https://github.com/JuliaLang/julia
Tip revision: 483dbf5279efe29f94cf35d42de0ab0a673fb34e authored by Tony Kelman on 27 July 2015, 06:18:41 UTC
Tag v0.3.11
Tag v0.3.11
Tip revision: 483dbf5
llvm-3.3.patch
diff -u -r -N llvm-3.3.src/lib/Support/Host.cpp llvm-3.3/lib/Support/Host.cpp
--- llvm-3.3.src/lib/Support/Host.cpp 2013-05-04 02:36:23.000000000 -0500
+++ llvm-3.3/lib/Support/Host.cpp 2014-06-05 15:08:41.975312974 -0500
@@ -52,8 +52,54 @@
/// GetX86CpuIDAndInfo - Execute the specified cpuid and return the 4 values in the
/// specified arguments. If we can't run cpuid on the host, return true.
-static bool GetX86CpuIDAndInfo(unsigned value, unsigned *rEAX,
- unsigned *rEBX, unsigned *rECX, unsigned *rEDX) {
+static bool GetX86CpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX,
+ unsigned *rECX, unsigned *rEDX) {
+#if defined(__GNUC__) || defined(__clang__)
+ #if defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64)
+ // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually.
+ asm ("movq\t%%rbx, %%rsi\n\t"
+ "cpuid\n\t"
+ "xchgq\t%%rbx, %%rsi\n\t"
+ : "=a" (*rEAX),
+ "=S" (*rEBX),
+ "=c" (*rECX),
+ "=d" (*rEDX)
+ : "a" (value));
+ return false;
+ #elif defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)
+ asm ("movl\t%%ebx, %%esi\n\t"
+ "cpuid\n\t"
+ "xchgl\t%%ebx, %%esi\n\t"
+ : "=a" (*rEAX),
+ "=S" (*rEBX),
+ "=c" (*rECX),
+ "=d" (*rEDX)
+ : "a" (value));
+ return false;
+// pedantic #else returns to appease -Wunreachable-code (so we don't generate
+// postprocessed code that looks like "return true; return false;")
+ #else
+ return true;
+ #endif
+#elif defined(_MSC_VER)
+ // The MSVC intrinsic is portable across x86 and x64.
+ int registers[4];
+ __cpuid(registers, value);
+ *rEAX = registers[0];
+ *rEBX = registers[1];
+ *rECX = registers[2];
+ *rEDX = registers[3];
+ return false;
+#else
+ return true;
+#endif
+}
+
+/// GetX86CpuIDAndInfoEx - Execute the specified cpuid with subleaf and return the
+/// 4 values in the specified arguments. If we can't run cpuid on the host,
+/// return true.
+bool GetX86CpuIDAndInfoEx(unsigned value, unsigned subleaf, unsigned *rEAX,
+ unsigned *rEBX, unsigned *rECX, unsigned *rEDX) {
#if defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64)
#if defined(__GNUC__)
// gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually.
@@ -64,16 +110,22 @@
"=S" (*rEBX),
"=c" (*rECX),
"=d" (*rEDX)
- : "a" (value));
+ : "a" (value),
+ "c" (subleaf));
return false;
#elif defined(_MSC_VER)
- int registers[4];
- __cpuid(registers, value);
- *rEAX = registers[0];
- *rEBX = registers[1];
- *rECX = registers[2];
- *rEDX = registers[3];
- return false;
+ // __cpuidex was added in MSVC++ 9.0 SP1
+ #if (_MSC_VER > 1500) || (_MSC_VER == 1500 && _MSC_FULL_VER >= 150030729)
+ int registers[4];
+ __cpuidex(registers, value, subleaf);
+ *rEAX = registers[0];
+ *rEBX = registers[1];
+ *rECX = registers[2];
+ *rEDX = registers[3];
+ return false;
+ #else
+ return true;
+ #endif
#else
return true;
#endif
@@ -86,11 +138,13 @@
"=S" (*rEBX),
"=c" (*rECX),
"=d" (*rEDX)
- : "a" (value));
+ : "a" (value),
+ "c" (subleaf));
return false;
#elif defined(_MSC_VER)
__asm {
mov eax,value
+ mov ecx,subleaf
cpuid
mov esi,rEAX
mov dword ptr [esi],eax
@@ -102,8 +156,6 @@
mov dword ptr [esi],edx
}
return false;
-// pedantic #else returns to appease -Wunreachable-code (so we don't generate
-// postprocessed code that looks like "return true; return false;")
#else
return true;
#endif
@@ -148,21 +200,27 @@
unsigned Model = 0;
DetectX86FamilyModel(EAX, Family, Model);
+ union {
+ unsigned u[3];
+ char c[12];
+ } text;
+
+ GetX86CpuIDAndInfo(0, &EAX, text.u+0, text.u+2, text.u+1);
+
+ unsigned MaxLeaf = EAX;
bool HasSSE3 = (ECX & 0x1);
+ bool HasSSE41 = (ECX & 0x80000);
// If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV
// indicates that the AVX registers will be saved and restored on context
// switch, then we have full AVX support.
const unsigned AVXBits = (1 << 27) | (1 << 28);
bool HasAVX = ((ECX & AVXBits) == AVXBits) && OSHasAVXSupport();
+ bool HasAVX2 = HasAVX && MaxLeaf >= 0x7 &&
+ !GetX86CpuIDAndInfoEx(0x7, 0x0, &EAX, &EBX, &ECX, &EDX) &&
+ (EBX & 0x20);
GetX86CpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
bool Em64T = (EDX >> 29) & 0x1;
- union {
- unsigned u[3];
- char c[12];
- } text;
-
- GetX86CpuIDAndInfo(0, &EAX, text.u+0, text.u+2, text.u+1);
if (memcmp(text.c, "GenuineIntel", 12) == 0) {
switch (Family) {
case 3:
@@ -244,7 +302,8 @@
// 17h. All processors are manufactured using the 45 nm process.
//
// 45nm: Penryn , Wolfdale, Yorkfield (XE)
- return "penryn";
+ // Not all Penryn processors support SSE 4.1 (such as the Pentium brand)
+ return HasSSE41 ? "penryn" : "core2";
case 26: // Intel Core i7 processor and Intel Xeon processor. All
// processors are manufactured using the 45 nm process.
@@ -269,10 +328,20 @@
// Ivy Bridge:
case 58:
+ case 62: // Ivy Bridge EP
// Not all Ivy Bridge processors support AVX (such as the Pentium
// versions instead of the i7 versions).
return HasAVX ? "core-avx-i" : "corei7";
+ // Haswell:
+ case 60:
+ case 63:
+ case 69:
+ case 70:
+ // Not all Haswell processors support AVX too (such as the Pentium
+ // versions instead of the i7 versions).
+ return HasAVX2 ? "core-avx2" : "corei7";
+
case 28: // Most 45 nm Intel Atom processors
case 38: // 45 nm Atom Lincroft
case 39: // 32 nm Atom Medfield