spantmp2.h

Mesa is an open-source implementation of the OpenGL specification - a system for rendering interacti

		        WRITE_PIXEL( x1, y, p );
	       }
	       else
	       {
	          for (;n1>0;i++,x1++,n1--)
		     WRITE_PIXEL( x1, y, p );
	       }
	    }
	 HW_ENDCLIPLOOP();
      }
   HW_WRITE_UNLOCK();
}


static void TAG(WriteMonoRGBAPixels)( GLcontext *ctx,
                                      struct gl_renderbuffer *rb,
				      GLuint n,
				      const GLint x[], const GLint y[],
				      const void *value,
				      const GLubyte mask[] ) 
{
   HW_WRITE_LOCK()
      {
         const GLubyte *color = (const GLubyte *) value;
	 GLint i;
	 LOCAL_VARS;
	 INIT_MONO_PIXEL(p, color);

	 if (DBG) fprintf(stderr, "WriteMonoRGBAPixels\n");

	 HW_WRITE_CLIPLOOP()
	    {
	       if (mask)
	       {
		  for (i=0;i<n;i++)
		     if (mask[i]) {
			int fy = Y_FLIP(y[i]);
			if (CLIPPIXEL( x[i], fy ))
			   WRITE_PIXEL( x[i], fy, p );
		     }
	       }
	       else
	       {
		  for (i=0;i<n;i++) {
		     int fy = Y_FLIP(y[i]);
		     if (CLIPPIXEL( x[i], fy ))
			WRITE_PIXEL( x[i], fy, p );
		  }
	       }
	    }
	 HW_ENDCLIPLOOP();
      }
   HW_WRITE_UNLOCK();
}


static void TAG(ReadRGBASpan)( GLcontext *ctx,
                               struct gl_renderbuffer *rb,
			       GLuint n, GLint x, GLint y, void *values)
{
   HW_READ_LOCK()
      {
         GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
	 GLint x1,n1;
	 LOCAL_VARS;

	 y = Y_FLIP(y);

	 if (DBG) fprintf(stderr, "ReadRGBASpan\n");

	 HW_READ_CLIPLOOP()
	    {
	       GLint i = 0;
	       CLIPSPAN(x,y,n,x1,n1,i);
	       for (;n1>0;i++,x1++,n1--)
		  READ_RGBA( rgba[i], x1, y );
	    }
         HW_ENDCLIPLOOP();
      }
   HW_READ_UNLOCK();
}


#if defined(USE_MMX_ASM) && \
   (((SPANTMP_PIXEL_FMT == GL_BGRA) && \
	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)) || \
    ((SPANTMP_PIXEL_FMT == GL_RGB) && \
	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_SHORT_5_6_5)))
static void TAG2(ReadRGBASpan,_MMX)( GLcontext *ctx,
                                     struct gl_renderbuffer *rb,
                                     GLuint n, GLint x, GLint y, void *values)
{
#ifndef USE_INNER_EMMS
   /* The EMMS instruction is directly in-lined here because using GCC's
    * built-in _mm_empty function was found to utterly destroy performance.
    */
   __asm__ __volatile__( "emms" );
#endif

   HW_READ_LOCK()
     {
        GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
	GLint x1,n1;
	LOCAL_VARS;

	y = Y_FLIP(y);

	if (DBG) fprintf(stderr, "ReadRGBASpan\n");

	HW_READ_CLIPLOOP()
	  {
	     GLint i = 0;
	     CLIPSPAN(x,y,n,x1,n1,i);

	       {
		  const void * src = GET_PTR( x1, y );
#if (SPANTMP_PIXEL_FMT == GL_RGB) && \
		  (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_SHORT_5_6_5)
		  _generic_read_RGBA_span_RGB565_MMX( src, rgba[i], n1 );
#else
		  _generic_read_RGBA_span_BGRA8888_REV_MMX( src, rgba[i], n1 );
#endif
	       }
	  }
	HW_ENDCLIPLOOP();
     }
   HW_READ_UNLOCK();
#ifndef USE_INNER_EMMS
   __asm__ __volatile__( "emms" );
#endif
}
#endif


#if defined(USE_SSE_ASM) && \
   (SPANTMP_PIXEL_FMT == GL_BGRA) && \
     (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
static void TAG2(ReadRGBASpan,_SSE2)( GLcontext *ctx,
                                      struct gl_renderbuffer *rb,
                                      GLuint n, GLint x, GLint y,
                                      void *values)
{
   HW_READ_LOCK()
     {
        GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
	GLint x1,n1;
	LOCAL_VARS;

	y = Y_FLIP(y);

	if (DBG) fprintf(stderr, "ReadRGBASpan\n");

	HW_READ_CLIPLOOP()
	  {
	     GLint i = 0;
	     CLIPSPAN(x,y,n,x1,n1,i);

	       {
		  const void * src = GET_PTR( x1, y );
		  _generic_read_RGBA_span_BGRA8888_REV_SSE2( src, rgba[i], n1 );
	       }
	  }
	HW_ENDCLIPLOOP();
     }
   HW_READ_UNLOCK();
}
#endif

#if defined(USE_SSE_ASM) && \
   (SPANTMP_PIXEL_FMT == GL_BGRA) && \
     (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
static void TAG2(ReadRGBASpan,_SSE)( GLcontext *ctx,
                                     struct gl_renderbuffer *rb,
                                     GLuint n, GLint x, GLint y,
                                     void *values)
{
#ifndef USE_INNER_EMMS
   /* The EMMS instruction is directly in-lined here because using GCC's
    * built-in _mm_empty function was found to utterly destroy performance.
    */
   __asm__ __volatile__( "emms" );
#endif

   HW_READ_LOCK()
     {
        GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
	GLint x1,n1;
	LOCAL_VARS;

	y = Y_FLIP(y);

	if (DBG) fprintf(stderr, "ReadRGBASpan\n");

	HW_READ_CLIPLOOP()
	  {
	     GLint i = 0;
	     CLIPSPAN(x,y,n,x1,n1,i);

	       {
		  const void * src = GET_PTR( x1, y );
		  _generic_read_RGBA_span_BGRA8888_REV_SSE( src, rgba[i], n1 );
	       }
	  }
	HW_ENDCLIPLOOP();
     }
   HW_READ_UNLOCK();
#ifndef USE_INNER_EMMS
   __asm__ __volatile__( "emms" );
#endif
}
#endif


static void TAG(ReadRGBAPixels)( GLcontext *ctx,
                                 struct gl_renderbuffer *rb,
				 GLuint n, const GLint x[], const GLint y[],
				 void *values )
{
   HW_READ_LOCK()
      {
         GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
         GLubyte *mask = NULL; /* remove someday */
	 GLint i;
	 LOCAL_VARS;

	 if (DBG) fprintf(stderr, "ReadRGBAPixels\n");

	 HW_READ_CLIPLOOP()
	    {
	       if (mask)
	       {
		  for (i=0;i<n;i++)
		     if (mask[i]) {
			int fy = Y_FLIP( y[i] );
			if (CLIPPIXEL( x[i], fy ))
			   READ_RGBA( rgba[i], x[i], fy );
		     }
	       }
	       else
	       {
		  for (i=0;i<n;i++) {
		     int fy = Y_FLIP( y[i] );
		     if (CLIPPIXEL( x[i], fy ))
			READ_RGBA( rgba[i], x[i], fy );
		  }
	       }
	    }
	 HW_ENDCLIPLOOP();
      }
   HW_READ_UNLOCK();
}

static void TAG(InitPointers)(struct gl_renderbuffer *rb)
{
   rb->PutRow = TAG(WriteRGBASpan);
   rb->PutRowRGB = TAG(WriteRGBSpan);
   rb->PutMonoRow = TAG(WriteMonoRGBASpan);
   rb->PutValues = TAG(WriteRGBAPixels);
   rb->PutMonoValues = TAG(WriteMonoRGBAPixels);
   rb->GetValues = TAG(ReadRGBAPixels);

#if defined(USE_SSE_ASM) && \
   (SPANTMP_PIXEL_FMT == GL_BGRA) && \
     (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
   if ( cpu_has_xmm2 ) {
      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "SSE2" );
      rb->GetRow = TAG2(ReadRGBASpan, _SSE2);
   }
   else
#endif
#if defined(USE_SSE_ASM) && \
   (SPANTMP_PIXEL_FMT == GL_BGRA) && \
     (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
   if ( cpu_has_xmm ) {
      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "SSE" );
      rb->GetRow = TAG2(ReadRGBASpan, _SSE);
   }
   else
#endif
#if defined(USE_MMX_ASM) && \
   (((SPANTMP_PIXEL_FMT == GL_BGRA) && \
	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)) || \
    ((SPANTMP_PIXEL_FMT == GL_RGB) && \
	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_SHORT_5_6_5)))
   if ( cpu_has_mmx ) {
      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "MMX" );
      rb->GetRow = TAG2(ReadRGBASpan, _MMX);
   }
   else
#endif
   {
      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "C" );
      rb->GetRow = TAG(ReadRGBASpan);
   }

}


#undef INIT_MONO_PIXEL
#undef WRITE_PIXEL
#undef WRITE_RGBA
#undef READ_RGBA
#undef TAG
#undef TAG2
#undef GET_PTR
#undef SPANTMP_PIXEL_FMT
#undef SPANTMP_PIXEL_TYPE