Fixing cpuid function on some compilers/architectures.

ahknight · Feb 1, 2015 · 3831f5b · 3831f5b
1 parent 6eeded0
commit 3831f5b
Show file tree

Hide file tree

Showing 2 changed files with 34 additions and 19 deletions.
diff --git a/crc32c/crc32c.c b/crc32c/crc32c.c
@@ -18,6 +18,21 @@ static uint32_t crc32c_CPUDetection(uint32_t crc, const void* data, size_t lengt
 
 CRC32CFunctionPtr crc32c = crc32c_CPUDetection;
 
+// Use the compiler's cpuid, if it exists.
+#if (defined __GNUC__) || (defined __clang__)
+#include <cpuid.h>
+
+static uint32_t cpuid(uint32_t functionInput) {
+  uint32_t eax;
+  uint32_t ebx;
+  uint32_t ecx;
+  uint32_t edx;
+  __cpuid(functionInput, eax, ebx, ecx, edx);
+  return ecx;
+}
+
+#else
+// Basic implementation.  Seems to only cover x86, not x64.
 static uint32_t cpuid(uint32_t functionInput) {
     uint32_t eax;
     uint32_t ebx;
@@ -26,16 +41,16 @@ static uint32_t cpuid(uint32_t functionInput) {
 #ifdef __PIC__
     // PIC: Need to save and restore ebx See:
     // http://sam.zoy.org/blog/2007-04-13-shlib-with-non-pic-code-have-inline-assembly-and-pic-mix-well
-    asm("pushl %%ebx\n\t" /* save %ebx */
-        "cpuid\n\t"
-        "movl %%ebx, %[ebx]\n\t" /* save what cpuid just put in %ebx */
-        "popl %%ebx" : "=a"(eax), [ebx] "=r"(ebx), "=c"(ecx), "=d"(edx) : "a" (functionInput)
-        : "cc");
+    __asm__("pushl %%ebx\n\t" /* save %ebx */
+            "cpuid\n\t"
+            "movl %%ebx, %[ebx]\n\t" /* save what cpuid just put in %ebx */
+            "popl %%ebx" : "=a"(eax), [ebx] "=r"(ebx), "=c"(ecx), "=d"(edx) : "a" (functionInput) : "cc");
 #else
-    asm("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (functionInput));
+    __asm__("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (functionInput));
 #endif
     return ecx;
 }
+#endif
 
 CRC32CFunctionPtr detectBestCRC32C() {
     static const int SSE42_BIT = 20;

diff --git a/crc32c/crc32c_tables.c b/crc32c/crc32c_tables.c
@@ -14,20 +14,20 @@
  #define CRCINIT 0xFFFFFFFF
  
  void init() {
- for (uint32_t i = 0; i <= 0xFF; i++) {
- uint32_t x = i;
- for (uint32_t j = 0; j < 8; j++)
- x = (x>>1) ^ (CRCPOLY & (-(int32_t)(x & 1)));
- g_crc_slicing[0][i] = x;
- }
+     for (uint32_t i = 0; i <= 0xFF; i++) {
+         uint32_t x = i;
+         for (uint32_t j = 0; j < 8; j++)
+             x = (x>>1) ^ (CRCPOLY & (-(int32_t)(x & 1)));
+         g_crc_slicing[0][i] = x;
+     }
  
- for (uint32_t i = 0; i <= 0xFF; i++) {
- uint32_t c = g_crc_slicing[0][i];
- for (uint32_t j = 1; j < 8; j++) {
- c = g_crc_slicing[0][c & 0xFF] ^ (c >> 8);
- g_crc_slicing[j][i] = c;
- }
- }
+     for (uint32_t i = 0; i <= 0xFF; i++) {
+         uint32_t c = g_crc_slicing[0][i];
+         for (uint32_t j = 1; j < 8; j++) {
+             c = g_crc_slicing[0][c & 0xFF] ^ (c >> 8);
+             g_crc_slicing[j][i] = c;
+         }
+     }
  }
  */