diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 675ae9458111f6414e0bdccd51099f31fa8eb593..a16fb6a82d1a76fa51394db5440e94ba26da1b5d 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,23 @@
+2008-10-11 Uros Bizjak <ubizjak@gmail.com>
+ Andi Kleen <ak@linux.intel.com>
+
+ * config/i386/cpuid.h (__cpuid_count): New defines.
+ * config/i386/driver-i386.c (struct cache_desc): New structure.
+ (describe_cache): Use struct cache_desc to pass cache descriptions.
+ (detect_l2_cache): Ditto. Rename from decode_l2_cache.
+ (detect_caches_amd): Use struct cache_desc to describe caches.
+ (decode_caches_intel): Use struct cache_desc to pass cache
+ descriptions. Update descriptions to match latest (rev -032,
+ December 2007) CPUID documentation. Do not check valid bit here.
+ Check for Xeon MP value 0x49 problems.
+ (detect_caches_cpuid2): New function, split from detect_caches_intel.
+ Check valid bit before calling decode_caches_intel. Detect number
+ of times to repeat CPUID instruction.
+ (detect_caches_cpuid4): New function.
+ (detect_caches_intel): Depending on max_level, call
+ detect_caches_cpuid2 or detect_caches_cpuid4. Call detect_l2_cache
+ only when other methods fail to provide valid L2 cache description.
+
2008-10-11 John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
PR middle-end/37608
@@ -13,10 +33,10 @@
2008-10-11 David Edelsohn <edelsohn@gnu.org>
- * config/rs6000/rs6000.md (aux_truncdfsf2): Remove TARGET_SINGLE_FLOAT.
- (addsf3, subsf3, mulsf3 ! TARGET_POWERPC): Remove TARGET_SINGLE_FLOAT
- and fp_type.
- (divdf3): Reformat long line.
+ * config/rs6000/rs6000.md (aux_truncdfsf2): Remove TARGET_SINGLE_FLOAT.
+ (addsf3, subsf3, mulsf3 ! TARGET_POWERPC): Remove TARGET_SINGLE_FLOAT
+ and fp_type.
+ (divdf3): Reformat long line.
2008-10-11 Michael J. Eager <eager@eagercon.com>
@@ -25,8 +45,7 @@
(rs6000_handle_option): Process -mfpu options.
* config/rs6000/rs6000.h: (TARGET_XILINX_FPU): New.
(enum fpu_type_t): New.
- * config/rs6000/rs6000.md (attr fp_type): New.
- Include xfpu.md.
+ * config/rs6000/rs6000.md (attr fp_type): New. Include xfpu.md.
(addsf3, subsf3, mulsf3, adddf3, subdf3, muldf3, trunctfdf2): Set
fp_type.
(floatsisf2): Remove TARGET_SINGLE_FPU condition.
diff --git a/gcc/config/i386/cpuid.h b/gcc/config/i386/cpuid.h
index 90a28137afb86c24c28c9e302fe0427a5b26906c..69d25a2fe1d0188c7d66cac9d4653295eefa2214 100644
--- a/gcc/config/i386/cpuid.h
+++ b/gcc/config/i386/cpuid.h
@@ -74,6 +74,13 @@
"xchg{l}\t{%%}ebx, %1\n\t" \
: "=a" (a), "=r" (b), "=c" (c), "=d" (d) \
: "0" (level))
+
+#define __cpuid_count(level, count, a, b, c, d) \
+ __asm__ ("xchg{l}\t{%%}ebx, %1\n\t" \
+ "cpuid\n\t" \
+ "xchg{l}\t{%%}ebx, %1\n\t" \
+ : "=a" (a), "=r" (b), "=c" (c), "=d" (d) \
+ : "0" (level), "2" (count))
#else
/* Host GCCs older than 3.0 weren't supporting Intel asm syntax
nor alternatives in i386 code. */
@@ -83,12 +90,24 @@
"xchgl\t%%ebx, %1\n\t" \
: "=a" (a), "=r" (b), "=c" (c), "=d" (d) \
: "0" (level))
+
+#define __cpuid_count(level, count, a, b, c, d) \
+ __asm__ ("xchgl\t%%ebx, %1\n\t" \
+ "cpuid\n\t" \
+ "xchgl\t%%ebx, %1\n\t" \
+ : "=a" (a), "=r" (b), "=c" (c), "=d" (d) \
+ : "0" (level), "2" (count))
#endif
#else
#define __cpuid(level, a, b, c, d) \
__asm__ ("cpuid\n\t" \
: "=a" (a), "=b" (b), "=c" (c), "=d" (d) \
: "0" (level))
+
+#define __cpuid_count(level, count, a, b, c, d) \
+ __asm__ ("cpuid\n\t" \
+ : "=a" (a), "=b" (b), "=c" (c), "=d" (d) \
+ : "0" (level), "2" (count))
#endif
/* Return highest supported input value for cpuid instruction. ext can
diff --git a/gcc/config/i386/driver-i386.c b/gcc/config/i386/driver-i386.c
index 013fe93b631f9e7f2be9736cf9f112aae8c4c5b5..72e2c48d385b459f9d957fb4071f50eac9f611fd 100644
--- a/gcc/config/i386/driver-i386.c
+++ b/gcc/config/i386/driver-i386.c
@@ -1,5 +1,5 @@
/* Subroutines for the gcc driver.
- Copyright (C) 2006, 2007 Free Software Foundation, Inc.
+ Copyright (C) 2006, 2007, 2008 Free Software Foundation, Inc.
This file is part of GCC.
@@ -28,34 +28,45 @@ const char *host_detect_local_cpu (int argc, const char **argv);
#ifdef __GNUC__
#include "cpuid.h"
-/* Returns parameters that describe L1_ASSOC associative cache of size
- L1_SIZEKB with lines of size L1_LINE. */
+struct cache_desc
+{
+ unsigned sizekb;
+ unsigned assoc;
+ unsigned line;
+};
+
+/* Returns command line parameters that describe size and
+ cache line size of the processor caches. */
static char *
-describe_cache (unsigned l1_sizekb, unsigned l1_line,
- unsigned l1_assoc ATTRIBUTE_UNUSED, unsigned l2_sizekb)
+describe_cache (struct cache_desc level1, struct cache_desc level2)
{
char size[100], line[100], size2[100];
- /* At the moment, gcc middle-end does not use the information about the
- associativity of the cache. */
+ /* At the moment, gcc does not use the information
+ about the associativity of the cache. */
+
+ sprintf (size, "--param l1-cache-size=%u", level1.sizekb);
+ sprintf (line, "--param l1-cache-line-size=%u", level1.line);
- sprintf (size, "--param l1-cache-size=%u", l1_sizekb);
- sprintf (line, "--param l1-cache-line-size=%u", l1_line);
- sprintf (size2, "--param l2-cache-size=%u", l2_sizekb);
+ sprintf (size2, "--param l2-cache-size=%u", level2.sizekb);
return concat (size, " ", line, " ", size2, " ", NULL);
}
+/* Detect L2 cache parameters using CPUID extended function 0x80000006. */
+
static void
-decode_l2_cache (unsigned *l2_size, unsigned *l2_line, unsigned *l2_assoc)
+detect_l2_cache (struct cache_desc *level2)
{
- unsigned eax, ebx, ecx, edx, assoc;
+ unsigned eax, ebx, ecx, edx;
+ unsigned assoc;
__cpuid (0x80000006, eax, ebx, ecx, edx);
- *l2_size = (ecx >> 16) & 0xffff;
- *l2_line = ecx & 0xff;
+ level2->sizekb = (ecx >> 16) & 0xffff;
+ level2->line = ecx & 0xff;
+
assoc = (ecx >> 12) & 0xf;
if (assoc == 6)
assoc = 8;
@@ -65,7 +76,8 @@ decode_l2_cache (unsigned *l2_size, unsigned *l2_line, unsigned *l2_assoc)
assoc = 32 + (assoc - 0xa) * 16;
else if (assoc >= 0xd && assoc <= 0xe)
assoc = 96 + (assoc - 0xd) * 32;
- *l2_assoc = assoc;
+
+ level2->assoc = assoc;
}
/* Returns the description of caches for an AMD processor. */
@@ -74,243 +86,261 @@ static const char *
detect_caches_amd (unsigned max_ext_level)
{
unsigned eax, ebx, ecx, edx;
- unsigned l1_sizekb, l1_line, l1_assoc;
- unsigned l2_sizekb = 0, l2_line = 0, l2_assoc = 0;
+
+ struct cache_desc level1, level2 = {0, 0, 0};
if (max_ext_level < 0x80000005)
return "";
__cpuid (0x80000005, eax, ebx, ecx, edx);
- l1_line = ecx & 0xff;
- l1_sizekb = (ecx >> 24) & 0xff;
- l1_assoc = (ecx >> 16) & 0xff;
+ level1.sizekb = (ecx >> 24) & 0xff;
+ level1.assoc = (ecx >> 16) & 0xff;
+ level1.line = ecx & 0xff;
if (max_ext_level >= 0x80000006)
- decode_l2_cache (&l2_sizekb, &l2_line, &l2_assoc);
+ detect_l2_cache (&level2);
- return describe_cache (l1_sizekb, l1_line, l1_assoc, l2_sizekb);
+ return describe_cache (level1, level2);
}
-/* Stores the size of the L1/2 cache and cache line, and the associativity
- of the cache according to REG to L1_SIZEKB, L1_LINE, L1_ASSOC and
- L2_SIZEKB. */
+/* Decodes the size, the associativity and the cache line size of
+ L1/L2 caches of an Intel processor. Values are based on
+ "Intel Processor Identification and the CPUID Instruction"
+ [Application Note 485], revision -032, December 2007. */
static void
-decode_caches_intel (unsigned reg, unsigned *l1_sizekb, unsigned *l1_line,
- unsigned *l1_assoc, unsigned *l2_sizekb,
- unsigned *l2_line, unsigned *l2_assoc)
+decode_caches_intel (unsigned reg, bool xeon_mp,
+ struct cache_desc *level1, struct cache_desc *level2)
{
- unsigned i, val;
+ int i;
+
+ for (i = 24; i >= 0; i -= 8)
+ switch ((reg >> i) & 0xff)
+ {
+ case 0x0a:
+ level1->sizekb = 8; level1->assoc = 2; level1->line = 32;
+ break;
+ case 0x0c:
+ level1->sizekb = 16; level1->assoc = 4; level1->line = 32;
+ break;
+ case 0x2c:
+ level1->sizekb = 32; level1->assoc = 8; level1->line = 64;
+ break;
+ case 0x39:
+ level2->sizekb = 128; level2->assoc = 4; level2->line = 64;
+ break;
+ case 0x3a:
+ level2->sizekb = 192; level2->assoc = 6; level2->line = 64;
+ break;
+ case 0x3b:
+ level2->sizekb = 128; level2->assoc = 2; level2->line = 64;
+ break;
+ case 0x3c:
+ level2->sizekb = 256; level2->assoc = 4; level2->line = 64;
+ break;
+ case 0x3d:
+ level2->sizekb = 384; level2->assoc = 6; level2->line = 64;
+ break;
+ case 0x3e:
+ level2->sizekb = 512; level2->assoc = 4; level2->line = 64;
+ break;
+ case 0x41:
+ level2->sizekb = 128; level2->assoc = 4; level2->line = 32;
+ break;
+ case 0x42:
+ level2->sizekb = 256; level2->assoc = 4; level2->line = 32;
+ break;
+ case 0x43:
+ level2->sizekb = 512; level2->assoc = 4; level2->line = 32;
+ break;
+ case 0x44:
+ level2->sizekb = 1024; level2->assoc = 4; level2->line = 32;
+ break;
+ case 0x45:
+ level2->sizekb = 2048; level2->assoc = 4; level2->line = 32;
+ break;
+ case 0x49:
+ if (xeon_mp)
+ break;
+ level2->sizekb = 4096; level2->assoc = 16; level2->line = 64;
+ break;
+ case 0x4e:
+ level2->sizekb = 6144; level2->assoc = 24; level2->line = 64;
+ break;
+ case 0x60:
+ level1->sizekb = 16; level1->assoc = 8; level1->line = 64;
+ break;
+ case 0x66:
+ level1->sizekb = 8; level1->assoc = 4; level1->line = 64;
+ break;
+ case 0x67:
+ level1->sizekb = 16; level1->assoc = 4; level1->line = 64;
+ break;
+ case 0x68:
+ level1->sizekb = 32; level1->assoc = 4; level1->line = 64;
+ break;
+ case 0x78:
+ level2->sizekb = 1024; level2->assoc = 4; level2->line = 64;
+ break;
+ case 0x79:
+ level2->sizekb = 128; level2->assoc = 8; level2->line = 64;
+ break;
+ case 0x7a:
+ level2->sizekb = 256; level2->assoc = 8; level2->line = 64;
+ break;
+ case 0x7b:
+ level2->sizekb = 512; level2->assoc = 8; level2->line = 64;
+ break;
+ case 0x7c:
+ level2->sizekb = 1024; level2->assoc = 8; level2->line = 64;
+ break;
+ case 0x7d:
+ level2->sizekb = 2048; level2->assoc = 8; level2->line = 64;
+ break;
+ case 0x7f:
+ level2->sizekb = 512; level2->assoc = 2; level2->line = 64;
+ break;
+ case 0x82:
+ level2->sizekb = 256; level2->assoc = 8; level2->line = 32;
+ break;
+ case 0x83:
+ level2->sizekb = 512; level2->assoc = 8; level2->line = 32;
+ break;
+ case 0x84:
+ level2->sizekb = 1024; level2->assoc = 8; level2->line = 32;
+ break;
+ case 0x85:
+ level2->sizekb = 2048; level2->assoc = 8; level2->line = 32;
+ break;
+ case 0x86:
+ level2->sizekb = 512; level2->assoc = 4; level2->line = 64;
+ break;
+ case 0x87:
+ level2->sizekb = 1024; level2->assoc = 8; level2->line = 64;
+
+ default:
+ break;
+ }
+}
- if (((reg >> 31) & 1) != 0)
- return;
+/* Detect cache parameters using CPUID function 2. */
- for (i = 0; i < 4; i++)
+static void
+detect_caches_cpuid2 (bool xeon_mp,
+ struct cache_desc *level1, struct cache_desc *level2)
+{
+ unsigned eax, ebx, ecx, edx;
+ int nreps;
+
+ __cpuid (2, eax, ebx, ecx, edx);
+
+ nreps = eax & 0x0f;
+ eax &= ~0x0f;
+
+ while (--nreps >= 0)
{
- val = reg & 0xff;
- reg >>= 8;
+ if (!((eax >> 31) & 1))
+ decode_caches_intel (eax, xeon_mp, level1, level2);
+ if (!((ebx >> 31) & 1))
+ decode_caches_intel (ebx, xeon_mp, level1, level2);
+ if (!((ecx >> 31) & 1))
+ decode_caches_intel (ecx, xeon_mp, level1, level2);
+ if (!((edx >> 31) & 1))
+ decode_caches_intel (edx, xeon_mp, level1, level2);
+
+ if (nreps)
+ __cpuid (2, eax, ebx, ecx, edx);
+ }
+}
- switch (val)
- {
- case 0xa:
- *l1_sizekb = 8;
- *l1_line = 32;
- *l1_assoc = 2;
- break;
- case 0xc:
- *l1_sizekb = 16;
- *l1_line = 32;
- *l1_assoc = 4;
- break;
- case 0x2c:
- *l1_sizekb = 32;
- *l1_line = 64;
- *l1_assoc = 8;
- break;
- case 0x39:
- *l2_sizekb = 128;
- *l2_line = 64;
- *l2_assoc = 4;
- break;
- case 0x3a:
- *l2_sizekb = 192;
- *l2_line = 64;
- *l2_assoc = 6;
- break;
- case 0x3b:
- *l2_sizekb = 128;
- *l2_line = 64;
- *l2_assoc = 2;
- break;
- case 0x3c:
- *l2_sizekb = 256;
- *l2_line = 64;
- *l2_assoc = 4;
- break;
- case 0x3d:
- *l2_sizekb = 384;
- *l2_line = 64;
- *l2_assoc = 6;
- break;
- case 0x3e:
- *l2_sizekb = 512;
- *l2_line = 64;
- *l2_assoc = 4;
- break;
- case 0x41:
- *l2_sizekb = 128;
- *l2_line = 32;
- *l2_assoc = 4;
- break;
- case 0x42:
- *l2_sizekb = 256;
- *l2_line = 32;
- *l2_assoc = 4;
- break;
- case 0x43:
- *l2_sizekb = 512;
- *l2_line = 32;
- *l2_assoc = 4;
- break;
- case 0x44:
- *l2_sizekb = 1024;
- *l2_line = 32;
- *l2_assoc = 4;
- break;
- case 0x45:
- *l2_sizekb = 2048;
- *l2_line = 32;
- *l2_assoc = 4;
- break;
- case 0x49:
- *l2_sizekb = 4096;
- *l2_line = 64;
- *l2_assoc = 16;
- break;
- case 0x60:
- *l1_sizekb = 16;
- *l1_line = 64;
- *l1_assoc = 8;
- break;
- case 0x66:
- *l1_sizekb = 8;
- *l1_line = 64;
- *l1_assoc = 4;
- break;
- case 0x67:
- *l1_sizekb = 16;
- *l1_line = 64;
- *l1_assoc = 4;
- break;
- case 0x68:
- *l1_sizekb = 32;
- *l1_line = 64;
- *l1_assoc = 4;
- break;
- case 0x78:
- *l2_sizekb = 1024;
- *l2_line = 64;
- *l2_assoc = 4;
- break;
- case 0x79:
- *l2_sizekb = 128;
- *l2_line = 64;
- *l2_assoc = 8;
- break;
- case 0x7a:
- *l2_sizekb = 256;
- *l2_line = 64;
- *l2_assoc = 8;
- break;
- case 0x7b:
- *l2_sizekb = 512;
- *l2_line = 64;
- *l2_assoc = 8;
- break;
- case 0x7c:
- *l2_sizekb = 1024;
- *l2_line = 64;
- *l2_assoc = 8;
- break;
- case 0x7d:
- *l2_sizekb = 2048;
- *l2_line = 64;
- *l2_assoc = 8;
- break;
- case 0x7f:
- *l2_sizekb = 512;
- *l2_line = 64;
- *l2_assoc = 2;
- break;
- case 0x82:
- *l2_sizekb = 256;
- *l2_line = 32;
- *l2_assoc = 8;
- break;
- case 0x83:
- *l2_sizekb = 512;
- *l2_line = 32;
- *l2_assoc = 8;
- break;
- case 0x84:
- *l2_sizekb = 1024;
- *l2_line = 32;
- *l2_assoc = 8;
- break;
- case 0x85:
- *l2_sizekb = 2048;
- *l2_line = 32;
- *l2_assoc = 8;
- break;
- case 0x86:
- *l2_sizekb = 512;
- *l2_line = 64;
- *l2_assoc = 4;
- break;
- case 0x87:
- *l2_sizekb = 1024;
- *l2_line = 64;
- *l2_assoc = 8;
- break;
+/* Detect cache parameters using CPUID function 4. This
+ method doesn't require hardcoded tables. */
+enum cache_type
+{
+ CACHE_END = 0,
+ CACHE_DATA = 1,
+ CACHE_INST = 2,
+ CACHE_UNIFIED = 3
+};
+
+static void
+detect_caches_cpuid4 (struct cache_desc *level1, struct cache_desc *level2)
+{
+ struct cache_desc *cache;
+
+ unsigned eax, ebx, ecx, edx;
+ int count;
+
+ for (count = 0;; count++)
+ {
+ __cpuid_count(4, count, eax, ebx, ecx, edx);
+ switch (eax & 0x1f)
+ {
+ case CACHE_END:
+ return;
+ case CACHE_DATA:
+ case CACHE_UNIFIED:
+ {
+ switch ((eax >> 5) & 0x07)
+ {
+ case 1:
+ cache = level1;
+ break;
+ case 2:
+ cache = level2;
+ break;
+ default:
+ cache = NULL;
+ }
+
+ if (cache)
+ {
+ unsigned sets = ecx + 1;
+ unsigned part;
+
+ cache->line = (ebx & 0x0fff) + 1;
+ ebx >>= 12;
+
+ part = (ebx & 0x03ff) + 1;
+ ebx >>= 10;
+
+ cache->assoc = (ebx & 0x03ff) + 1;
+
+ cache->sizekb = (cache->assoc * part
+ * cache->line * sets) / 1024;
+ }
+ }
default:
break;
}
}
}
-/* Returns the description of caches for an intel processor. */
+/* Returns the description of caches for an Intel processor. */
static const char *
-detect_caches_intel (unsigned max_level, unsigned max_ext_level)
+detect_caches_intel (bool xeon_mp, unsigned max_level, unsigned max_ext_level)
{
- unsigned eax, ebx, ecx, edx;
- unsigned l1_sizekb = 0, l1_line = 0, assoc = 0;
- unsigned l2_sizekb = 0, l2_line = 0, l2_assoc = 0;
+ struct cache_desc level1 = {0, 0, 0}, level2 = {0, 0, 0};
- if (max_level < 2)
+ if (max_level >= 4)
+ detect_caches_cpuid4 (&level1, &level2);
+ else if (max_level >= 2)
+ detect_caches_cpuid2 (xeon_mp, &level1, &level2);
+ else
return "";
- __cpuid (2, eax, ebx, ecx, edx);
-
- decode_caches_intel (eax, &l1_sizekb, &l1_line, &assoc,
- &l2_sizekb, &l2_line, &l2_assoc);
- decode_caches_intel (ebx, &l1_sizekb, &l1_line, &assoc,
- &l2_sizekb, &l2_line, &l2_assoc);
- decode_caches_intel (ecx, &l1_sizekb, &l1_line, &assoc,
- &l2_sizekb, &l2_line, &l2_assoc);
- decode_caches_intel (edx, &l1_sizekb, &l1_line, &assoc,
- &l2_sizekb, &l2_line, &l2_assoc);
-
- if (!l1_sizekb)
+ if (level1.sizekb == 0)
return "";
- /* Newer Intel CPUs are equipped with AMD style L2 cache info */
- if (max_ext_level >= 0x80000006)
- decode_l2_cache (&l2_sizekb, &l2_line, &l2_assoc);
+ /* Intel CPUs are equipped with AMD style L2 cache info. Try this
+ method if other methods fail to provide L2 cache parameters. */
+ if (level2.sizekb == 0 && max_ext_level >= 0x80000006)
+ detect_l2_cache (&level2);
- return describe_cache (l1_sizekb, l1_line, assoc, l2_sizekb);
+ return describe_cache (level1, level2);
}
/* This will be called by the spec parser in gcc.c when it sees
@@ -334,11 +364,12 @@ const char *host_detect_local_cpu (int argc, const char **argv)
const char *cache = "";
const char *options = "";
- unsigned int eax, ebx, ecx, edx;
+ unsigned int eax, ebx, ecx, edx;
unsigned int max_level, ext_level;
+
unsigned int vendor;
- unsigned int family;
+ unsigned int model, family;
unsigned int has_sse3, has_ssse3, has_cmpxchg16b;
unsigned int has_cmpxchg8b, has_cmov, has_mmx, has_sse, has_sse2;
@@ -364,6 +395,7 @@ const char *host_detect_local_cpu (int argc, const char **argv)
__cpuid (1, eax, ebx, ecx, edx);
/* We don't care for extended family. */
+ model = (eax >> 4) & 0x0f;
family = (eax >> 8) & 0x0f;
has_sse3 = ecx & bit_SSE3;
@@ -396,7 +428,10 @@ const char *host_detect_local_cpu (int argc, const char **argv)
if (vendor == *(const unsigned int*) "Auth")
cache = detect_caches_amd (ext_level);
else if (vendor == *(const unsigned int*) "Genu")
- cache = detect_caches_intel (max_level, ext_level);
+ {
+ bool xeon_mp = (family == 15 && model == 6);
+ cache = detect_caches_intel (xeon_mp, max_level, ext_level);
+ }
}
if (vendor == *(const unsigned int*) "Auth")