#ifndef ZMM_H #define ZMM_H #define ATL_mmMULADD #define ATL_mmLAT 10 #define ATL_mmMU 5 #define ATL_mmNU 4 #define ATL_mmKU 48 #define MB 48 #define NB 48 #define KB 48 #define NBNB 2304 #define MBNB 2304 #define MBKB 2304 #define NBKB 2304 #define NB2 96 #define NBNB2 4608 #define ATL_MulByNB(N_) ((N_) * 48) #define ATL_DivByNB(N_) ((N_) / 48) #define ATL_MulByNBNB(N_) ((N_) * 2304) void ATL_zJIK48x48x48TN48x48x0_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_zJIK48x48x48TN48x48x0_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_zJIK48x48x48TN48x48x0_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_zJIK48x48x48TN48x48x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, ATL_rnone, C_, ldc_); \ ATL_zJIK48x48x48TN48x48x0_a1_b1(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, ATL_rone, (C_)+1, ldc_); \ ATL_zJIK48x48x48TN48x48x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \ ATL_zJIK48x48x48TN48x48x0_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_zJIK48x48x48TN48x48x0_a1_b0(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, ATL_rzero, C_, ldc_); \ ATL_zJIK48x48x48TN48x48x0_a1_b0(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, ATL_rzero, (C_)+1, ldc_); \ ATL_zJIK48x48x48TN48x48x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \ ATL_zJIK48x48x48TN48x48x0_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_zJIK48x48x48TN48x48x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, -(be_), C_, ldc_); \ ATL_zJIK48x48x48TN48x48x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, be_, (C_)+1, ldc_); \ ATL_zJIK48x48x48TN48x48x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \ ATL_zJIK48x48x48TN48x48x0_a1_b1(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_), ldb_, ATL_rone, (C_)+1, ldc_); \ } #define rNBmm_b1 ATL_dJIK48x48x48TN48x48x0_a1_b1 #define rNBmm_b0 ATL_dJIK48x48x48TN48x48x0_a1_b0 #define rNBmm_bX ATL_dJIK48x48x48TN48x48x0_a1_bX #endif