Added cpu_clock_by_ic() function to measure cpu clock using instruction counting. Still not translated to MSVC, and isn't tested on CPUs other than Core i7 (but there it works beatifully). Bumped version to 0.1.3.

git-svn-id: https://svn.code.sf.net/p/libcpuid/code/HEAD/libcpuid@77 3b4be424-7ac5-41d7-8526-f4ddcb85d872
2024-12-16 16:35:45 +00:00 · 2010-04-20 09:01:07 +00:00 · 2010-04-20 09:01:07 +00:00 · 44c313126b
commit 44c313126b
parent af2c364e85
14 changed files with 444 additions and 36 deletions
--- a/configure.ac
+++ b/configure.ac
@ -1,4 +1,4 @@
-AC_INIT([libcpuid CPU Identification library], [0.1.1], [libcpuid-devel@lists.sourceforge.net], [libcpuid])
+AC_INIT([libcpuid CPU Identification library], [0.1.3], [libcpuid-devel@lists.sourceforge.net], [libcpuid])
 AC_CONFIG_SRCDIR([libcpuid/libcpuid.h])
 AC_CONFIG_HEADERS([config.h])
 AM_INIT_AUTOMAKE([-Wall dist-bzip2 dist-zip foreign])
@ -12,9 +12,11 @@ dnl When you make a change that adds features, increment AGE, and set REVISION t
 dnl C:A:R
 dnl 10:0:0   Version 0.1.0
 dnl 11:0:0   Version 0.1.1: four more fields to cpu_raw_data_t
+dnl 11:0:1   Version 0.1.2: added cpu_msr* functions
+dnl 11:0:2   Version 0.1.3: added cpu_clock_by_ic() function
 LIBCPUID_CURRENT=11
 LIBCPUID_AGE=0
-LIBCPUID_REVISION=0
+LIBCPUID_REVISION=2
 AC_SUBST([LIBCPUID_AGE])
 AC_SUBST([LIBCPUID_REVISION])
 AC_SUBST([LIBCPUID_CURRENT])
--- a/cpuid_tool/cpuid_tool.c
+++ b/cpuid_tool/cpuid_tool.c
@ -83,6 +83,7 @@ typedef enum {
 	NEED_CLOCK,
 	NEED_CLOCK_OS,
 	NEED_CLOCK_RDTSC,
+	NEED_CLOCK_IC,
 	NEED_RDMSR,
 } output_data_switch;

@ -133,6 +134,7 @@ matchtable[] = {
 	{ NEED_CLOCK        , "--clock"        , 0},
 	{ NEED_CLOCK_OS     , "--clock-os"     , 0},
 	{ NEED_CLOCK_RDTSC  , "--clock-rdtsc"  , 1},
+	{ NEED_CLOCK_IC     , "--clock-ic"     , 1},
 	{ NEED_RDMSR        , "--rdmsr"        , 0},
 };

@ -404,6 +406,9 @@ static void print_info(output_data_switch query, struct cpu_raw_data_t* raw,
 		case NEED_CLOCK_RDTSC:
 			fprintf(fout, "%d\n", cpu_clock_measure(400, 1));
 			break;
+		case NEED_CLOCK_IC:
+			fprintf(fout, "%d\n", cpu_clock_by_ic(25, 16));
+			break;
 		case NEED_RDMSR:
 		{
 			if ((handle = cpu_msr_driver_open()) == NULL) {
--- a/libcpuid/Makefile.am
+++ b/libcpuid/Makefile.am
@ -28,6 +28,7 @@ noinst_HEADERS = 		\
 	asm-bits.h		\
 	libcpuid_util.h		\
 	recog_intel.h		\
-	recog_amd.h
+	recog_amd.h		\
+	rdtsc.h

 EXTRA_DIST += libcpuid.sym libcpuid.vcproj
--- a/libcpuid/Makefile.x64
+++ b/libcpuid/Makefile.x64
@ -5,7 +5,7 @@ all: libcpuid.lib
 ASM = ml64 /nologo
 CC = cl.exe /nologo /TC
 OPTFLAGS = /MT
-DEFINES = /D "VERSION=\"0.1.1\""
+DEFINES = /D "VERSION=\"0.1.3\""
 OBJECTS = masm-x64.obj asm-bits.obj cpuid_main.obj libcpuid_util.obj recog_amd.obj recog_intel.obj rdtsc.obj

 libcpuid.lib: $(OBJECTS)
--- a/libcpuid/Makefile.x86
+++ b/libcpuid/Makefile.x86
@ -12,7 +12,7 @@ all: libcpuid.lib

 CC = cl.exe /nologo /TC
 OPTFLAGS = /MT
-DEFINES = /D "VERSION=\"0.1.1\""
+DEFINES = /D "VERSION=\"0.1.3\""
 OBJECTS = asm-bits.obj cpuid_main.obj libcpuid_util.obj recog_amd.obj recog_intel.obj rdtsc.obj

 libcpuid.lib: $(OBJECTS)
--- a/libcpuid/asm-bits.c
+++ b/libcpuid/asm-bits.c
@ -197,3 +197,316 @@ void cpu_rdtsc(uint64_t* result)
 	*result = (uint64_t)low_part + (((uint64_t) hi_part) << 32);
 }
 #endif /* INLINE_ASM_SUPPORTED */
+
+#ifdef INLINE_ASM_SUPPORTED
+void busy_sse_loop(int cycles)
+{
+#ifdef COMPILER_GCC
+#ifndef __APPLE__
+#	define XALIGN ".balign 16\n"
+#else
+#	define XALIGN ".align 4\n"
+#endif
+	__asm __volatile (
+		"	xorps	%%xmm0,	%%xmm0\n"
+		"	xorps	%%xmm1,	%%xmm1\n"
+		"	xorps	%%xmm2,	%%xmm2\n"
+		"	xorps	%%xmm3,	%%xmm3\n"
+		"	xorps	%%xmm4,	%%xmm4\n"
+		"	xorps	%%xmm5,	%%xmm5\n"
+		"	xorps	%%xmm6,	%%xmm6\n"
+		"	xorps	%%xmm7,	%%xmm7\n"
+		XALIGN
+		".bsLoop:\n"
+		// 0:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 2:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 3:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 4:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 5:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 6:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 7:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 8:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		// 9:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//10:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//11:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//12:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//13:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//14:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//15:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//16:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//17:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//18:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//19:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//20:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//21:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//22:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//23:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//24:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//25:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//26:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//27:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//28:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//29:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//30:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		//31:
+		"	addps	%%xmm1, %%xmm0\n"
+		"	addps	%%xmm2, %%xmm1\n"
+		"	addps	%%xmm3, %%xmm2\n"
+		"	addps	%%xmm4, %%xmm3\n"
+		"	addps	%%xmm5, %%xmm4\n"
+		"	addps	%%xmm6, %%xmm5\n"
+		"	addps	%%xmm7, %%xmm6\n"
+		"	addps	%%xmm0, %%xmm7\n"
+		
+		"	dec	%%eax\n"
+		"	jnz	.bsLoop\n"
+		::"eax"(cycles)
+	);
+#else
+#  ifdef COMPILER_MICROSOFT
+#  else
+#    error "Unsupported compiler"
+#  endif /* COMPILER_MICROSOFT */
+#endif /* COMPILER_GCC */
+}
+#endif /* INLINE_ASSEMBLY_SUPPORTED */
--- a/libcpuid/asm-bits.h
+++ b/libcpuid/asm-bits.h
@ -48,5 +48,6 @@

 int cpuid_exists_by_eflags(void);
 void exec_cpuid(uint32_t *regs);
+void busy_sse_loop(int cycles);

 #endif /* __ASM_BITS_H__ */
--- a/libcpuid/exports.def
+++ b/libcpuid/exports.def
@ -22,3 +22,8 @@ cpuid_set_warn_function @18
 cpuid_set_verbosiness_level @19
 cpuid_get_cpu_list @20
 cpuid_free_cpu_list @21
+cpu_msr_driver_open @22
+cpu_rdmsr @23
+cpu_msrinfo @24
+cpu_msr_driver_close @25
+cpu_clock_by_ic @26
--- a/libcpuid/libcpuid.h
+++ b/libcpuid/libcpuid.h
@ -29,7 +29,7 @@
 * @File     libcpuid.h
 * @Author   Veselin Georgiev
 * @Date     Oct 2008
- * @Version  0.1.2
+ * @Version  0.1.3
 *
 * Version history:
 *
@ -38,6 +38,8 @@
 *                      new processor topology enumeration required on Core i7
 *  0.1.2 (2009-09-26): Added support for MSR reading through self-extracting
 *                      kernel driver on Win32.
+ *  0.1.3 (2010-04-20): Added support for greater more accurate CPU clock measurements
+ *                      with cpu_clock_by_ic()
 */

 /** @mainpage A simple libcpuid introduction
@ -615,6 +617,38 @@ int cpu_clock_by_os(void);
 */
 int cpu_clock_measure(int millis, int quad_check);

+/**
+ * @brief Measure the CPU clock frequency using instruction-counting
+ *
+ * @param millis - how much time to allocate for each run, in milliseconds
+ * @param runs - how many runs to perform
+ *
+ * The function performs a busy-wait cycle using a known number of "heavy" (SSE)
+ * instructions. These instructions run at (more or less guaranteed) 1 IPC rate,
+ * so by running a busy loop for a fixed amount of time, and measuring the
+ * amount of instructions done, the CPU clock is accurately measured.
+ *
+ * Of course, this function is still affected by the power-saving schemes, so
+ * the warnings as of cpu_clock_measure() still apply. However, this function is
+ * immune to problems with detection, related to the Intel Nehalem's "Turbo"
+ * mode, where the internal clock is raised, but the RDTSC rate is unaffected.
+ *
+ * The function will run for about (millis * runs) milliseconds.
+ * You can make only a single busy-wait run (runs == 1); however, this can
+ * be affected by task scheduling (which will break the counting), so allowing
+ * more than one run is recommended. As run length is not imperative for
+ * accurate readings (e.g., 50ms is sufficient), you can afford a lot of short
+ * runs, e.g. 10 runs of 50ms or 20 runs of 25ms.
+ *
+ * Recommended values - millis = 50, runs = 4. For more robustness,
+ * increase the number of runs.
+ *
+ * @returns the CPU clock frequency in MHz (within some measurement error
+ * margin). If SSE is not supported, the result is -1. If the input parameters
+ * are incorrect, or some other internal fault is detected, the result is -2.
+ */
+int cpu_clock_by_ic(int millis, int runs);
+
 /**
 * @brief Get the CPU clock frequency (all-in-one method)
 *
--- a/libcpuid/libcpuid.sym
+++ b/libcpuid/libcpuid.sym
@ -23,3 +23,4 @@ cpu_msr_driver_open
 cpu_rdmsr
 cpu_msrinfo
 cpu_msr_driver_close
+cpu_clock_by_ic
--- a/libcpuid/libcpuid_vc71.vcproj
+++ b/libcpuid/libcpuid_vc71.vcproj
@ -19,7 +19,7 @@
 			<Tool
 				Name="VCCLCompilerTool"
 				Optimization="0"
-				PreprocessorDefinitions="VERSION=\&quot;0.1.1\&quot;"
+				PreprocessorDefinitions="VERSION=\&quot;0.1.3\&quot;"
 				MinimalRebuild="TRUE"
 				BasicRuntimeChecks="3"
 				RuntimeLibrary="3"
@ -60,7 +60,7 @@
 			CharacterSet="2">
 			<Tool
 				Name="VCCLCompilerTool"
-				PreprocessorDefinitions="VERSION=\&quot;0.1.1\&quot;"
+				PreprocessorDefinitions="VERSION=\&quot;0.1.3\&quot;"
 				RuntimeLibrary="2"
 				UsePrecompiledHeader="0"
 				WarningLevel="3"
@ -99,7 +99,7 @@
 			CharacterSet="2">
 			<Tool
 				Name="VCCLCompilerTool"
-				PreprocessorDefinitions="VERSION=\&quot;0.1.1\&quot;"
+				PreprocessorDefinitions="VERSION=\&quot;0.1.3\&quot;"
 				RuntimeLibrary="2"
 				UsePrecompiledHeader="0"
 				WarningLevel="3"
@ -199,6 +199,9 @@
 			<File
 				RelativePath=".\recog_intel.h">
 			</File>
+			<File
+				RelativePath=".\rdtsc.h">
+			</File>
 		</Filter>
 		<Filter
 			Name="Resource Files"
--- a/libcpuid/libcpuid_vc9.vcproj
+++ b/libcpuid/libcpuid_vc9.vcproj
@ -40,7 +40,7 @@
 			<Tool
 				Name="VCCLCompilerTool"
 				Optimization="0"
-				PreprocessorDefinitions="WIN32;_DEBUG;_LIB;VERSION=&quot;\&quot;0.1.1\&quot;&quot;"
+				PreprocessorDefinitions="WIN32;_DEBUG;_LIB;VERSION=&quot;\&quot;0.1.3\&quot;&quot;"
 				MinimalRebuild="true"
 				BasicRuntimeChecks="3"
 				RuntimeLibrary="1"
@ -104,7 +104,7 @@
 			/>
 			<Tool
 				Name="VCCLCompilerTool"
-				PreprocessorDefinitions="WIN32;NDEBUG;_LIB;VERSION=&quot;\&quot;0.1.1\&quot;&quot;"
+				PreprocessorDefinitions="WIN32;NDEBUG;_LIB;VERSION=&quot;\&quot;0.1.3\&quot;&quot;"
 				RuntimeLibrary="0"
 				UsePrecompiledHeader="0"
 				WarningLevel="3"
@ -208,6 +208,10 @@
 				RelativePath=".\recog_intel.h"
 				>
 			</File>
+			<File
+				RelativePath=".\rdtsc.h"
+				>
+			</File>
 		</Filter>
 		<Filter
 			Name="Resource Files"
--- a/libcpuid/rdmsr.c
+++ b/libcpuid/rdmsr.c
@ -29,6 +29,7 @@
 #include "libcpuid.h"
 #include "asm-bits.h"
 #include "libcpuid_util.h"
+#include "rdtsc.h"

 #ifndef _WIN32
 #  ifdef __APPLE__
@ -359,10 +360,6 @@ int cpu_msr_driver_close(struct msr_driver_t* drv)

 #endif /* _WIN32 */

- /* from rdtsc.c: */
-void sys_precise_clock(uint64_t *result);
-int busy_loop_delay(int milliseconds);
-
 static int rdmsr_supported(void)
 {
 	struct cpu_id_t* id = get_cached_cpuid();
--- a/libcpuid/rdtsc.c
+++ b/libcpuid/rdtsc.c
@ -26,6 +26,9 @@
 #include <stdio.h>
 #include <string.h>
 #include "libcpuid.h"
+#include "libcpuid_util.h"
+#include "asm-bits.h"
+#include "rdtsc.h"

 #ifdef _WIN32
 #include <windows.h>
@ -223,6 +226,45 @@ int cpu_clock_measure(int millis, int quad_check)
 	return (results[bi] + results[bj] + _zero) / 2;
 }

+int cpu_clock_by_ic(int millis, int runs)
+{
+	int max_value = 0, cur_value, i, ri, cycles_inner, cycles_outer, c;
+	struct cpu_id_t* id;
+	uint64_t t0, t1, tl, hz;
+	if (millis <= 0 || runs <= 0) return -2;
+	id = get_cached_cpuid();
+	if (!id || !id->flags[CPU_FEATURE_SSE]) return -1;
+	//
+	tl = millis * 125; // (*1000 / 8)
+	cycles_inner = 128;
+	cycles_outer = 1;
+	do {
+		if (cycles_inner < 1000000000) cycles_inner *= 2;
+		else cycles_outer *= 2;
+		sys_precise_clock(&t0);
+		for (i = 0; i < cycles_outer; i++)
+			busy_sse_loop(cycles_inner);
+		sys_precise_clock(&t1);
+	} while (t1 - t0 < tl);
+	debugf(2, "inner: %d, outer: %d\n", cycles_inner, cycles_outer);
+	for (ri = 0; ri < runs; ri++) {
+		sys_precise_clock(&t0);
+		c = 0;
+		do {
+			c++;
+			for (i = 0; i < cycles_outer; i++)
+				busy_sse_loop(cycles_inner);
+			sys_precise_clock(&t1);
+		} while (t1 - t0 < tl * (uint64_t) 8);
+		// cpu_Hz = cycles_inner * cycles_outer * 256 / (t1 - t0) * 1000000
+		debugf(2, "c = %d, td = %llu\n", c, t1 - t0);
+		hz = (uint64_t) cycles_inner * (uint64_t) cycles_outer * (uint64_t) c * (uint64_t) 256000000 / (t1 - t0);
+		cur_value = (int) (hz / 1000000);
+		if (cur_value > max_value) max_value = cur_value;
+	}
+	return max_value;
+}
+
 int cpu_clock(void)
 {
 	int result;