Add flag for not converting A & B - will be used in future to do conversion during packing.

Author: ChipKerchner

kernel/riscv64/sbgemm_kernel_16x8_zvl256b.c

@@ -42,7 +42,9 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
     if ((M >= 4) && (N >= 4) && (K > 0)) {
         CONV = (FLOAT *)(malloc((K * (8 + (M & -4))) * sizeof(FLOAT)));
         if (!CONV) return 1;
+#ifndef BF16_DONT_CONV
         B_CONV(AA, CONV + (K * 8), (M & -4) * K);
+#endif
     }
 #endif
 
@@ -52,9 +54,11 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
         BLASLONG gvl = __riscv_vsetvl_e16m1(16);
 #ifdef BF16_WIDEN_ONE
         BLASLONG bi2 = K * 8;
+#ifndef BF16_DONT_CONV
         if (M >= 4) {
             B_CONV(BB + (n_top*K), CONV, bi2);
         }
+#endif
         BLASLONG ai2 = K * 8;
 #endif
 
@@ -476,9 +480,11 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
 
 #ifdef BF16_WIDEN_ONE
         BLASLONG bi2 = K * 4;
+#ifndef BF16_DONT_CONV
         if (M >= 4) {
             B_CONV(BB + (n_top*K), CONV, bi2);
         }
+#endif
         BLASLONG ai2 = K * 8;
 #endif
 

kernel/riscv64/sbgemm_kernel_8x8_zvl128b.c

@@ -42,7 +42,9 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
     if ((M >= 4) && (N >= 4) && (K > 0)) {
         CONV = (FLOAT *)(malloc((K * (8 + (M & -4))) * sizeof(FLOAT)));
         if (!CONV) return 1;
+#ifndef BF16_DONT_CONV
         B_CONV(AA, CONV + (K * 8), (M & -4) * K);
+#endif
     }
 #endif
 
@@ -53,9 +55,11 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
 
 #ifdef BF16_WIDEN_ONE
         BLASLONG bi2 = K * 8;
+#ifndef BF16_DONT_CONV
         if (M >= 4) {
           B_CONV(BB + (n_top*K), CONV, bi2);
         }
+#endif
         BLASLONG ai2 = K * 8;
 #endif
 
@@ -379,9 +383,11 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
 
 #ifdef BF16_WIDEN_ONE
         BLASLONG bi2 = K * 4;
+#ifndef BF16_DONT_CONV
         if (M >= 4) {
             B_CONV(BB + (n_top*K), CONV, bi2);
         }
+#endif
         BLASLONG ai2 = K * 8;
 #endif