Merge pull request #5408 from ChipKerchner/fixRISCV64GEMVInitializationAndWarnings

Fix bad vector zero initializer and other compiler warnings for RISC-V.

Author: martin-frbg

kernel/riscv64/gemv_n_vector.c

@@ -35,7 +35,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define VSSEV_FLOAT RISCV_RVV(vsse32_v_f32m8)
 #define VFMACCVF_FLOAT RISCV_RVV(vfmacc_vf_f32m8)
 #define VFMUL_VF_FLOAT RISCV_RVV(vfmul_vf_f32m8)
-#define VFILL_ZERO_FLOAT RISCV_RVV(vfsub_vv_f32m8)
+#define VREINTERPRET_FLOAT RISCV_RVV(vreinterpret_v_i32m8_f32m8)
+#define VFILL_INT RISCV_RVV(vmv_v_x_i32m8)
 #else
 #define VSETVL(n) RISCV_RVV(vsetvl_e64m4)(n)
 #define FLOAT_V_T vfloat64m4_t
@@ -45,7 +46,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define VSSEV_FLOAT RISCV_RVV(vsse64_v_f64m4)
 #define VFMACCVF_FLOAT RISCV_RVV(vfmacc_vf_f64m4)
 #define VFMUL_VF_FLOAT RISCV_RVV(vfmul_vf_f64m4)
-#define VFILL_ZERO_FLOAT RISCV_RVV(vfsub_vv_f64m4)
+#define VREINTERPRET_FLOAT RISCV_RVV(vreinterpret_v_i64m4_f64m4)
+#define VFILL_INT RISCV_RVV(vmv_v_x_i64m4)
 #endif
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
@@ -56,7 +58,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
     if(n < 0)  return(0);
     FLOAT *a_ptr = a;
     FLOAT temp[4];
-    FLOAT_V_T va0, va1, vy0, vy1,vy0_temp, vy1_temp , temp_v ,va0_0 , va0_1 , va1_0 ,va1_1 ,va2_0 ,va2_1 ,va3_0 ,va3_1 ;
+    FLOAT_V_T va0, va1, vy0, vy1,vy0_temp, vy1_temp ,va0_0 , va0_1 , va1_0 ,va1_1 ,va2_0 ,va2_1 ,va3_0 ,va3_1 ;
     unsigned int gvl = 0;
     if(inc_y == 1 && inc_x == 1){
         gvl = VSETVL(m);
@@ -66,8 +68,8 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
                 ix = 0;
                 vy0_temp = VLEV_FLOAT(&y[j], gvl);
                 vy1_temp = VLEV_FLOAT(&y[j+gvl], gvl);
-                vy0 = VFILL_ZERO_FLOAT(vy0 , vy0 , gvl);
-                vy1 = VFILL_ZERO_FLOAT(vy1 , vy1 , gvl);
+                vy0 = VREINTERPRET_FLOAT(VFILL_INT(0, gvl));
+                vy1 = VREINTERPRET_FLOAT(VFILL_INT(0, gvl));
                 int i;
 
                 int remainder = n % 4;
@@ -118,7 +120,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
 			a_ptr = a;
 			ix = 0;
 			vy0_temp = VLEV_FLOAT(&y[j], gvl);
-			vy0 = VFILL_ZERO_FLOAT(vy0 , vy0 , gvl);
+			vy0 = VREINTERPRET_FLOAT(VFILL_INT(0, gvl));
 			int i;
 
 			int remainder = n % 4;
@@ -251,4 +253,4 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha, FLOAT *a, BLASLO
         }
     }
     return(0);
-}
+}

kernel/riscv64/omatcopy_cn_vector.c

@@ -51,7 +51,7 @@ int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alpha, FLOAT *a, BLASLONG lda, FLO
 	FLOAT *aptr,*bptr;
 	size_t vl;
 
-	FLOAT_V_T va, vb,va1,vb1;
+	FLOAT_V_T va,va1;
 	if ( rows <= 0 )  return(0);
 	if ( cols <= 0 )  return(0);
 

kernel/riscv64/zamax_vector.c

@@ -81,7 +81,6 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         FLOAT_V_T_M1 v_res;
         v_res = VFMVVF_FLOAT_M1(0, 1);
 
-        MASK_T mask0, mask1;
         BLASLONG stride_x = inc_x * sizeof(FLOAT) * 2;
         gvl = VSETVL(n);
         v_max = VFMVVF_FLOAT(0, gvl);

kernel/riscv64/zamin_vector.c

@@ -83,7 +83,6 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         FLOAT_V_T_M1 v_res;
         v_res = VFMVVF_FLOAT_M1(FLT_MAX, 1);
 
-        MASK_T mask0, mask1;
         BLASLONG stride_x = inc_x * sizeof(FLOAT) * 2;
         gvl = VSETVL(n);
         v_min = VFMVVF_FLOAT(FLT_MAX, gvl);

kernel/riscv64/zasum_vector.c

@@ -76,14 +76,13 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	FLOAT asumf=0.0;
 	if (n <= 0 || inc_x <= 0) return(asumf);
         unsigned int gvl = 0;
-        FLOAT_V_T v0, v1, v_zero,v_sum;
+        FLOAT_V_T v0, v1,v_sum;
         FLOAT_V_T_M1 v_res;
         v_res = VFMVVF_FLOAT_M1(0, 1);
 
         if(inc_x == 1){
                 BLASLONG n2 = n * 2;
                 gvl = VSETVL(n2);
-                v_zero = VFMVVF_FLOAT(0, gvl);
                 if(gvl <= n2/2){
                         v_sum = VFMVVF_FLOAT(0, gvl);
                         for(i=0,j=0; i<n2/(gvl*2); i++){
@@ -108,7 +107,6 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         }else{
                 gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT) * 2;
-                v_zero = VFMVVF_FLOAT(0, gvl);
 
                 BLASLONG inc_xv = inc_x * 2 * gvl;
                 v_sum = VFMVVF_FLOAT(0, gvl);

kernel/riscv64/zgemv_n_vector.c

@@ -55,8 +55,8 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i,
         BLASLONG i = 0, j = 0, k = 0;
         BLASLONG ix = 0, iy = 0;
         FLOAT *a_ptr = a;
-        FLOAT temp_r = 0.0, temp_i = 0.0, temp_r1, temp_i1, temp_r2, temp_i2, temp_r3, temp_i3, temp_rr[4], temp_ii[4];
-        FLOAT_V_T va0, va1, vy0, vy1, vy0_new, vy1_new, va2, va3, va4, va5, va6, va7, temp_iv, temp_rv, x_v0, x_v1, temp_v1, temp_v2, temp_v3, temp_v4;
+        FLOAT temp_r = 0.0, temp_i = 0.0, temp_rr[4], temp_ii[4];
+        FLOAT_V_T va0, va1, vy0, vy1, vy0_new, vy1_new, va2, va3, va4, va5, va6, va7, temp_iv, temp_rv, x_v0, x_v1;
         unsigned int gvl = 0;
         BLASLONG stride_a = sizeof(FLOAT) * 2;
         BLASLONG stride_y = inc_y * sizeof(FLOAT) * 2;

kernel/riscv64/zsum_vector.c

@@ -71,14 +71,13 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	FLOAT asumf=0.0;
 	if (n <= 0 || inc_x <= 0) return(asumf);
         unsigned int gvl = 0;
-        FLOAT_V_T v0, v1, v_zero,v_sum;
+        FLOAT_V_T v0, v1,v_sum;
         FLOAT_V_T_M1 v_res;
         v_res = VFMVVF_FLOAT_M1(0, 1);
 
         if(inc_x == 1){
                 BLASLONG n2 = n * 2;
                 gvl = VSETVL(n2);
-                v_zero = VFMVVF_FLOAT(0, gvl);
                 if(gvl <= n2/2){
                         v_sum = VFMVVF_FLOAT(0, gvl);
                         for(i=0,j=0; i<n2/(gvl*2); i++){
@@ -100,7 +99,6 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
         }else{
                 gvl = VSETVL(n);
                 unsigned int stride_x = inc_x * sizeof(FLOAT) * 2;
-                v_zero = VFMVVF_FLOAT(0, gvl);
 
                 BLASLONG inc_xv = inc_x * 2 * gvl;
                 v_sum = VFMVVF_FLOAT(0, gvl);