Use float in vector dot products

SIMD implementation is limited to floats anyway and the return value
is also a float, a float-double-float roundtrip is useless.

Author: sw

ggml.c

@@ -1336,7 +1336,7 @@ inline static void ggml_vec_mul_f32 (const int n, float * z, const float * x, co
 inline static void ggml_vec_div_f32 (const int n, float * z, const float * x, const float * y) { for (int i = 0; i < n; ++i) z[i]  = x[i]/y[i];   }
 
 inline static void ggml_vec_dot_f32(const int n, float * restrict s, const float * restrict x, const float * restrict y) {
-    ggml_float sumf = 0.0;
+    float sumf = 0.0f;
 
 #ifdef GGML_SIMD
     const int np = (n & ~(GGML_F32_STEP - 1));
@@ -1360,12 +1360,12 @@ inline static void ggml_vec_dot_f32(const int n, float * restrict s, const float
 
     // leftovers
     for (int i = np; i < n; ++i) {
-        sumf += (ggml_float)(x[i]*y[i]);
+        sumf += x[i]*y[i];
     }
 #else
     // scalar
     for (int i = 0; i < n; ++i) {
-        sumf += (ggml_float)(x[i]*y[i]);
+        sumf += x[i]*y[i];
     }
 #endif
 
@@ -1414,7 +1414,7 @@ static inline __m512 dot_q4_0_oneblock_avx512(
 #endif
 
 inline static void ggml_vec_dot_f16(const int n, float * restrict s, ggml_fp16_t * restrict x, ggml_fp16_t * restrict y) {
-    ggml_float sumf = 0.0;
+    float sumf = 0.0f;
 
 #if defined(GGML_SIMD)
     const int np = (n & ~(GGML_F16_STEP - 1));
@@ -1438,11 +1438,11 @@ inline static void ggml_vec_dot_f16(const int n, float * restrict s, ggml_fp16_t
 
     // leftovers
     for (int i = np; i < n; ++i) {
-        sumf += (ggml_float)(GGML_FP16_TO_FP32(x[i])*GGML_FP16_TO_FP32(y[i]));
+        sumf += GGML_FP16_TO_FP32(x[i])*GGML_FP16_TO_FP32(y[i]);
     }
 #else
     for (int i = 0; i < n; ++i) {
-        sumf += (ggml_float)(GGML_FP16_TO_FP32(x[i])*GGML_FP16_TO_FP32(y[i]));
+        sumf += GGML_FP16_TO_FP32(x[i])*GGML_FP16_TO_FP32(y[i]);
     }
 #endif
 
@@ -1899,7 +1899,7 @@ inline static void ggml_vec_dot_q4_1(const int n, float * restrict s, const void
 // compute GGML_VEC_DOT_UNROLL dot products at once
 // xs - x row stride in bytes
 inline static void ggml_vec_dot_f16_unroll(const int n, const int xs, float * restrict s, void * restrict xv, ggml_fp16_t * restrict y) {
-    ggml_float sumf[GGML_VEC_DOT_UNROLL] = { 0.0 };
+    float sumf[GGML_VEC_DOT_UNROLL] = { 0.0f };
 
     ggml_fp16_t * restrict x[GGML_VEC_DOT_UNROLL];
 
@@ -1935,13 +1935,13 @@ inline static void ggml_vec_dot_f16_unroll(const int n, const int xs, float * re
     // leftovers
     for (int i = np; i < n; ++i) {
         for (int j = 0; j < GGML_VEC_DOT_UNROLL; ++j) {
-            sumf[j] += (ggml_float)(GGML_FP16_TO_FP32(x[j][i])*GGML_FP16_TO_FP32(y[i]));
+            sumf[j] += GGML_FP16_TO_FP32(x[j][i])*GGML_FP16_TO_FP32(y[i]);
         }
     }
 #else
     for (int i = 0; i < n; ++i) {
         for (int j = 0; j < GGML_VEC_DOT_UNROLL; ++j) {
-            sumf[j] += (ggml_float)(GGML_FP16_TO_FP32(x[j][i])*GGML_FP16_TO_FP32(y[i]));
+            sumf[j] += GGML_FP16_TO_FP32(x[j][i])*GGML_FP16_TO_FP32(y[i]);
         }
     }
 #endif
@@ -2663,7 +2663,7 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
                 const float f = table_f32_f16[i] = GGML_COMPUTE_FP16_TO_FP32(ii);
                 table_gelu_f16[i] = GGML_FP32_TO_FP16(ggml_gelu_f32(f));
                 table_silu_f16[i] = GGML_FP32_TO_FP16(ggml_silu_f32(f));
-                table_exp_f16[i]  = GGML_FP32_TO_FP16(exp(f));
+                table_exp_f16[i]  = GGML_FP32_TO_FP16(expf(f));
             }
 
             const uint64_t t_end = ggml_time_us(); UNUSED(t_end);