mirror of
https://github.com/amd/blis.git
synced 2026-05-24 18:34:40 +00:00
Fix SVE Compil.
This commit is contained in:
committed by
Devin Matthews
parent
b5df1811f1
commit
9cc897f374
@@ -41,6 +41,16 @@
|
||||
#define BLIS_SIMD_ALIGN_SIZE 256
|
||||
#define BLIS_SIMD_NUM_REGISTERS 32
|
||||
|
||||
// SVE-specific configs.
|
||||
#define N_L1_SVE_DEFAULT 64
|
||||
#define W_L1_SVE_DEFAULT 4
|
||||
#define C_L1_SVE_DEFAULT 256
|
||||
#define N_L2_SVE_DEFAULT 2048
|
||||
#define W_L2_SVE_DEFAULT 16
|
||||
#define C_L2_SVE_DEFAULT 256
|
||||
#define N_L3_SVE_DEFAULT 8192
|
||||
#define W_L3_SVE_DEFAULT 16
|
||||
#define C_L3_SVE_DEFAULT 256
|
||||
|
||||
//#endif
|
||||
|
||||
|
||||
@@ -140,7 +140,7 @@ GEMM_ACOLCMPLX_CONTIGUOUS_LOAD_FWD(z28,z29,p0,%0,x2)
|
||||
" prfm PLDL1KEEP, [x16] \n\t"
|
||||
" add x16, x16, %4 \n\t"
|
||||
" prfm PLDL1KEEP, [x16] \n\t"
|
||||
// " END_CCOL_PRFM: \n\t"
|
||||
" END_CCOL_PRFM: \n\t"
|
||||
" \n\t"
|
||||
CLEAR_COL20(z0,z1,z2,z3,z4,z5,z6,z7,z8,z9,z10,z11,z12,z13,z14,z15,z16,z17,z18,z19)
|
||||
" \n\t"
|
||||
|
||||
@@ -139,7 +139,7 @@ GEMM_ACOL_CONTIGUOUS_LOAD(z28,z29,p0,p0,x0)
|
||||
" prfm PLDL1KEEP, [x16] \n\t"
|
||||
" add x16, x16, x7 \n\t"
|
||||
" prfm PLDL1KEEP, [x16] \n\t"
|
||||
// " END_CCOL_PRFM: \n\t"
|
||||
" END_CCOL_PRFM: \n\t"
|
||||
" \n\t"
|
||||
CLEAR_COL20(z0,z1,z2,z3,z4,z5,z6,z7,z8,z9,z10,z11,z12,z13,z14,z15,z16,z17,z18,z19)
|
||||
" \n\t"
|
||||
|
||||
@@ -139,7 +139,7 @@ GEMM_ACOL_CONTIGUOUS_LOAD(z28,z29,p0,p0,x0)
|
||||
" prfm PLDL1STRM, [x16] \n\t"
|
||||
" add x16, x16, x7 \n\t"
|
||||
" prfm PLDL1STRM, [x16] \n\t"
|
||||
// " END_CCOL_PRFM: \n\t"
|
||||
" END_CCOL_PRFM: \n\t"
|
||||
" \n\t"
|
||||
CLEAR_COL20(z0,z1,z2,z3,z4,z5,z6,z7,z8,z9,z10,z11,z12,z13,z14,z15,z16,z17,z18,z19)
|
||||
" \n\t"
|
||||
|
||||
@@ -140,7 +140,7 @@ GEMM_ACOLCMPLX_CONTIGUOUS_LOAD_FWD(z28,z29,p0,%0,x2)
|
||||
" prfm PLDL1KEEP, [x16] \n\t"
|
||||
" add x16, x16, %4 \n\t"
|
||||
" prfm PLDL1KEEP, [x16] \n\t"
|
||||
// " END_CCOL_PRFM: \n\t"
|
||||
" END_CCOL_PRFM: \n\t"
|
||||
" \n\t"
|
||||
CLEAR_COL20(z0,z1,z2,z3,z4,z5,z6,z7,z8,z9,z10,z11,z12,z13,z14,z15,z16,z17,z18,z19)
|
||||
" \n\t"
|
||||
|
||||
Reference in New Issue
Block a user