[graphics] reduce the size of fr3 files (#1175)

* first pass

* first pass at shrinking fr3s

* only need to load vertices once

* avx2 detect and switch

* fix build

* another ifx'

* one more

* fix the sky and stupid math bug in size check

common/util/os.cpp

@@ -1,5 +1,7 @@
 #include "os.h"
 
+#include "common/common_types.h"
+
 #ifdef __linux__
 
 #include <sys/resource.h>
@@ -14,4 +16,72 @@ size_t get_peak_rss() {
 size_t get_peak_rss() {
   return 0;
 }
-#endif
+#endif
+
+#ifdef _WIN32
+// windows has a __cpuid
+#include <intrin.h>
+#else
+// using int to be compatible with msvc's intrinsic
+void __cpuidex(int result[4], int eax, int ecx) {
+  asm("cpuid\n\t"
+      : "=a"(result[0]), "=b"(result[1]), "=c"(result[2]), "=d"(result[3])
+      : "0"(eax), "2"(ecx));
+}
+#endif
+
+CpuInfo gCpuInfo;
+
+void setup_cpu_info() {
+  if (gCpuInfo.initialized) {
+    return;
+  }
+
+  // as a test, get the brand and model
+  for (u32 i = 0x80000002; i <= 0x80000004; i++) {
+    int result[4];
+    __cpuidex(result, i, 0);
+    for (auto reg : result) {
+      for (int c = 0; c < 4; c++) {
+        gCpuInfo.model.push_back(reg);
+        reg >>= 8;
+      }
+    }
+  }
+
+  {
+    int result[4];
+    __cpuidex(result, 0, 0);
+    for (auto r : {1, 3, 2}) {
+      for (int c = 0; c < 4; c++) {
+        gCpuInfo.brand.push_back(result[r]);
+        result[r] >>= 8;
+      }
+    }
+  }
+
+  // check for AVX2
+  {
+    int result[4];
+    __cpuidex(result, 7, 0);
+    gCpuInfo.has_avx2 = result[1] & (1 << 5);
+  }
+
+  {
+    int result[4];
+    __cpuidex(result, 1, 0);
+    gCpuInfo.has_avx = result[2] & (1 << 28);
+  }
+
+  printf("-------- CPU Information --------\n");
+  printf(" Brand: %s\n", gCpuInfo.brand.c_str());
+  printf(" Model: %s\n", gCpuInfo.model.c_str());
+  printf(" AVX  : %s\n", gCpuInfo.has_avx ? "true" : "false");
+  printf(" AVX2 : %s\n", gCpuInfo.has_avx2 ? "true" : "false");
+
+  gCpuInfo.initialized = true;
+}
+
+CpuInfo& get_cpu_info() {
+  return gCpuInfo;
+}