mirror of
https://github.com/amd/blis.git
synced 2026-05-13 02:25:39 +00:00
5bdf5e2aaa
- Re-designed the new edge kernels that uses masked load-store instructions for handling corner cases. - Mask load-store instruction macros are added. vmovdqu, VMOVDQU for setting up the mask. vmaskmovpd, VMASKMOVPD for masked load-store - Following edge kernels are added for 6x8m dgemm sup. n-left edge kernels - bli_dgemmsup_rv_haswell_asm_6x7m - bli_dgemmsup_rv_haswell_asm_6x5m - bli_dgemmsup_rv_haswell_asm_6x3m m-left edge kernels - bli_dgemmsup_rv_haswell_asm_5x7 - bli_dgemmsup_rv_haswell_asm_4x7 - bli_dgemmsup_rv_haswell_asm_3x7 - bli_dgemmsup_rv_haswell_asm_2x7 - bli_dgemmsup_rv_haswell_asm_1x7 - bli_dgemmsup_rv_haswell_asm_5x5 - bli_dgemmsup_rv_haswell_asm_4x5 - bli_dgemmsup_rv_haswell_asm_3x5 - bli_dgemmsup_rv_haswell_asm_2x5 - bli_dgemmsup_rv_haswell_asm_1x5 - bli_dgemmsup_rv_haswell_asm_5x3 - bli_dgemmsup_rv_haswell_asm_4x3 - bli_dgemmsup_rv_haswell_asm_3x3 - bli_dgemmsup_rv_haswell_asm_2x3 - bli_dgemmsup_rv_haswell_asm_1x3 - For 16x3 dgemm_small, m_left computation is handled with masked load-store instructions avoid overhead of conditional checks for edge cases. - It improves performance by reducing branching overhead and by being more cache friendly. AMD-Internal: [CPUPL-3574] Change-Id: I976d6a9209d2a1a02b2830d03d21d200a5aad173