From 24091cb62b951b4323065c7a3bd076f742a88e09 Mon Sep 17 00:00:00 2001
From: akielaries <akiel@akiel.org>
Date: Mon, 26 Feb 2024 21:58:46 -0700
Subject: [PATCH] #EDITS: some mass updated to linear algebra module and tests

---
 modules/linalg/dgemm_arr.cpp | 107 +----------------------------------
 1 file changed, 3 insertions(+), 104 deletions(-)

diff --git a/modules/linalg/dgemm_arr.cpp b/modules/linalg/dgemm_arr.cpp
index aa12b92d1..ac732c51d 100644
--- a/modules/linalg/dgemm_arr.cpp
+++ b/modules/linalg/dgemm_arr.cpp
@@ -41,7 +41,7 @@
 #include <stdlib.h>
 #include <time.h>
 
-#if defined(__SSE__)
+#if defined(__SSE2__)
 
 #ifdef __cplusplus
 extern "C" {
@@ -168,33 +168,7 @@ void gpmp::linalg::DGEMM::pack_buffer_B(int kc,
     }
 }
 
-// micro kernel that multiplies panels from A and B using assembly kernels
-void gpmp::linalg::DGEMM::dgemm_micro_kernel(long kc,
-                                             double alpha,
-                                             const double *A,
-                                             const double *B,
-                                             double beta,
-                                             double *C,
-                                             long incRowC,
-                                             long incColC,
-                                             const double *nextA,
-                                             const double *nextB) {
-    long kb = kc / 4;
-    long kl = kc % 4;
-
-    dgemm_kernel_asm(A,
-                     B,
-                     C,
-                     nextA,
-                     nextB,
-                     kl,
-                     kb,
-                     incRowC,
-                     incColC,
-                     alpha,
-                     beta);
-}
-
+// micro kernel that multiplies panels from A and B
 void gpmp::linalg::DGEMM::dgemm_micro_kernel(int kc,
                                              double alpha,
                                              const double *A,
@@ -333,77 +307,6 @@ void gpmp::linalg::DGEMM::dgemm_macro_kernel(int mc,
     int mr, nr;
     int i, j;
 
-// use assembly kernel function
-#if defined(__SSE__)
-    const double *nextA;
-    const double *nextB;
-
-    for (j = 0; j < np; ++j) {
-        nr = (j != np - 1 || _nr == 0) ? BLOCK_SZ_NR : _nr;
-        nextB = &DGEMM_BUFF_B[j * kc * BLOCK_SZ_NR];
-
-        for (i = 0; i < mp; ++i) {
-            mr = (i != mp - 1 || _mr == 0) ? BLOCK_SZ_MR : _mr;
-            nextA = &DGEMM_BUFF_A[(i + 1) * kc * BLOCK_SZ_MR];
-
-            if (i == mp - 1) {
-                nextA = DGEMM_BUFF_A;
-                nextB = &DGEMM_BUFF_B[(j + 1) * kc * BLOCK_SZ_NR];
-                if (j == np - 1) {
-                    nextB = DGEMM_BUFF_B;
-                }
-            }
-
-            if (mr == BLOCK_SZ_MR && nr == BLOCK_SZ_NR) {
-
-                dgemm_micro_kernel(
-                    kc,
-                    alpha,
-                    &DGEMM_BUFF_A[i * kc * BLOCK_SZ_MR],
-                    &DGEMM_BUFF_B[j * kc * BLOCK_SZ_NR],
-                    beta,
-                    &C[i * BLOCK_SZ_MR * incRowC + j * BLOCK_SZ_NR * incColC],
-                    incRowC,
-                    incColC,
-                    nextA,
-                    nextB);
-            }
-
-            else {
-                dgemm_micro_kernel(kc,
-                                   alpha,
-                                   &DGEMM_BUFF_A[i * kc * BLOCK_SZ_MR],
-                                   &DGEMM_BUFF_B[j * kc * BLOCK_SZ_NR],
-                                   0.0,
-                                   DGEMM_BUFF_C,
-                                   1,
-                                   BLOCK_SZ_MR,
-                                   nextA,
-                                   nextB);
-                dgescal(
-                    mr,
-                    nr,
-                    beta,
-                    &C[i * BLOCK_SZ_MR * incRowC + j * BLOCK_SZ_NR * incColC],
-                    incRowC,
-                    incColC);
-                dgeaxpy(mr,
-                        nr,
-                        1.0,
-                        DGEMM_BUFF_C,
-                        1,
-                        BLOCK_SZ_MR,
-                        &DGEMM_BUFF_C[i * BLOCK_SZ_MR * incRowC +
-                                      j * BLOCK_SZ_NR * incColC],
-                        incRowC,
-                        incColC);
-            }
-        }
-    }
-
-// default implementation
-#else
-
     for (j = 0; j < np; ++j) {
         nr = (j != np - 1 || _nr == 0) ? BLOCK_SZ_NR : _nr;
 
@@ -420,9 +323,7 @@ void gpmp::linalg::DGEMM::dgemm_macro_kernel(int mc,
                     &C[i * BLOCK_SZ_MR * incRowC + j * BLOCK_SZ_NR * incColC],
                     incRowC,
                     incColC);
-            }
-
-            else {
+            } else {
                 dgemm_micro_kernel(kc,
                                    alpha,
                                    &DGEMM_BUFF_A[i * kc * BLOCK_SZ_MR],
@@ -451,8 +352,6 @@ void gpmp::linalg::DGEMM::dgemm_macro_kernel(int mc,
             }
         }
     }
-
-#endif
 }
 
 // Main DGEMM entrypoint, compute C <- beta*C + alpha*A*B