#if defined(PIPE_OS_WINDOWS)
#include <windows.h>
-#if defined(MSVC)
+#if defined(PIPE_CC_MSVC)
#include <intrin.h>
#endif
#endif
+#ifdef DEBUG
DEBUG_GET_ONCE_BOOL_OPTION(dump_cpu, "GALLIUM_DUMP_CPU", FALSE)
+#endif
struct util_cpu_caps util_cpu_caps;
#endif
-#if defined(PIPE_ARCH_X86)
-
-/* The sigill handlers */
-#if defined(PIPE_OS_LINUX) /*&& defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)*/
-static void
-sigill_handler_sse(int signal, struct sigcontext sc)
-{
- /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
- * instructions are 3 bytes long. We must increment the instruction
- * pointer manually to avoid repeated execution of the offending
- * instruction.
- *
- * If the SIGILL is caused by a divide-by-zero when unmasked
- * exceptions aren't supported, the SIMD FPU status and control
- * word will be restored at the end of the test, so we don't need
- * to worry about doing it here. Besides, we may not be able to...
- */
- sc.eip += 3;
-
- util_cpu_caps.has_sse=0;
-}
-
-static void
-sigfpe_handler_sse(int signal, struct sigcontext sc)
-{
- if (sc.fpstate->magic != 0xffff) {
- /* Our signal context has the extended FPU state, so reset the
- * divide-by-zero exception mask and clear the divide-by-zero
- * exception bit.
- */
- sc.fpstate->mxcsr |= 0x00000200;
- sc.fpstate->mxcsr &= 0xfffffffb;
- } else {
- /* If we ever get here, we're completely hosed.
- */
- }
-}
-#endif /* PIPE_OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
-
-#if defined(PIPE_OS_WINDOWS)
-static LONG CALLBACK
-win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
-{
- if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
- ep->ContextRecord->Eip +=3;
- util_cpu_caps.has_sse=0;
- return EXCEPTION_CONTINUE_EXECUTION;
- }
- return EXCEPTION_CONTINUE_SEARCH;
-}
-#endif /* PIPE_OS_WINDOWS */
-
-#endif /* PIPE_ARCH_X86 */
-
-
#if defined(PIPE_ARCH_PPC) && !defined(PIPE_OS_APPLE)
static jmp_buf __lv_powerpc_jmpbuf;
static volatile sig_atomic_t __lv_powerpc_canjump = 0;
static INLINE void
cpuid(uint32_t ax, uint32_t *p)
{
-#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
+#if (defined(PIPE_CC_GCC) || defined(PIPE_CC_SUNPRO)) && defined(PIPE_ARCH_X86)
__asm __volatile (
"xchgl %%ebx, %1\n\t"
"cpuid\n\t"
"=d" (p[3])
: "0" (ax)
);
-#elif defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86_64)
+#elif (defined(PIPE_CC_GCC) || defined(PIPE_CC_SUNPRO)) && defined(PIPE_ARCH_X86_64)
__asm __volatile (
"cpuid\n\t"
: "=a" (p[0]),
p[2] = 0;
p[3] = 0;
#endif
+}
+
+static INLINE uint64_t xgetbv(void)
+{
+#if defined(PIPE_CC_GCC)
+ uint32_t eax, edx;
+
+ __asm __volatile (
+ ".byte 0x0f, 0x01, 0xd0" // xgetbv isn't supported on gcc < 4.4
+ : "=a"(eax),
+ "=d"(edx)
+ : "c"(0)
+ );
+
+ return ((uint64_t)edx << 32) | eax;
+#elif defined(PIPE_CC_MSVC) && defined(_MSC_FULL_VER) && defined(_XCR_XFEATURE_ENABLED_MASK)
+ return _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
+#else
+ return 0;
+#endif
+
}
#endif /* X86 or X86_64 */
util_cpu_caps.nr_cpus = 1;
#endif
+ /* Make the fallback cacheline size nonzero so that it can be
+ * safely passed to align().
+ */
+ util_cpu_caps.cacheline = sizeof(void *);
+
#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
if (has_cpuid()) {
uint32_t regs[4];
util_cpu_caps.x86_cpu_type = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
/* general feature flags */
- util_cpu_caps.has_tsc = (regs2[3] & (1 << 8 )) >> 8; /* 0x0000010 */
- util_cpu_caps.has_mmx = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
- util_cpu_caps.has_sse = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
- util_cpu_caps.has_sse2 = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
- util_cpu_caps.has_sse3 = (regs2[2] & (1)); /* 0x0000001 */
- util_cpu_caps.has_ssse3 = (regs2[2] & (1 << 9 )) >> 9; /* 0x0000020 */
- util_cpu_caps.has_sse4_1 = (regs2[2] & (1 << 19)) >> 19;
+ util_cpu_caps.has_tsc = (regs2[3] >> 4) & 1; /* 0x0000010 */
+ util_cpu_caps.has_mmx = (regs2[3] >> 23) & 1; /* 0x0800000 */
+ util_cpu_caps.has_sse = (regs2[3] >> 25) & 1; /* 0x2000000 */
+ util_cpu_caps.has_sse2 = (regs2[3] >> 26) & 1; /* 0x4000000 */
+ util_cpu_caps.has_sse3 = (regs2[2] >> 0) & 1; /* 0x0000001 */
+ util_cpu_caps.has_ssse3 = (regs2[2] >> 9) & 1; /* 0x0000020 */
+ util_cpu_caps.has_sse4_1 = (regs2[2] >> 19) & 1;
+ util_cpu_caps.has_sse4_2 = (regs2[2] >> 20) & 1;
+ util_cpu_caps.has_popcnt = (regs2[2] >> 23) & 1;
+ util_cpu_caps.has_avx = ((regs2[2] >> 28) & 1) && // AVX
+ ((regs2[2] >> 27) & 1) && // OSXSAVE
+ ((xgetbv() & 6) == 6); // XMM & YMM
+ util_cpu_caps.has_f16c = (regs2[2] >> 29) & 1;
util_cpu_caps.has_mmx2 = util_cpu_caps.has_sse; /* SSE cpus supports mmxext too */
cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
util_cpu_caps.cacheline = cacheline;
}
+ if (regs[1] == 0x756e6547 && regs[2] == 0x6c65746e && regs[3] == 0x49656e69) {
+ /* GenuineIntel */
+ util_cpu_caps.has_intel = 1;
+ }
+
cpuid(0x80000000, regs);
if (regs[0] >= 0x80000001) {
cpuid(0x80000001, regs2);
- util_cpu_caps.has_mmx |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
- util_cpu_caps.has_mmx2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
- util_cpu_caps.has_3dnow = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
- util_cpu_caps.has_3dnow_ext = (regs2[3] & (1 << 30 )) >> 30;
+ util_cpu_caps.has_mmx |= (regs2[3] >> 23) & 1;
+ util_cpu_caps.has_mmx2 |= (regs2[3] >> 22) & 1;
+ util_cpu_caps.has_3dnow = (regs2[3] >> 31) & 1;
+ util_cpu_caps.has_3dnow_ext = (regs2[3] >> 30) & 1;
}
if (regs[0] >= 0x80000006) {
debug_printf("util_cpu_caps.has_sse3 = %u\n", util_cpu_caps.has_sse3);
debug_printf("util_cpu_caps.has_ssse3 = %u\n", util_cpu_caps.has_ssse3);
debug_printf("util_cpu_caps.has_sse4_1 = %u\n", util_cpu_caps.has_sse4_1);
+ debug_printf("util_cpu_caps.has_sse4_2 = %u\n", util_cpu_caps.has_sse4_2);
+ debug_printf("util_cpu_caps.has_avx = %u\n", util_cpu_caps.has_avx);
debug_printf("util_cpu_caps.has_3dnow = %u\n", util_cpu_caps.has_3dnow);
debug_printf("util_cpu_caps.has_3dnow_ext = %u\n", util_cpu_caps.has_3dnow_ext);
debug_printf("util_cpu_caps.has_altivec = %u\n", util_cpu_caps.has_altivec);
projects / mesa.git / blobdiff