r"(packed) : "h"(bits));
return float(reinterpret_cast<half2 const &>(packed).x);
return __half2float(reinterpret_cast<half2 const &>(packed).x);
#else
return Base::convert_fp8_to_float(x.storage);
#endif