Vulkan: revert the order of the index calculation and bound check in conv_2d shader

Author: relent95

ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp

@@ -229,16 +229,14 @@ void main() {
             uint32_t B_ly    = r_offset + Ar;
             uint32_t B_lx    = Ac;
             uint32_t K_idx   = B_idx_K * BS_K + B_ly; /* Global K_idx (row index of A)*/
-            float    val;
-            if (K_idx >= K || CRS_idx_a >= CRS) {
-                val = 0.0;
-            } else {
 #ifdef TRANSPOSE
-                uint32_t knl_idx = min(KW_idx_a + KH_idx_a * p.nb01 + K_idx * p.nb02 + Cin_idx_a * p.nb03, K * CRS - 1);
+            uint32_t knl_idx = min(KW_idx_a + KH_idx_a * p.nb01 + K_idx * p.nb02 + Cin_idx_a * p.nb03, K * CRS - 1);
 #else
-                uint32_t knl_idx = min(KW_idx_a + KH_idx_a * p.nb01 + Cin_idx_a * p.nb02 + K_idx * p.nb03, K * CRS - 1);
+            uint32_t knl_idx = min(KW_idx_a + KH_idx_a * p.nb01 + Cin_idx_a * p.nb02 + K_idx * p.nb03, K * CRS - 1);
 #endif
-                val = knl_data[knl_idx];
+            float    val     = knl_data[knl_idx];
+            if (K_idx >= K || CRS_idx_a >= CRS) {
+                val = 0.0;
             }
             Ash[B_ly * Ash_stride + B_lx] = SHMEM_TYPE(val);
         }
@@ -286,18 +284,16 @@ void main() {
             uint32_t H_idx = OH_idx * p.s1 + KH_idx_b * p.d1 - p.p1;
             uint32_t W_idx = OW_idx * p.s0 + KW_idx_b * p.d0 - p.p0;
 #endif
-            float val;
+            uint32_t src_idx =
+                min(max(W_idx + H_idx * p.nb11 + Cin_idx_b * p.nb12 + N_idx * p.nb13, 0), p.Cin * p.N * p.W * p.H - 1);
+            float val = src_data[src_idx];
             if (CRS_idx_b >= CRS || NPQ_idx >= NPQ
                 || int32_t(H_idx) < 0 || H_idx >= p.H || int32_t(W_idx) < 0 || W_idx >= p.W
 #ifdef TRANSPOSE
                 || (H_idx_x_s1 - H_idx * p.s1 != 0) || (W_idx_x_s0 - W_idx * p.s0 != 0)
 #endif
                 ) {
                 val = 0.0;
-            } else {
-                uint32_t src_idx =
-                    min(max(W_idx + H_idx * p.nb11 + Cin_idx_b * p.nb12 + N_idx * p.nb13, 0), p.Cin * p.N * p.W * p.H - 1);
-                val = src_data[src_idx];
             }
             Bsh[B_ly * Bsh_stride + B_lx] = SHMEM_TYPE(val);
         }