c@64: #ifndef CMM_H
c@64:    #define CMM_H
c@64: 
c@64:    #define ATL_mmMULADD
c@64:    #define ATL_mmLAT 1
c@64:    #define ATL_mmMU  12
c@64:    #define ATL_mmNU  1
c@64:    #define ATL_mmKU  72
c@64:    #define MB 72
c@64:    #define NB 72
c@64:    #define KB 72
c@64:    #define NBNB 5184
c@64:    #define MBNB 5184
c@64:    #define MBKB 5184
c@64:    #define NBKB 5184
c@64:    #define NB2 144
c@64:    #define NBNB2 10368
c@64: 
c@64:    #define ATL_MulByNB(N_) ((N_) * 72)
c@64:    #define ATL_DivByNB(N_) ((N_) / 72)
c@64:    #define ATL_MulByNBNB(N_) ((N_) * 5184)
c@64: 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);
c@64: 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);
c@64: 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);
c@64: 
c@64:    #define NBmm_b1(m_, n_, k_, al_, A_, lda_, B_, ldb_, be_, C_, ldc_) \
c@64: { \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, ATL_rnone, C_, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, ATL_rone, (C_)+1, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_), ldb_, ATL_rone, (C_)+1, ldc_); \
c@64:    }
c@64:    #define NBmm_b0(m_, n_, k_, al_, A_, lda_, B_, ldb_, be_, C_, ldc_) \
c@64: { \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_b0(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, ATL_rzero, C_, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_b0(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, ATL_rzero, (C_)+1, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_), ldb_, ATL_rone, (C_)+1, ldc_); \
c@64:    }
c@64:    #define NBmm_bX(m_, n_, k_, al_, A_, lda_, B_, ldb_, be_, C_, ldc_) \
c@64: { \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_), ldb_, -(be_), C_, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_), lda_, (B_)+NBNB, ldb_, be_, (C_)+1, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_bX(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_)+NBNB, ldb_, ATL_rnone, C_, ldc_); \
c@64:    ATL_cJIK72x72x72TN72x72x0_a1_b1(m_, n_, k_, al_, (A_)+NBNB, lda_, (B_), ldb_, ATL_rone, (C_)+1, ldc_); \
c@64:    }
c@64:    #define rNBmm_b1 ATL_sJIK72x72x72TN72x72x0_a1_b1
c@64:    #define rNBmm_b0 ATL_sJIK72x72x72TN72x72x0_a1_b0
c@64:    #define rNBmm_bX ATL_sJIK72x72x72TN72x72x0_a1_bX
c@64: 
c@64: #endif