Mercurial > hg > qm-vamp-plugins
view build/linux/amd64/atlas/cmm.h @ 80:e7c785094e7b
* update readmes &c
author | Chris Cannam <c.cannam@qmul.ac.uk> |
---|---|
date | Tue, 18 Nov 2008 14:51:49 +0000 |
parents | ddea89113517 |
children |
line wrap: on
line source
#ifndef CMM_H #define CMM_H #define ATL_mmMULADD #define ATL_mmLAT 1 #define ATL_mmMU 12 #define ATL_mmNU 1 #define ATL_mmKU 72 #define MB 72 #define NB 72 #define KB 72 #define NBNB 5184 #define MBNB 5184 #define MBKB 5184 #define NBKB 5184 #define NB2 144 #define NBNB2 10368 #define ATL_MulByNB(N_) ((N_) * 72) #define ATL_DivByNB(N_) ((N_) / 72) #define ATL_MulByNBNB(N_) ((N_) * 5184) void ATL_cJIK72x72x72TN72x72x0_a1_b0(const int M, const int N, const int K, const TYPE alpha, const TYPE *A, const int lda, const TYPE *B, const int ldb, const TYPE beta, TYPE *C, const int ldc); void ATL_cJIK72x72x72TN72x72x0_a1_b1(const int M, const int N, const int K, const TYPE alpha, const TYPE *A, const int lda, const TYPE *B, const int ldb, const TYPE beta, TYPE *C, const int ldc); void ATL_cJIK72x72x72TN72x72x0_a1_bX(const int M, const int N, const int K, const TYPE alpha, const TYPE *A, const int lda, const TYPE *B, const int ldb, const TYPE beta, TYPE *C, const int ldc); #define NBmm_b1(m_, n_, k_, al_, A_, lda_, B_, ldb_, be_, C_, ldc_) \ { \ ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, ATL_rnone, C_, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, ATL_rone, (C_)+1, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_), ldb_, ATL_rone, (C_)+1, ldc_); \ } #define NBmm_b0(m_, n_, k_, al_, A_, lda_, B_, ldb_, be_, C_, ldc_) \ { \ ATL_cJIK72x72x72TN72x72x0_a1_b0(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, ATL_rzero, C_, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_b0(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, ATL_rzero, (C_)+1, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_), ldb_, ATL_rone, (C_)+1, ldc_); \ } #define NBmm_bX(m_, n_, k_, al_, A_, lda_, B_, ldb_, be_, C_, ldc_) \ { \ ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, -(be_), C_, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, be_, (C_)+1, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \ ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_), ldb_, ATL_rone, (C_)+1, ldc_); \ } #define rNBmm_b1 ATL_sJIK72x72x72TN72x72x0_a1_b1 #define rNBmm_b0 ATL_sJIK72x72x72TN72x72x0_a1_b0 #define rNBmm_bX ATL_sJIK72x72x72TN72x72x0_a1_bX #endif