Fix usage of F16C intrinsics in AVX code (#563)

* Fix usage of F16C intrinsics in AVX code when F16C is not defined

ggml.c

@@ -1122,13 +1122,36 @@ void dequantize_row_q4_1(const void * restrict x, float * restrict y, int k) {
 #define GGML_F16_EPR  8
 
 // F16 arithmetic is not supported by AVX, so we use F32 instead
-// we take advantage of the _mm256_cvt intrinsics to convert F16 <-> F32
 
 #define GGML_F32Cx8             __m256
 #define GGML_F32Cx8_ZERO        _mm256_setzero_ps()
 #define GGML_F32Cx8_SET1(x)     _mm256_set1_ps(x)
+
+#if defined(__F16C__)
+// the  _mm256_cvt intrinsics require F16C
 #define GGML_F32Cx8_LOAD(x)     _mm256_cvtph_ps(_mm_loadu_si128((__m128i *)(x)))
 #define GGML_F32Cx8_STORE(x, y) _mm_storeu_si128((__m128i *)(x), _mm256_cvtps_ph(y, 0))
+#else
+static inline __m256 __avx_f32cx8_load(ggml_fp16_t *x) {
+    float tmp[8];
+
+    for (int i = 0; i < 8; i++)
+        tmp[i] = GGML_FP16_TO_FP32(x[i]);
+
+    return _mm256_loadu_ps(tmp);
+}
+static inline void __avx_f32cx8_store(ggml_fp16_t *x, __m256 y) {
+    float arr[8];
+
+    _mm256_storeu_ps(arr, y);
+
+    for (int i = 0; i < 8; i++)
+        x[i] = GGML_FP16_TO_FP32(arr[i]);
+}
+#define GGML_F32Cx8_LOAD(x)     __avx_f32cx8_load(x)
+#define GGML_F32Cx8_STORE(x, y) __avx_f32cx8_store(x, y)
+#endif
+
 #define GGML_F32Cx8_FMA         GGML_F32x8_FMA
 #define GGML_F32Cx8_ADD         _mm256_add_ps
 #define GGML_F32Cx8_MUL         _mm256_mul_ps