sbgemm: cooperlake: prefetch A & B

diff --git a/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c b/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c
index 0280b44..7af51b6 100644 (file)
--- a/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c
+++ b/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c
@@ -64,6 +64,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define DECLARE_B_PAIR() \
        __m512i B_lo; __m512i B_hi;
 
+#define PREFETCH_B_STEP 32
+#define PREFETCH_B(Bx, By) \
+       if (By == 0) asm("prefetcht0 %c1(%0)": : "r"(ptr_b##Bx), "n"(PREFETCH_B_STEP * 2)); \
+       else asm("prefetcht0 %c3(%0, %1, %c2)": : "r"(ptr_b##Bx), "r"(n_blksize), "n"(By*2), "n"(PREFETCH_B_STEP * 2))
+
 #define BROADCAST_B_PAIR(Bx, By) \
        BROADCAST64(ptr_b##Bx, n_blksize, By, 0, B_lo); \
        BROADCAST64(ptr_b##Bx, n_blksize, By, 4, B_hi);
@@ -204,17 +209,19 @@ int CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A, IFLOAT *
                        k_count = k;
                        for (; k_count > 3; k_count -=4) {
                                LOAD_A_PAIR(0);
+                               _mm_prefetch(ptr_a0 + 128, _MM_HINT_T0);
                                ptr_a0 += 16 * 2;
-                               BROADCAST_B_PAIR(0, 0); MATMUL_4X(0, 0, 0);
-                               BROADCAST_B_PAIR(0, 1); MATMUL_4X(0, 0, 1);
-                               BROADCAST_B_PAIR(0, 2); MATMUL_4X(0, 0, 2);
+                               BROADCAST_B_PAIR(0, 0); PREFETCH_B(0, 0); MATMUL_4X(0, 0, 0);
+                               BROADCAST_B_PAIR(0, 1); PREFETCH_B(0, 1); MATMUL_4X(0, 0, 1);
+                               BROADCAST_B_PAIR(0, 2); PREFETCH_B(0, 2); MATMUL_4X(0, 0, 2);
                                ptr_b0 += 4 * 2;
-                               BROADCAST_B_PAIR(1, 0); MATMUL_4X(0, 1, 0);
-                               BROADCAST_B_PAIR(1, 1); MATMUL_4X(0, 1, 1);
-                               BROADCAST_B_PAIR(1, 2); MATMUL_4X(0, 1, 2);
+                               BROADCAST_B_PAIR(1, 0); PREFETCH_B(1, 0); MATMUL_4X(0, 1, 0);
+                               BROADCAST_B_PAIR(1, 1); PREFETCH_B(1, 1); MATMUL_4X(0, 1, 1);
+                               BROADCAST_B_PAIR(1, 2); PREFETCH_B(1, 2); MATMUL_4X(0, 1, 2);
                                ptr_b1 += 4 * 2;
 
                                LOAD_A_PAIR(0);
+                               _mm_prefetch(ptr_a0 + 128, _MM_HINT_T0);
                                ptr_a0 += 16 * 2;
                                BROADCAST_B_PAIR(0, 0); MATMUL_4X(0, 0, 0);
                                BROADCAST_B_PAIR(0, 1); MATMUL_4X(0, 0, 1);