[Mplayerxp-cvslog] SF.net SVN: mplayerxp:[135] mplayerxp

[<nic...@us...>]

Revision: 135
          http://mplayerxp.svn.sourceforge.net/mplayerxp/?rev=135&view=rev
Author:   nickols_k
Date:     2010-01-29 17:57:20 +0000 (Fri, 29 Jan 2010)

Log Message:
-----------
gcc-related code-generation fixed + gcc-44 related workaround

Modified Paths:
--------------
    functions
    mplayerxp/configure
    mplayerxp/cpudetect.c
    mplayerxp/libvo/osd.c
    mplayerxp/postproc/dsp.c
    mplayerxp/postproc/dsp_accel.h
    mplayerxp/pvector/pvector_int_x86.h

Modified: functions
===================================================================
--- functions	2010-01-28 19:09:03 UTC (rev 134)
+++ functions	2010-01-29 17:57:20 UTC (rev 135)
@@ -321,31 +321,40 @@
     check_cmd $cc $CFLAGS "$@" -E -o $TMPO $TMPC
 }
 
-check_asm(){
+test_asm(){
     log check_asm "$@"
     name="$1"
     asm="$2"
     shift 2
-    check_cc "$@" <<EOF && enable $name || disable $name
+    check_cc "$@" <<EOF && return 0 || return 1
 int foo(void){
     asm volatile($asm);
 }
 EOF
 }
 
+check_asm(){
+    name="$1"
+    test_asm "$@" && enable $name || disable $name
+}
+
 check_ld(){
     log check_ld "$@"
     check_cc || return
     check_cmd $cc $LDFLAGS "$@" -o $TMPE $TMPO $extralibs
 }
 
-check_cflags(){
+test_cflags(){
     log check_cflags "$@"
-    check_cc "$@" <<EOF && add_cflags "$@"
+    check_cc "$@" <<EOF && return 0 || return 1
 int x;
 EOF
 }
 
+check_cflags(){
+    test_cflags "$@" && add_cflags "$@"
+}
+
 is_in_cflags(){
     value=$1
     shift
@@ -356,15 +365,19 @@
     return 1
 }
 
-check_ldflags(){
+test_ldflags(){
     log check_ldflags "$@"
-    check_ld "$@" <<EOF && add_ldflags "$@"
+    check_ld "$@" <<EOF && return 0 || return 1
 int main(void){
     return 0;
 }
 EOF
 }
 
+check_ldflags(){
+    test_ldflags "$@" && add_ldflags "$@"
+}
+
 check_header(){
     log check_header "$@"
     header=$1
@@ -877,6 +890,13 @@
 }
 
 # Display error message, flushes tempfile, exit
+warning() {
+  log "Warning: $@"
+  echo
+  echo "Warning: $@" >&2
+  echo >&2
+}
+
 die() {
   log "Error: $@"
   echo
@@ -1128,8 +1148,13 @@
 fi
 
 # agree with user-defined architecture/tuning here
+if enabled gcc44_workaround ; then
+is_in_cflags "-march" || check_cflags "-march=native"
+is_in_cflags "-mtune" || check_cflags "-mtune=native"
+else
 is_in_cflags "-march" || check_cflags $locarch
 is_in_cflags "-mtune" || check_cflags "-mtune=generic"
+fi
 if enabled profile || test $debug -gt 2 || enabled gcov; then
     add_cflags "-O2 -fno-builtin"
 if enabled profile ; then
@@ -1145,7 +1170,7 @@
 add_extralibs "-lgcov"
 fi
 else
-  add_cflags "-O3"
+  is_in_cflags "-O3" || add_cflags "-O3"
   mips   || check_cflags "-pipe"
   x86_32 && check_cflags "-fomit-frame-pointer"
 # -fomit-frame-pointer is not needed for x86_64
@@ -1155,12 +1180,14 @@
 is_in_cflags "-mfancy-math-387" || check_cflags "-mfancy-math-387"
 is_in_cflags "-fno-math-errno" || check_cflags "-fno-math-errno"
 is_in_cflags "-fno-signed-zeros" || check_cflags "-fno-signed-zeros"
-# depend on compiler version
-#check_cflags "-fno-tree-vectorize"
+# it seems that -mtune=generic doesn't prevent generation of SSE4 opcodes for i686 arch
+check_cflags "-fno-tree-vectorize"
+check_cflags "-fno-tree-vect-loop-version"
+check_cflags "-fno-vect-cost-model"
 
 disable mmx_test
 x86 && enable mmx_test
-
+enabled gcc44_workaround && disable mmx_test
 if enabled mmx_test ; then
 check_cflags -mmmx
 check_cflags -m3dnow

Modified: mplayerxp/configure
===================================================================
--- mplayerxp/configure	2010-01-28 19:09:03 UTC (rev 134)
+++ mplayerxp/configure	2010-01-29 17:57:20 UTC (rev 135)
@@ -238,6 +238,7 @@
   add_ldflags "-L/usr/bin"
 fi
 
+disable gcc44_workaround
 # Checking CC version...
 # gcc-3.0 merges optimizations coming from egcs, pgcc, agcc, ...
   echocheck "$cc version"
@@ -247,51 +248,16 @@
   cc_version=`$cc -dumpversion`
   cc_v=$cc_version
   case $cc_version in
-    '')
-      cc_version="v. ?.??, bad"
-      cc_verc_fail=yes
+    4.[4-9]|4.[4-9].[0-9])
+      enable gcc44_workaround
       ;;
-    3.2|3.2.[0-2])
-      cc_version="$cc_version, bad"
-      cc_verc_fail=yes
+    3.[3-9]|3.[2-9].[3-9]|4.[0-3]|4.[0-3].[0-9])
       ;;
-    3.[2-9]|3.[2-9].[0-9]|4.[0-9]|4.[0-9].[0-9])
-      cc_version="$cc_version, ok"
-      cc_verc_fail=no
-      ;;
     *)
-      cc_version="$cc_version, bad"
-      cc_verc_fail=yes
+      die "*** Bad gcc-$cc_version: Please upgrade C compiler at least to gcc-3.2.3+! ***"
       ;;
   esac
   echores "$cc_version"
-  test "$cc_verc_fail" = yes && disable fastcall
-  if enabled fastcall ; then
-  echocheck "__fastcall optimization abilities"
-  case $cc_v in
-    '')
-      fastcall=no
-      ;;
-    # avoid fastcall usage on gcc-2.95.2 and older
-    3.[2-9]|3.[2-9].[0-9]|4.[0-9].[0-9])
-      fastcall=yes
-      ;;
-    *)
-      fastcall=no
-      ;;
-  esac
-  echores "$fastcall"
-  fi
-
-  if test "$cc_verc_fail" = yes ; then
-    cat <<EOF
-
-*** Please upgrade C compiler to gcc-3.2.3+ version! ***
-EOF
-    die "Bad gcc version"
-  fi
-# ---
-
 # now that we know what compiler should be used for compilation, try to find
 # out which assembler is used by the $cc compiler
 if cygwin ; then
@@ -1012,6 +978,13 @@
 EOF
 fi
 
+if enabled gcc44_workaround ; then
+warning "*** gcc-4.4.x sometime generates AVX opcodes for -mavx key ignoring -fno-tree-vectorize option.
+That breaks current logic of mplayerxp configuring.
+It is stronly recommended to use this version of gcc for home purposes only!
+Please don't use this compiler to produce redistributable executables of mplayerxp!!! ***"
+fi
+
 # Last move:
 rmtmps
 

Modified: mplayerxp/cpudetect.c
===================================================================
--- mplayerxp/cpudetect.c	2010-01-28 19:09:03 UTC (rev 134)
+++ mplayerxp/cpudetect.c	2010-01-29 17:57:20 UTC (rev 135)
@@ -155,44 +155,6 @@
 		gCpuCaps.hasAVX,
 		gCpuCaps.hasFMA
 );
-
-		/* FIXME: Does SSE2 need more OS support, too? */
-#if defined(__linux__) || defined(__FreeBSD__)
-		if (caps->hasSSE)
-			check_os_katmai_support();
-		if (!caps->hasSSE)
-			caps->hasSSE2 = 0;
-#else
-		caps->hasSSE=0;
-		caps->hasSSE2 = 0;
-#endif
-
-#ifndef ARCH_X86_64
-#ifndef CAN_COMPILE_MMX
-	if(caps->hasMMX) MSG_WARN("MMX supported but disabled\n");
-	caps->hasMMX=0;
-#endif
-#ifndef CAN_COMPILE_MMX2
-	if(caps->hasMMX2) MSG_WARN("MMX2 supported but disabled\n");
-	caps->hasMMX2=0;
-#endif
-#ifndef CAN_COMPILE_SSE
-	if(caps->hasSSE) MSG_WARN("SSE supported but disabled\n");
-	caps->hasSSE=0;
-#endif
-#ifndef CAN_COMPILE_SSE2
-	if(caps->hasSSE2) MSG_WARN("SSE2 supported but disabled\n");
-	caps->hasSSE2=0;
-#endif
-#ifndef CAN_COMPILE_3DNOW
-	if(caps->has3DNow) MSG_WARN("3DNow supported but disabled\n");
-	caps->has3DNow=0;
-#endif
-#ifndef CAN_COMPILE_3DNOW2
-	if(caps->has3DNowExt) MSG_WARN("3DNowExt supported but disabled\n");
-	caps->has3DNowExt=0;
-#endif
-#endif
 }
 
 

Modified: mplayerxp/libvo/osd.c
===================================================================
--- mplayerxp/libvo/osd.c	2010-01-28 19:09:03 UTC (rev 134)
+++ mplayerxp/libvo/osd.c	2010-01-29 17:57:20 UTC (rev 135)
@@ -115,6 +115,17 @@
 #endif
 /*FIXME the optimized stuff is a lie for 15/16bpp as they arent optimized yet*/
 // ordered per speed fasterst first
+#ifdef __AVX__
+if(gCpuCaps.hasSSE41)
+{
+	MSG_V("Using SSE4 Optimized OnScreenDisplay\n");
+	vo_draw_alpha_yv12_ptr=vo_draw_alpha_yv12_AVX;
+	vo_draw_alpha_yuy2_ptr=vo_draw_alpha_yuy2_AVX;
+	vo_draw_alpha_rgb24_ptr=vo_draw_alpha_rgb24_AVX;
+	vo_draw_alpha_rgb32_ptr=vo_draw_alpha_rgb32_AVX;
+}
+else
+#endif
 #ifdef __SSE4_1__
 if(gCpuCaps.hasSSE41)
 {

Modified: mplayerxp/postproc/dsp.c
===================================================================
--- mplayerxp/postproc/dsp.c	2010-01-28 19:09:03 UTC (rev 134)
+++ mplayerxp/postproc/dsp.c	2010-01-29 17:57:20 UTC (rev 135)
@@ -666,6 +666,10 @@
 
 static void __FASTCALL__ init_change_bps(const void* in, void* out, unsigned len, unsigned inbps, unsigned outbps,int final)
 {
+#ifdef __AVX__
+	if(gCpuCaps.hasAVX) change_bps = change_bps_AVX;
+	else
+#endif
 #ifdef __SSE4_1__
 	if(gCpuCaps.hasSSE41) change_bps = change_bps_SSE4;
 	else
@@ -683,10 +687,6 @@
 	if(gCpuCaps.hasMMX2) change_bps = change_bps_SSE;
 	else
 #endif
-#ifdef __MMX__
-	if(gCpuCaps.hasMMX) change_bps = change_bps_MMX;
-	else
-#endif
 #endif /* __x86_64__ */
 	change_bps = change_bps_c;
 	(*change_bps)(in,out,len,inbps,outbps,final);
@@ -779,10 +779,6 @@
 	if(gCpuCaps.hasMMX2) FIR_i16 = FIR_i16_SSE;
 	else
 #endif
-#ifdef __MMX__
-	if(gCpuCaps.hasMMX) FIR_i16 = FIR_i16_MMX;
-	else
-#endif
 #endif /*__x86_64__*/
 	FIR_i16 = FIR_i16_c;
 	return (*FIR_i16)(x,w);

Modified: mplayerxp/postproc/dsp_accel.h
===================================================================
--- mplayerxp/postproc/dsp_accel.h	2010-01-28 19:09:03 UTC (rev 134)
+++ mplayerxp/postproc/dsp_accel.h	2010-01-29 17:57:20 UTC (rev 135)
@@ -276,7 +276,7 @@
 #endif
 static int32_t __FASTCALL__ PVECTOR_RENAME(FIR_i16)(int16_t *x,int16_t *w)
 {
-#ifdef OPTIMIZE_MMX
+#ifdef OPTIMIZE_SSE
     __m64 mm[8];
     mm[0] = _m_load(&w[0]);
     mm[1] = _m_load(&w[4]);
@@ -291,12 +291,8 @@
     mm[0] = _m_paddd(mm[4],mm[5]);
     mm[1] = _m_paddd(mm[6],mm[7]);
     mm[2] = _m_paddd(mm[0],mm[1]);
-#ifdef OPTIMIZE_SSE
     mm[0] = _m_pshufw(mm[2],0xFE);
-#else
-    mm[0] = mm[2];
-    mm[0] = _m_psrlqi(mm[0],32);
-#endif
+
     mm[0] = _m_paddd(mm[0],mm[2]);
     mm[0] = _m_psrldi(mm[0],16);
     return _mm_cvtsi64_si32(mm[0]);

Modified: mplayerxp/pvector/pvector_int_x86.h
===================================================================
--- mplayerxp/pvector/pvector_int_x86.h	2010-01-28 19:09:03 UTC (rev 134)
+++ mplayerxp/pvector/pvector_int_x86.h	2010-01-29 17:57:20 UTC (rev 135)
@@ -2,10 +2,10 @@
   pvector_int_x86.h
 */
 
-//#if defined( OPTIMIZE_AVX )
-//#define _VEC(a) a ## _AVX
-//#include <immintrin.h>
-#if defined(OPTIMIZE_AES)
+#if defined( OPTIMIZE_AVX )
+#define _VEC(a) a ## _AVX
+#include <immintrin.h>
+#elif defined(OPTIMIZE_AES)
 #include <wmmintrin.h>
 #elif defined (OPTIMIZE_SSE4)
 #include <smmintrin.h>


This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.