From a4abf7828efe2bd40321366505b75deb2e200cc7 Mon Sep 17 00:00:00 2001
From: Chip Kerchner <ckerchner@tenstorrent.com>
Date: Thu, 18 Sep 2025 21:30:22 +0000
Subject: [PATCH] Fix _Float16 casting issue and reduce LMUL for certain vector
 instruction from m2 to m1.

---
 kernel/riscv64/shgemm_kernel_16x8_zvl256b.c |  92 +++----
 kernel/riscv64/shgemm_kernel_8x8_zvl128b.c  | 258 ++++++++++----------
 2 files changed, 175 insertions(+), 175 deletions(-)

diff --git a/kernel/riscv64/shgemm_kernel_16x8_zvl256b.c b/kernel/riscv64/shgemm_kernel_16x8_zvl256b.c
index fb98f564c3..746e07d55f 100644
--- a/kernel/riscv64/shgemm_kernel_16x8_zvl256b.c
+++ b/kernel/riscv64/shgemm_kernel_16x8_zvl256b.c
@@ -295,22 +295,22 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
             
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
-                result2+=(float)(A[ai+0]*B[bi+1]);
-                result3+=(float)(A[ai+1]*B[bi+1]);
-                result4+=(float)(A[ai+0]*B[bi+2]);
-                result5+=(float)(A[ai+1]*B[bi+2]);
-                result6+=(float)(A[ai+0]*B[bi+3]);
-                result7+=(float)(A[ai+1]*B[bi+3]);
-                result8+=(float)(A[ai+0]*B[bi+4]);
-                result9+=(float)(A[ai+1]*B[bi+4]);
-                result10+=(float)(A[ai+0]*B[bi+5]);
-                result11+=(float)(A[ai+1]*B[bi+5]);
-                result12+=(float)(A[ai+0]*B[bi+6]);
-                result13+=(float)(A[ai+1]*B[bi+6]);
-                result14+=(float)(A[ai+0]*B[bi+7]);
-                result15+=(float)(A[ai+1]*B[bi+7]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result3+=(float)(A[ai+1])*(float)(B[bi+1]);
+                result4+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result5+=(float)(A[ai+1])*(float)(B[bi+2]);
+                result6+=(float)(A[ai+0])*(float)(B[bi+3]);
+                result7+=(float)(A[ai+1])*(float)(B[bi+3]);
+                result8+=(float)(A[ai+0])*(float)(B[bi+4]);
+                result9+=(float)(A[ai+1])*(float)(B[bi+4]);
+                result10+=(float)(A[ai+0])*(float)(B[bi+5]);
+                result11+=(float)(A[ai+1])*(float)(B[bi+5]);
+                result12+=(float)(A[ai+0])*(float)(B[bi+6]);
+                result13+=(float)(A[ai+1])*(float)(B[bi+6]);
+                result14+=(float)(A[ai+0])*(float)(B[bi+7]);
+                result15+=(float)(A[ai+1])*(float)(B[bi+7]);
                 ai+=2;
                 bi+=8;
             }
@@ -353,14 +353,14 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+0]*B[bi+1]);
-                result2+=(float)(A[ai+0]*B[bi+2]);
-                result3+=(float)(A[ai+0]*B[bi+3]);
-                result4+=(float)(A[ai+0]*B[bi+4]);
-                result5+=(float)(A[ai+0]*B[bi+5]);
-                result6+=(float)(A[ai+0]*B[bi+6]);
-                result7+=(float)(A[ai+0]*B[bi+7]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result3+=(float)(A[ai+0])*(float)(B[bi+3]);
+                result4+=(float)(A[ai+0])*(float)(B[bi+4]);
+                result5+=(float)(A[ai+0])*(float)(B[bi+5]);
+                result6+=(float)(A[ai+0])*(float)(B[bi+6]);
+                result7+=(float)(A[ai+0])*(float)(B[bi+7]);
                 ai+=1;
                 bi+=8;
             }
@@ -569,14 +569,14 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
-                result2+=(float)(A[ai+0]*B[bi+1]);
-                result3+=(float)(A[ai+1]*B[bi+1]);
-                result4+=(float)(A[ai+0]*B[bi+2]);
-                result5+=(float)(A[ai+1]*B[bi+2]);
-                result6+=(float)(A[ai+0]*B[bi+3]);
-                result7+=(float)(A[ai+1]*B[bi+3]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result3+=(float)(A[ai+1])*(float)(B[bi+1]);
+                result4+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result5+=(float)(A[ai+1])*(float)(B[bi+2]);
+                result6+=(float)(A[ai+0])*(float)(B[bi+3]);
+                result7+=(float)(A[ai+1])*(float)(B[bi+3]);
                 ai+=2;
                 bi+=4;
             }
@@ -607,10 +607,10 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+0]*B[bi+1]);
-                result2+=(float)(A[ai+0]*B[bi+2]);
-                result3+=(float)(A[ai+0]*B[bi+3]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result3+=(float)(A[ai+0])*(float)(B[bi+3]);
                 ai+=1;
                 bi+=4;
             }
@@ -770,10 +770,10 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
-                result2+=(float)(A[ai+0]*B[bi+1]);
-                result3+=(float)(A[ai+1]*B[bi+1]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result3+=(float)(A[ai+1])*(float)(B[bi+1]);
                 ai+=2;
                 bi+=2;
             }
@@ -797,8 +797,8 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+0]*B[bi+1]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+0])*(float)(B[bi+1]);
                 ai+=1;
                 bi+=2;
             }
@@ -930,8 +930,8 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
                 ai+=2;
                 bi+=1;
             }
@@ -953,7 +953,7 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
                 ai+=1;
                 bi+=1;
             }
@@ -966,4 +966,4 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
         n_top += 1;
     }
     return 0;
-}
\ No newline at end of file
+}
diff --git a/kernel/riscv64/shgemm_kernel_8x8_zvl128b.c b/kernel/riscv64/shgemm_kernel_8x8_zvl128b.c
index 74e8c33181..3197f7c005 100644
--- a/kernel/riscv64/shgemm_kernel_8x8_zvl128b.c
+++ b/kernel/riscv64/shgemm_kernel_8x8_zvl128b.c
@@ -115,17 +115,17 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             _Float16 B7 = B[bi+7];
             bi += 8;
 
-            vfloat16m1_t A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+            vfloat16mf2_t A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
             ai += 4;
 
-            vfloat32m2_t result0 = __riscv_vfwmul_vf_f32m2( A0, B0, gvl);
-            vfloat32m2_t result1 = __riscv_vfwmul_vf_f32m2( A0, B1, gvl);
-            vfloat32m2_t result2 = __riscv_vfwmul_vf_f32m2( A0, B2, gvl);
-            vfloat32m2_t result3 = __riscv_vfwmul_vf_f32m2( A0, B3, gvl);
-            vfloat32m2_t result4 = __riscv_vfwmul_vf_f32m2( A0, B4, gvl);
-            vfloat32m2_t result5 = __riscv_vfwmul_vf_f32m2( A0, B5, gvl);
-            vfloat32m2_t result6 = __riscv_vfwmul_vf_f32m2( A0, B6, gvl);
-            vfloat32m2_t result7 = __riscv_vfwmul_vf_f32m2( A0, B7, gvl);
+            vfloat32m1_t result0 = __riscv_vfwmul_vf_f32m1( A0, B0, gvl);
+            vfloat32m1_t result1 = __riscv_vfwmul_vf_f32m1( A0, B1, gvl);
+            vfloat32m1_t result2 = __riscv_vfwmul_vf_f32m1( A0, B2, gvl);
+            vfloat32m1_t result3 = __riscv_vfwmul_vf_f32m1( A0, B3, gvl);
+            vfloat32m1_t result4 = __riscv_vfwmul_vf_f32m1( A0, B4, gvl);
+            vfloat32m1_t result5 = __riscv_vfwmul_vf_f32m1( A0, B5, gvl);
+            vfloat32m1_t result6 = __riscv_vfwmul_vf_f32m1( A0, B6, gvl);
+            vfloat32m1_t result7 = __riscv_vfwmul_vf_f32m1( A0, B7, gvl);
 
             for(BLASLONG k=1; k < K; ++k) {
                 B0 = B[bi+0];
@@ -138,55 +138,55 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
                 B7 = B[bi+7];
                 bi += 8;
 
-                A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+                A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
                 ai += 4;
 
-                result0 = __riscv_vfwmacc_vf_f32m2(result0, B0, A0, gvl);
-                result1 = __riscv_vfwmacc_vf_f32m2(result1, B1, A0, gvl);
-                result2 = __riscv_vfwmacc_vf_f32m2(result2, B2, A0, gvl);
-                result3 = __riscv_vfwmacc_vf_f32m2(result3, B3, A0, gvl);
-                result4 = __riscv_vfwmacc_vf_f32m2(result4, B4, A0, gvl);
-                result5 = __riscv_vfwmacc_vf_f32m2(result5, B5, A0, gvl);
-                result6 = __riscv_vfwmacc_vf_f32m2(result6, B6, A0, gvl);
-                result7 = __riscv_vfwmacc_vf_f32m2(result7, B7, A0, gvl);
+                result0 = __riscv_vfwmacc_vf_f32m1(result0, B0, A0, gvl);
+                result1 = __riscv_vfwmacc_vf_f32m1(result1, B1, A0, gvl);
+                result2 = __riscv_vfwmacc_vf_f32m1(result2, B2, A0, gvl);
+                result3 = __riscv_vfwmacc_vf_f32m1(result3, B3, A0, gvl);
+                result4 = __riscv_vfwmacc_vf_f32m1(result4, B4, A0, gvl);
+                result5 = __riscv_vfwmacc_vf_f32m1(result5, B5, A0, gvl);
+                result6 = __riscv_vfwmacc_vf_f32m1(result6, B6, A0, gvl);
+                result7 = __riscv_vfwmacc_vf_f32m1(result7, B7, A0, gvl);
             }
 
             BLASLONG ci = n_top * ldc + m_top;
 
-            vfloat32m2_t c0 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c0 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c1 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c1 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c2 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c2 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c3 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c3 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c4 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c4 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c5 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c5 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c6 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c6 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c7 = __riscv_vle32_v_f32m2(&C[ci], gvl);
-            c0 = __riscv_vfmacc_vf_f32m2(c0, alpha, result0, gvl);
-            c1 = __riscv_vfmacc_vf_f32m2(c1, alpha, result1, gvl);
-            c2 = __riscv_vfmacc_vf_f32m2(c2, alpha, result2, gvl);
-            c3 = __riscv_vfmacc_vf_f32m2(c3, alpha, result3, gvl);
-            c4 = __riscv_vfmacc_vf_f32m2(c4, alpha, result4, gvl);
-            c5 = __riscv_vfmacc_vf_f32m2(c5, alpha, result5, gvl);
-            c6 = __riscv_vfmacc_vf_f32m2(c6, alpha, result6, gvl);
-            c7 = __riscv_vfmacc_vf_f32m2(c7, alpha, result7, gvl);
+            vfloat32m1_t c7 = __riscv_vle32_v_f32m1(&C[ci], gvl);
+            c0 = __riscv_vfmacc_vf_f32m1(c0, alpha, result0, gvl);
+            c1 = __riscv_vfmacc_vf_f32m1(c1, alpha, result1, gvl);
+            c2 = __riscv_vfmacc_vf_f32m1(c2, alpha, result2, gvl);
+            c3 = __riscv_vfmacc_vf_f32m1(c3, alpha, result3, gvl);
+            c4 = __riscv_vfmacc_vf_f32m1(c4, alpha, result4, gvl);
+            c5 = __riscv_vfmacc_vf_f32m1(c5, alpha, result5, gvl);
+            c6 = __riscv_vfmacc_vf_f32m1(c6, alpha, result6, gvl);
+            c7 = __riscv_vfmacc_vf_f32m1(c7, alpha, result7, gvl);
 
             ci= n_top * ldc + m_top;
 
-            __riscv_vse32_v_f32m2(&C[ci], c0, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c1, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c2, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c3, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c4, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c5, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c6, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c7, gvl); 
+            __riscv_vse32_v_f32m1(&C[ci], c0, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c1, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c2, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c3, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c4, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c5, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c6, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c7, gvl); 
             
             m_top += 4;
         }
@@ -215,22 +215,22 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             float result15 = 0;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
-                result2+=(float)(A[ai+0]*B[bi+1]);
-                result3+=(float)(A[ai+1]*B[bi+1]);
-                result4+=(float)(A[ai+0]*B[bi+2]);
-                result5+=(float)(A[ai+1]*B[bi+2]);
-                result6+=(float)(A[ai+0]*B[bi+3]);
-                result7+=(float)(A[ai+1]*B[bi+3]);
-                result8+=(float)(A[ai+0]*B[bi+4]);
-                result9+=(float)(A[ai+1]*B[bi+4]);
-                result10+=(float)(A[ai+0]*B[bi+5]);
-                result11+=(float)(A[ai+1]*B[bi+5]);
-                result12+=(float)(A[ai+0]*B[bi+6]);
-                result13+=(float)(A[ai+1]*B[bi+6]);
-                result14+=(float)(A[ai+0]*B[bi+7]);
-                result15+=(float)(A[ai+1]*B[bi+7]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result3+=(float)(A[ai+1])*(float)(B[bi+1]);
+                result4+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result5+=(float)(A[ai+1])*(float)(B[bi+2]);
+                result6+=(float)(A[ai+0])*(float)(B[bi+3]);
+                result7+=(float)(A[ai+1])*(float)(B[bi+3]);
+                result8+=(float)(A[ai+0])*(float)(B[bi+4]);
+                result9+=(float)(A[ai+1])*(float)(B[bi+4]);
+                result10+=(float)(A[ai+0])*(float)(B[bi+5]);
+                result11+=(float)(A[ai+1])*(float)(B[bi+5]);
+                result12+=(float)(A[ai+0])*(float)(B[bi+6]);
+                result13+=(float)(A[ai+1])*(float)(B[bi+6]);
+                result14+=(float)(A[ai+0])*(float)(B[bi+7]);
+                result15+=(float)(A[ai+1])*(float)(B[bi+7]);
                 ai+=2;
                 bi+=8;
             }
@@ -273,14 +273,14 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+0]*B[bi+1]);
-                result2+=(float)(A[ai+0]*B[bi+2]);
-                result3+=(float)(A[ai+0]*B[bi+3]);
-                result4+=(float)(A[ai+0]*B[bi+4]);
-                result5+=(float)(A[ai+0]*B[bi+5]);
-                result6+=(float)(A[ai+0]*B[bi+6]);
-                result7+=(float)(A[ai+0]*B[bi+7]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result3+=(float)(A[ai+0])*(float)(B[bi+3]);
+                result4+=(float)(A[ai+0])*(float)(B[bi+4]);
+                result5+=(float)(A[ai+0])*(float)(B[bi+5]);
+                result6+=(float)(A[ai+0])*(float)(B[bi+6]);
+                result7+=(float)(A[ai+0])*(float)(B[bi+7]);
                 ai+=1;
                 bi+=8;
             }
@@ -372,13 +372,13 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             _Float16 B3 = B[bi+3];
             bi += 4;
 
-            vfloat16m1_t A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+            vfloat16mf2_t A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
             ai += 4;
 
-            vfloat32m2_t result0 = __riscv_vfwmul_vf_f32m2( A0, B0, gvl);
-            vfloat32m2_t result1 = __riscv_vfwmul_vf_f32m2( A0, B1, gvl);
-            vfloat32m2_t result2 = __riscv_vfwmul_vf_f32m2( A0, B2, gvl);
-            vfloat32m2_t result3 = __riscv_vfwmul_vf_f32m2( A0, B3, gvl);
+            vfloat32m1_t result0 = __riscv_vfwmul_vf_f32m1( A0, B0, gvl);
+            vfloat32m1_t result1 = __riscv_vfwmul_vf_f32m1( A0, B1, gvl);
+            vfloat32m1_t result2 = __riscv_vfwmul_vf_f32m1( A0, B2, gvl);
+            vfloat32m1_t result3 = __riscv_vfwmul_vf_f32m1( A0, B3, gvl);
 
             for(BLASLONG k=1; k < K; ++k) {
                 B0 = B[bi+0];
@@ -387,35 +387,35 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
                 B3 = B[bi+3];
                 bi += 4;
 
-                A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+                A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
                 ai += 4;
 
-                result0 = __riscv_vfwmacc_vf_f32m2(result0, B0, A0, gvl);
-                result1 = __riscv_vfwmacc_vf_f32m2(result1, B1, A0, gvl);
-                result2 = __riscv_vfwmacc_vf_f32m2(result2, B2, A0, gvl);
-                result3 = __riscv_vfwmacc_vf_f32m2(result3, B3, A0, gvl);
+                result0 = __riscv_vfwmacc_vf_f32m1(result0, B0, A0, gvl);
+                result1 = __riscv_vfwmacc_vf_f32m1(result1, B1, A0, gvl);
+                result2 = __riscv_vfwmacc_vf_f32m1(result2, B2, A0, gvl);
+                result3 = __riscv_vfwmacc_vf_f32m1(result3, B3, A0, gvl);
             }
 
             BLASLONG ci = n_top * ldc + m_top;
 
-            vfloat32m2_t c0 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c0 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c1 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c1 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c2 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c2 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c3 = __riscv_vle32_v_f32m2(&C[ci], gvl);
-            c0 = __riscv_vfmacc_vf_f32m2(c0, alpha, result0, gvl);
-            c1 = __riscv_vfmacc_vf_f32m2(c1, alpha, result1, gvl);
-            c2 = __riscv_vfmacc_vf_f32m2(c2, alpha, result2, gvl);
-            c3 = __riscv_vfmacc_vf_f32m2(c3, alpha, result3, gvl);
+            vfloat32m1_t c3 = __riscv_vle32_v_f32m1(&C[ci], gvl);
+            c0 = __riscv_vfmacc_vf_f32m1(c0, alpha, result0, gvl);
+            c1 = __riscv_vfmacc_vf_f32m1(c1, alpha, result1, gvl);
+            c2 = __riscv_vfmacc_vf_f32m1(c2, alpha, result2, gvl);
+            c3 = __riscv_vfmacc_vf_f32m1(c3, alpha, result3, gvl);
 
             ci= n_top * ldc + m_top;
 
-            __riscv_vse32_v_f32m2(&C[ci], c0, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c1, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c2, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c3, gvl); 
+            __riscv_vse32_v_f32m1(&C[ci], c0, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c1, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c2, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c3, gvl); 
             
             m_top += 4;
         }
@@ -436,14 +436,14 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             float result7 = 0;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
-                result2+=(float)(A[ai+0]*B[bi+1]);
-                result3+=(float)(A[ai+1]*B[bi+1]);
-                result4+=(float)(A[ai+0]*B[bi+2]);
-                result5+=(float)(A[ai+1]*B[bi+2]);
-                result6+=(float)(A[ai+0]*B[bi+3]);
-                result7+=(float)(A[ai+1]*B[bi+3]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result3+=(float)(A[ai+1])*(float)(B[bi+1]);
+                result4+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result5+=(float)(A[ai+1])*(float)(B[bi+2]);
+                result6+=(float)(A[ai+0])*(float)(B[bi+3]);
+                result7+=(float)(A[ai+1])*(float)(B[bi+3]);
                 ai+=2;
                 bi+=4;
             }
@@ -474,10 +474,10 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+0]*B[bi+1]);
-                result2+=(float)(A[ai+0]*B[bi+2]);
-                result3+=(float)(A[ai+0]*B[bi+3]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+2]);
+                result3+=(float)(A[ai+0])*(float)(B[bi+3]);
                 ai+=1;
                 bi+=4;
             }
@@ -551,36 +551,36 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             _Float16 B1 = B[bi+1];
             bi += 2;
 
-            vfloat16m1_t A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+            vfloat16mf2_t A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
             ai += 4;
 
-            vfloat32m2_t result0 = __riscv_vfwmul_vf_f32m2( A0, B0, gvl);
-            vfloat32m2_t result1 = __riscv_vfwmul_vf_f32m2( A0, B1, gvl);
+            vfloat32m1_t result0 = __riscv_vfwmul_vf_f32m1( A0, B0, gvl);
+            vfloat32m1_t result1 = __riscv_vfwmul_vf_f32m1( A0, B1, gvl);
 
             for(BLASLONG k=1; k < K; ++k) {
                 B0 = B[bi+0];
                 B1 = B[bi+1];
                 bi += 2;
 
-                A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+                A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
                 ai += 4;
 
-                result0 = __riscv_vfwmacc_vf_f32m2(result0, B0, A0, gvl);
-                result1 = __riscv_vfwmacc_vf_f32m2(result1, B1, A0, gvl);
+                result0 = __riscv_vfwmacc_vf_f32m1(result0, B0, A0, gvl);
+                result1 = __riscv_vfwmacc_vf_f32m1(result1, B1, A0, gvl);
             }
 
             BLASLONG ci = n_top * ldc + m_top;
 
-            vfloat32m2_t c0 = __riscv_vle32_v_f32m2(&C[ci], gvl);
+            vfloat32m1_t c0 = __riscv_vle32_v_f32m1(&C[ci], gvl);
             ci += ldc - gvl * 0;
-            vfloat32m2_t c1 = __riscv_vle32_v_f32m2(&C[ci], gvl);
-            c0 = __riscv_vfmacc_vf_f32m2(c0, alpha, result0, gvl);
-            c1 = __riscv_vfmacc_vf_f32m2(c1, alpha, result1, gvl);
+            vfloat32m1_t c1 = __riscv_vle32_v_f32m1(&C[ci], gvl);
+            c0 = __riscv_vfmacc_vf_f32m1(c0, alpha, result0, gvl);
+            c1 = __riscv_vfmacc_vf_f32m1(c1, alpha, result1, gvl);
 
             ci= n_top * ldc + m_top;
 
-            __riscv_vse32_v_f32m2(&C[ci], c0, gvl); ci += ldc - gvl * 0;
-            __riscv_vse32_v_f32m2(&C[ci], c1, gvl); 
+            __riscv_vse32_v_f32m1(&C[ci], c0, gvl); ci += ldc - gvl * 0;
+            __riscv_vse32_v_f32m1(&C[ci], c1, gvl); 
             
             m_top += 4;
         }
@@ -597,10 +597,10 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             float result3 = 0;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
-                result2+=(float)(A[ai+0]*B[bi+1]);
-                result3+=(float)(A[ai+1]*B[bi+1]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
+                result2+=(float)(A[ai+0])*(float)(B[bi+1]);
+                result3+=(float)(A[ai+1])*(float)(B[bi+1]);
                 ai+=2;
                 bi+=2;
             }
@@ -624,8 +624,8 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+0]*B[bi+1]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+0])*(float)(B[bi+1]);
                 ai+=1;
                 bi+=2;
             }
@@ -689,29 +689,29 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             _Float16 B0 = B[bi+0];
             bi += 1;
 
-            vfloat16m1_t A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+            vfloat16mf2_t A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
             ai += 4;
 
-            vfloat32m2_t result0 = __riscv_vfwmul_vf_f32m2( A0, B0, gvl);
+            vfloat32m1_t result0 = __riscv_vfwmul_vf_f32m1( A0, B0, gvl);
 
             for(BLASLONG k=1; k < K; ++k) {
                 B0 = B[bi+0];
                 bi += 1;
 
-                A0 = __riscv_vle16_v_f16m1(&A[ai + 0 * gvl], gvl);
+                A0 = __riscv_vle16_v_f16mf2(&A[ai + 0 * gvl], gvl);
                 ai += 4;
 
-                result0 = __riscv_vfwmacc_vf_f32m2(result0, B0, A0, gvl);
+                result0 = __riscv_vfwmacc_vf_f32m1(result0, B0, A0, gvl);
             }
 
             BLASLONG ci = n_top * ldc + m_top;
 
-            vfloat32m2_t c0 = __riscv_vle32_v_f32m2(&C[ci], gvl);
-            c0 = __riscv_vfmacc_vf_f32m2(c0, alpha, result0, gvl);
+            vfloat32m1_t c0 = __riscv_vle32_v_f32m1(&C[ci], gvl);
+            c0 = __riscv_vfmacc_vf_f32m1(c0, alpha, result0, gvl);
 
             ci= n_top * ldc + m_top;
 
-            __riscv_vse32_v_f32m2(&C[ci], c0, gvl);            
+            __riscv_vse32_v_f32m1(&C[ci], c0, gvl);
             m_top += 4;
         }
 
@@ -725,8 +725,8 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             float result1 = 0;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
-                result1+=(float)(A[ai+1]*B[bi+0]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
+                result1+=(float)(A[ai+1])*(float)(B[bi+0]);
                 ai+=2;
                 bi+=1;
             }
@@ -748,7 +748,7 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
             BLASLONG bi = n_top * K;
 
             for(BLASLONG k=0; k<K; k++) {
-                result0+=(float)(A[ai+0]*B[bi+0]);
+                result0+=(float)(A[ai+0])*(float)(B[bi+0]);
                 ai+=1;
                 bi+=1;
             }
@@ -764,4 +764,4 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, IFLOAT *A, IFLOAT *B,
 
     return 0;
 
-}
\ No newline at end of file
+}