Merge pull request #5030 from tingboliao/develop

Optimize the zgemm_tcopy_4_rvv function to be compatible with the situations where the vector lengths(vlens) are 128 and 256.
OpenMathLib · Dec 30, 2024 · 73527aa · 73527aa
2 parents 36b0fb3 + c4c3d9e
commit 73527aa
Showing 1 changed file with 25 additions and 111 deletions.
diff --git a/kernel/riscv64/zgemm_tcopy_4_rvv.c b/kernel/riscv64/zgemm_tcopy_4_rvv.c
@@ -28,35 +28,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 
 #if !defined(DOUBLE)
-#define VSETVL(n)               __riscv_vsetvl_e32m1(n)
-#define FLOAT_V_T               vfloat32m1_t
-#define FLOAT_VX2_T             vfloat32m1x2_t
-#define FLOAT_VX4_T             vfloat32m1x4_t
-#define FLOAT_VX8_T             vfloat32m1x8_t
-#define VLEV_FLOAT              __riscv_vle32_v_f32m1
-#define VSEV_FLOAT              __riscv_vse32_v_f32m1
-#define VLSSEG2_FLOAT           __riscv_vlsseg2e32_v_f32m1x2
-#define VLSSEG4_FLOAT           __riscv_vlsseg4e32_v_f32m1x4
-#define VLSSEG8_FLOAT           __riscv_vlsseg8e32_v_f32m1x8
-#define VSSEG2_FLOAT            __riscv_vsseg2e32_v_f32m1x2
-#define VSSEG4_FLOAT            __riscv_vsseg4e32_v_f32m1x4
-#define VSSEG8_FLOAT            __riscv_vsseg8e32_v_f32m1x8
+#define FLOAT_V_T               vfloat32m2_t
+#define FLOAT_V_T_HALF          vfloat32m1_t
+#define VLEV_FLOAT              __riscv_vle32_v_f32m2
+#define VLEV_FLOAT_HALF         __riscv_vle32_v_f32m1
+#define VSEV_FLOAT              __riscv_vse32_v_f32m2
+#define VSEV_FLOAT_HALF         __riscv_vse32_v_f32m1
 #else
-#define VSETVL(n)               __riscv_vsetvl_e64m1(n)
-#define FLOAT_V_T               vfloat64m1_t
-#define FLOAT_VX2_T             vfloat64m1x2_t
-#define FLOAT_VX4_T             vfloat64m1x4_t
-#define FLOAT_VX8_T             vfloat64m1x8_t
-#define VLEV_FLOAT              __riscv_vle64_v_f64m1
-#define VSEV_FLOAT              __riscv_vse64_v_f64m1
-#define VLSSEG2_FLOAT           __riscv_vlsseg2e64_v_f64m1x2
-#define VLSSEG4_FLOAT           __riscv_vlsseg4e64_v_f64m1x4
-#define VLSSEG8_FLOAT           __riscv_vlsseg8e64_v_f64m1x8
-#define VSSEG2_FLOAT            __riscv_vsseg2e64_v_f64m1x2
-#define VSSEG4_FLOAT            __riscv_vsseg4e64_v_f64m1x4
-#define VSSEG8_FLOAT            __riscv_vsseg8e64_v_f64m1x8
+#define FLOAT_V_T               vfloat64m4_t
+#define FLOAT_V_T_HALF          vfloat64m2_t
+#define VLEV_FLOAT              __riscv_vle64_v_f64m4
+#define VLEV_FLOAT_HALF         __riscv_vle64_v_f64m2
+#define VSEV_FLOAT              __riscv_vse64_v_f64m4
+#define VSEV_FLOAT_HALF         __riscv_vse64_v_f64m2
 #endif
 
+
 int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
 
     BLASLONG i, j;
@@ -67,9 +54,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
     IFLOAT *boffset, *boffset1, *boffset2, *boffset3;
 
     FLOAT_V_T v0;
-    FLOAT_VX2_T vx2;
-    FLOAT_VX4_T vx4;
-    FLOAT_VX8_T vx8;
+    FLOAT_V_T_HALF v1;
 
     size_t vl;
 
@@ -80,86 +65,12 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
     boffset2  = b + 2 * m  * (n & ~3);
     boffset3  = b + 2 * m  * (n & ~1);
 
-    for(j = (m >> 2); j > 0; j--) {
-
-        aoffset1  = aoffset;
-        aoffset += 8 * lda;
-
-        boffset1  = boffset;
-        boffset  += 32;
-
-        for(i = (n >> 2); i > 0; i--) {
-            vl = 4;
-
-            vx8 = VLSSEG8_FLOAT(aoffset1, lda * sizeof(FLOAT) * 2, vl);
-            VSSEG8_FLOAT(boffset1, vx8, vl);
-
-            aoffset1 += 8;
-            boffset1 += m * 8;
-        }
-
-        if (n & 2) {
-            vl = 4;
-
-            vx4 = VLSSEG4_FLOAT(aoffset1, lda * sizeof(FLOAT) * 2, vl);
-            VSSEG4_FLOAT(boffset2, vx4, vl);
-
-            aoffset1 += 4;
-            boffset2 += 16;
-        }
-
-        if (n & 1) {
-            vl = 4;
-
-            vx2 = VLSSEG2_FLOAT(aoffset1, lda * sizeof(FLOAT) * 2, vl);
-            VSSEG2_FLOAT(boffset3, vx2, vl);
-
-            aoffset1 += 2;
-            boffset3 += 8;
-        }
-    }
-
-    if (m & 2) {
+    for(j = m; j > 0; j--) {
         aoffset1  = aoffset;
-        aoffset += 4 * lda;
-
         boffset1  = boffset;
-        boffset  += 16;
-
-        for(i = (n >> 2); i > 0; i--) {
-            vl = 2;
-
-            vx8 = VLSSEG8_FLOAT(aoffset1, lda * sizeof(FLOAT) * 2, vl);
-            VSSEG8_FLOAT(boffset1, vx8, vl);
-
-            aoffset1 += 8;
-            boffset1 += m * 8;
-        }
-
-        if (n & 2) {
-            vl = 2;
-
-            vx4 = VLSSEG4_FLOAT(aoffset1, lda * sizeof(FLOAT) * 2, vl);
-            VSSEG4_FLOAT(boffset2, vx4, vl);
-
-            aoffset1 += 4;
-            boffset2 += 8;
-        }
-
-        if (n & 1) {
-            vl = 2;
 
-            vx2 = VLSSEG2_FLOAT(aoffset1, lda * sizeof(FLOAT) * 2, vl);
-            VSSEG2_FLOAT(boffset3, vx2, vl);
-
-            //aoffset1 += 2;
-            boffset3 += 4;
-        }
-    }
-
-    if (m & 1) {
-        aoffset1  = aoffset;
-        boffset1  = boffset;
+        aoffset += 2 * lda;
+        boffset  += 8;
 
         for(i = (n >> 2); i > 0; i--) {
             vl = 8;
@@ -174,16 +85,19 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
         if (n & 2) {
             vl = 4;
 
-            v0 = VLEV_FLOAT(aoffset1, vl);
-            VSEV_FLOAT(boffset2, v0, vl);
+            v1 = VLEV_FLOAT_HALF(aoffset1, vl);
+            VSEV_FLOAT_HALF(boffset2, v1, vl);
 
             aoffset1 += 4;
-            //boffset2 += 4;
+            boffset2 += 4;
         }
 
         if (n & 1) {
-           *(boffset3) = *(aoffset1);
-           *(boffset3 + 1) = *(aoffset1 + 1);
+            *(boffset3) = *(aoffset1);
+            *(boffset3 + 1) = *(aoffset1 + 1);
+
+            aoffset1 += 2;
+            boffset3 += 2;
         }
     }