sbgemm: cooperlake: unroll core loop by 2

diff --git a/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c b/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c
index 4c1f506..0280b44 100644 (file)
--- a/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c
+++ b/kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c
@@ -201,7 +201,31 @@ int CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A, IFLOAT *
                        DECLARE_B_PAIR();
                        DECLARE_RESULT_4X(0, 0, 0); DECLARE_RESULT_4X(0, 0, 1); DECLARE_RESULT_4X(0, 0, 2);
                        DECLARE_RESULT_4X(0, 1, 0); DECLARE_RESULT_4X(0, 1, 1); DECLARE_RESULT_4X(0, 1, 2);
-                       for (k_count = k; k_count > 1; k_count -=2) {
+                       k_count = k;
+                       for (; k_count > 3; k_count -=4) {
+                               LOAD_A_PAIR(0);
+                               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);
+                               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);
+                               ptr_b1 += 4 * 2;
+
+                               LOAD_A_PAIR(0);
+                               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);
+                               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);
+                               ptr_b1 += 4 * 2;
+                       }
+                       for (; k_count > 1; k_count -=2) {
                                LOAD_A_PAIR(0);
                                ptr_a0 += 16 * 2;
                                BROADCAST_B_PAIR(0, 0); MATMUL_4X(0, 0, 0);