Some more work on interal debugging, timing routines for things that

will have implementations in assembly code.  To come: texture image
conversions, more of internal T&L pipeline and so on.

1 files changed, 261 insertions, 0 deletions

diff --git a/src/mesa/math/m_debug_util.h b/src/mesa/math/m_debug_util.h
new file mode 100644
index 00000000000..e49f3615c72
--- /dev/null
+++ b/src/mesa/math/m_debug_util.h
@@ -0,0 +1,261 @@
+/* $Id: m_debug_util.h,v 1.1 2001/02/03 08:41:04 gareth Exp $ */
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Author:
+ *    Gareth Hughes <[email protected]>
+ */
+
+#ifndef __M_DEBUG_UTIL_H__
+#define __M_DEBUG_UTIL_H__
+
+
+#ifdef DEBUG  /* This code only used for debugging */
+
+
+/* Comment this out to deactivate the cycle counter.
+ * NOTE: it works only on CPUs which know the 'rdtsc' command (586 or higher)
+ * (hope, you don't try to debug Mesa on a 386 ;)
+ */
+#if defined(__GNUC__) && defined(__i386__) && defined(USE_X86_ASM)
+#define  RUN_DEBUG_BENCHMARK
+#endif
+
+#define TEST_COUNT		128	/* size of the tested vector array   */
+
+#define REQUIRED_PRECISION	10	/* allow 4 bits to miss              */
+#define MAX_PRECISION		24	/* max. precision possible           */
+
+
+#ifdef  RUN_DEBUG_BENCHMARK
+/* Overhead of profiling counter in cycles.  Automatically adjusted to
+ * your machine at run time - counter initialization should give very
+ * consistent results.
+ */
+extern long counter_overhead;
+
+/* This is the value of the environment variable MESA_PROFILE, and is
+ * used to determine if we should benchmark the functions as well as
+ * verify their correctness.
+ */
+extern char *mesa_profile;
+
+/* Modify the the number of tests if you like.
+ * We take the minimum of all results, because every error should be
+ * positive (time used by other processes, task switches etc).
+ * It is assumed that all calculations are done in the cache.
+ */
+
+#if 1 /* PPro, PII, PIII version */
+
+/* Profiling on the P6 architecture requires a little more work, due to
+ * the internal out-of-order execution.  We must perform a serializing
+ * 'cpuid' instruction before and after the 'rdtsc' instructions to make
+ * sure no other uops are executed when we sample the timestamp counter.
+ */
+#define  INIT_COUNTER()							\
+   do {									\
+      int cycle_i;							\
+      counter_overhead = LONG_MAX;					\
+      for ( cycle_i = 0 ; cycle_i < 8 ; cycle_i++ ) {			\
+	 long cycle_tmp1 = 0, cycle_tmp2 = 0;				\
+	 __asm__ __volatile__ ( "push %%ebx       \n"			\
+				"xor %%eax, %%eax \n"			\
+				"cpuid            \n"			\
+				"rdtsc            \n"			\
+				"mov %%eax, %0    \n"			\
+				"xor %%eax, %%eax \n"			\
+				"cpuid            \n"			\
+				"pop %%ebx        \n"			\
+				"push %%ebx       \n"			\
+				"xor %%eax, %%eax \n"			\
+				"cpuid            \n"			\
+				"rdtsc            \n"			\
+				"mov %%eax, %1    \n"			\
+				"xor %%eax, %%eax \n"			\
+				"cpuid            \n"			\
+				"pop %%ebx        \n"			\
+				: "=m" (cycle_tmp1), "=m" (cycle_tmp2)	\
+				: : "eax", "ecx", "edx" );		\
+	 if ( counter_overhead > (cycle_tmp2 - cycle_tmp1) ) {		\
+	    counter_overhead = cycle_tmp2 - cycle_tmp1;			\
+	 }								\
+      }									\
+   } while (0)
+
+#define  BEGIN_RACE(x)							\
+   x = LONG_MAX;							\
+   for ( cycle_i = 0 ; cycle_i < 10 ; cycle_i++ ) {			\
+      long cycle_tmp1 = 0, cycle_tmp2 = 0;				\
+      __asm__ __volatile__ ( "push %%ebx       \n"			\
+			     "xor %%eax, %%eax \n"			\
+			     "cpuid            \n"			\
+			     "rdtsc            \n"			\
+			     "mov %%eax, %0    \n"			\
+			     "xor %%eax, %%eax \n"			\
+			     "cpuid            \n"			\
+			     "pop %%ebx        \n"			\
+			     : "=m" (cycle_tmp1)			\
+			     : : "eax", "ecx", "edx" );
+
+#define END_RACE(x)							\
+      __asm__ __volatile__ ( "push %%ebx       \n"			\
+			     "xor %%eax, %%eax \n"			\
+			     "cpuid            \n"			\
+			     "rdtsc            \n"			\
+			     "mov %%eax, %0    \n"			\
+			     "xor %%eax, %%eax \n"			\
+			     "cpuid            \n"			\
+			     "pop %%ebx        \n"			\
+			     : "=m" (cycle_tmp2)			\
+			     : : "eax", "ecx", "edx" );			\
+      if ( x > (cycle_tmp2 - cycle_tmp1) ) {				\
+	 x = cycle_tmp2 - cycle_tmp1;					\
+      }									\
+   }									\
+   x -= counter_overhead;
+
+#else /* PPlain, PMMX version */
+
+/* To ensure accurate results, we stall the pipelines with the
+ * non-pairable 'cdq' instruction.  This ensures all the code being
+ * profiled is complete when the 'rdtsc' instruction executes.
+ */
+#define  INIT_COUNTER(x)						\
+   do {									\
+      int cycle_i;							\
+      x = LONG_MAX;							\
+      for ( cycle_i = 0 ; cycle_i < 32 ; cycle_i++ ) {			\
+	 long cycle_tmp1, cycle_tmp2, dummy;				\
+	 __asm__ ( "mov %%eax, %0" : "=a" (cycle_tmp1) );		\
+	 __asm__ ( "mov %%eax, %0" : "=a" (cycle_tmp2) );		\
+	 __asm__ ( "cdq" );						\
+	 __asm__ ( "cdq" );						\
+	 __asm__ ( "rdtsc" : "=a" (cycle_tmp1), "=d" (dummy) );		\
+	 __asm__ ( "cdq" );						\
+	 __asm__ ( "cdq" );						\
+	 __asm__ ( "rdtsc" : "=a" (cycle_tmp2), "=d" (dummy) );		\
+	 if ( x > (cycle_tmp2 - cycle_tmp1) )				\
+	    x = cycle_tmp2 - cycle_tmp1;				\
+      }									\
+   } while (0)
+
+#define  BEGIN_RACE(x)							\
+   x = LONG_MAX;							\
+   for ( cycle_i = 0 ; cycle_i < 16 ; cycle_i++ ) {			\
+      long cycle_tmp1, cycle_tmp2, dummy;				\
+      __asm__ ( "mov %%eax, %0" : "=a" (cycle_tmp1) );			\
+      __asm__ ( "mov %%eax, %0" : "=a" (cycle_tmp2) );			\
+      __asm__ ( "cdq" );						\
+      __asm__ ( "cdq" );						\
+      __asm__ ( "rdtsc" : "=a" (cycle_tmp1), "=d" (dummy) );
+
+
+#define END_RACE(x)							\
+      __asm__ ( "cdq" );						\
+      __asm__ ( "cdq" );						\
+      __asm__ ( "rdtsc" : "=a" (cycle_tmp2), "=d" (dummy) );		\
+      if ( x > (cycle_tmp2 - cycle_tmp1) )				\
+	 x = cycle_tmp2 - cycle_tmp1;					\
+   }									\
+   x -= counter_overhead;
+
+#endif
+
+#else
+
+#define BEGIN_RACE(x)
+#define END_RACE(x)
+
+#endif
+
+
+/* ================================================================
+ * Helper functions
+ */
+
+static GLfloat rnd( void )
+{
+   GLfloat f = (GLfloat)rand() / (GLfloat)RAND_MAX;
+   GLfloat gran = (GLfloat)(1 << 13);
+
+   f = (GLfloat)(GLint)(f * gran) / gran;
+
+   return f * 2.0 - 1.0;
+}
+
+static int significand_match( GLfloat a, GLfloat b )
+{
+   GLfloat d = a - b;
+   int a_ex, b_ex, d_ex;
+
+   if ( d == 0.0F ) {
+      return MAX_PRECISION;   /* Exact match */
+   }
+
+   if ( a == 0.0F || b == 0.0F ) {
+      /* It would probably be better to check if the
+       * non-zero number is denormalized and return
+       * the index of the highest set bit here.
+       */
+      return 0;
+   }
+
+   frexp( a, &a_ex );
+   frexp( b, &b_ex );
+   frexp( d, &d_ex );
+
+   if ( a_ex < b_ex ) {
+      return a_ex - d_ex;
+   } else {
+      return b_ex - d_ex;
+   }
+}
+
+enum { NIL = 0, ONE = 1, NEG = -1, VAR = 2 };
+
+static void init_matrix( GLfloat *m )
+{
+   m[0] = 63.0; m[4] = 43.0; m[ 8] = 29.0; m[12] = 43.0;
+   m[1] = 55.0; m[5] = 17.0; m[ 9] = 31.0; m[13] =  7.0;
+   m[2] = 44.0; m[6] =  9.0; m[10] =  7.0; m[14] =  3.0;
+   m[3] = 11.0; m[7] = 23.0; m[11] = 91.0; m[15] =  9.0;
+}
+
+
+/* Ensure our arrays are correctly aligned.
+ */
+#if defined(__GNUC__)
+#  define ALIGN16	__attribute__ ((aligned (16)))
+#elif defined(__MSC__)
+#  define ALIGN16	__declspec(align(16)) /* GH: Does this work? */
+#else
+#  warning "ALIGN16 will not 16-byte align!\n"
+#  define ALIGN16
+#endif
+
+
+#endif /* DEBUG */
+
+#endif /* __M_DEBUG_UTIL_H__ */