Minor, plus rebase on master

Author: Kawrakow

ggml.c

@@ -2000,7 +2000,7 @@ static const quantize_fns_t quantize_fns[GGML_TYPE_COUNT] = {
     [GGML_TYPE_Q4_3] = {
         .dequantize_row_q         = dequantize_row_q4_3,
         .quantize_row_q           = quantize_row_q4_3,
-        .quantize_row_q_reference = (quantize_row_q_t) quantize_row_q4_3_reference, // TODO: RMSE optimization
+        .quantize_row_q_reference = (quantize_row_q_t) quantize_row_q4_3_reference,
         .quantize_row_q_dot       = quantize_row_q8_0,
         .vec_dot_q                = ggml_vec_dot_q4_3_q8_0,
     },

tests/test-quantize.c

@@ -4,6 +4,9 @@
 #include <math.h>
 
 int main(void) {
+
+// Sorry, but I have to disable these for RMSE-optimized quantization
+#ifdef GGML_NO_RMSE
     #define QK 32
     float src[QK];
     uint8_t dst[24];
@@ -13,8 +16,6 @@ int main(void) {
         src[i] = (float)(i + 1);
     }
 
-// Sorry, but I have to disable these for RMSE-optimized quantization
-#ifdef GGML_NO_RMSE
     size_t size = ggml_quantize_q4_0(src, dst, QK, QK, hist);
     assert(size == 20);
     float max_result = ((float *)dst)[0];