test-quantize-fns: CI fixes

Fix issues uncovered in CI
 - need to use sizes divisible by 32*8 for loop unrolling
 - use intrinsic header that should work on Mac

Author: unbounded

tests/test-quantize-fns.cpp

@@ -86,15 +86,15 @@ float dot_product_error(quantize_fns_t & qfns, size_t test_size, const float * t
 
 int main(int argc, char * argv[]) {
     bool verbose = false;
-    const size_t test_size = 32 * 100;
+    const size_t test_size = 32 * 128;
 
     std::string arg;
     for (int i = 1; i < argc; i++) {
         arg = argv[i];
 
         if (arg == "-v") {
             verbose = true;
-	} else {
+        } else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             return 1;
         }

tests/test-quantize-perf.cpp

@@ -18,10 +18,10 @@
 #define WARMUP 5
 #define ITERATIONS 10
 
-#define L1_SIZE      32*100
-#define L2_SIZE     32*2000
-#define L3_SIZE    32*20000
-#define MEM_SIZE 32*2000000
+#define L1_SIZE      32*128
+#define L2_SIZE     32*2048
+#define L3_SIZE    32*20480
+#define MEM_SIZE 32*2048000
 
 struct quantize_perf_params {
     std::vector<std::string> include_types;
@@ -36,7 +36,7 @@ struct quantize_perf_params {
 
 #if defined(__x86_64__) || defined(__i386__)
 
-#include <immintrin.h>
+#include <x86intrin.h>
 inline int64_t cpu_cycles() {
 // Rough way to detect new-ish CPUs
 #ifdef __POPCNT__
@@ -71,29 +71,25 @@ void * align_with_offset(void * ptr, int offset) {
 }
 
 void benchmark_function(size_t size, size_t q_size, std::function<size_t(void)> function) {
-
-    size_t bytes_out = 0;
-
     int64_t min_time_us = INT64_MAX;
     int64_t total_time_us = 0;
     int64_t min_time_cycles = INT64_MAX;
     int64_t total_time_cycles = 0;
 
     for (int i = 0; i < WARMUP; i++) {
-        bytes_out |= function();
+        function();
     }
 
 
     for (int i = 0; i < ITERATIONS; i++) {
         const int64_t start_time = ggml_time_us();
         const int64_t start_cycles = cpu_cycles();
 
-        bytes_out |= function();
+        function();
 
         const int64_t end_cycles = cpu_cycles();
         const int64_t end_time = ggml_time_us();
 
-        //printf("    aostne %d\n",  end_cycles - start_cycles);
         total_time_cycles += end_cycles - start_cycles;
         min_time_cycles = std::min(min_time_cycles, end_cycles - start_cycles);
         total_time_us += end_time - start_time;