#AMD has re-released PDFs 69191 and 69192, removing the "AMD Confidential" headers.
#AVX512_BMM specification:
docs.amd.com/v/u/en-US/69...
#FRED specification:
docs.amd.com/v/u/en-US/69...
5
2
0
0
Hashtag
#AVX512_BMM
Advertisement · 728 × 90
One possible application of the #AVX512_BMM VPMACOR16x16x16 instruction is the compact index2msk function: if src2 is a full mask and src3 contains variably shifted masks, then VPMACOR16x16x16 will collect them into a single word. The order and repetitions are arbitrary.
2
0
1
0
Third official #AMD #Zen6 docs:
"AMD64 Bit Matrix Multiply and Bit Reversal Instructions" specification #AVX512_BMM 69192 v1.00 pdf:
docs.amd.com/v/u/en-US/69...
7
0
0
3
#AMD released the "AMD64 Architecture Programmer’s Manual Volume 4: 128-Bit, 256-Bit, and 512-Bit Media Instructions" 26568 pdf v3.26
docs.amd.com/v/u/en-US/26...
This describes instructions up to #Zen5
#AVX512_BMM
#AVX512_FP16
#AVX_NE_CONVERT
#AVX_IFMA
#AVX_VNNI_INT8
from #Zen6 are missing
3
1
0
0
#AMD #Zen6 znver6 ISA:
- #AVX512_BMM (CPUID.80000021.EAX[23], VBMACOR16x16x16, VBMACXOR16x16x16, VBITREV)
- #AVX512_FP16
- #AVX_NE_CONVERT
- #AVX_IFMA
- #AVX_VNNI_INT8
Source:
sourceware.org/pipermail/bi...
2
1
1
1