OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove old t_vertex.c codegen infrastructure, tie in new code.

Browse Source 
Currently disabled, can enable with MESA_EXPERIMENTAL=t.
This commit is contained in:
Keith Whitwell
2005-05-18 15:26:48 +00:00
parent dc7fc17396
commit 2b2bd08589
9 changed files with 1410 additions and 1999 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/mesa/sources
View File

@@ -133,8 +133,8 @@ TNL_SOURCES = \
tnl/t_vb_points.c \
tnl/t_vp_build.c \
tnl/t_vertex.c \
tnl/t_vertex_c.c \
tnl/t_vertex_codegen.c \
tnl/t_vertex_sse.c \
tnl/t_vertex_generic.c \
tnl/t_vtx_api.c \
tnl/t_vtx_generic.c \
tnl/t_vtx_x86.c \

60
src/mesa/tnl/t_context.h
View File

@@ -536,6 +536,7 @@ struct tnl_clipspace_attr
GLuint vertattrsize; /* size of the attribute in bytes */
GLubyte *inputptr;
GLuint inputstride;
GLuint inputsize;
const tnl_insert_func *insert;
tnl_insert_func emit;
tnl_extract_func extract;
@@ -543,39 +544,6 @@ struct tnl_clipspace_attr
};
struct tnl_clipspace_codegen {
GLboolean (*emit_header)( struct tnl_clipspace_codegen *,
struct tnl_clipspace *);
GLboolean (*emit_footer)( struct tnl_clipspace_codegen * );
GLboolean (*emit_attr_header)( struct tnl_clipspace_codegen *,
struct tnl_clipspace_attr *,
GLint j, GLenum out_type,
GLboolean need_vp );
GLboolean (*emit_attr_footer)( struct tnl_clipspace_codegen * );
GLboolean (*emit_mov)( struct tnl_clipspace_codegen *,
GLint, GLint );
GLboolean (*emit_const)( struct tnl_clipspace_codegen *,
GLint, GLfloat );
GLboolean (*emit_mad)( struct tnl_clipspace_codegen *,
GLint, GLint, GLint, GLint );
GLboolean (*emit_float_to_chan)( struct tnl_clipspace_codegen *,
GLint, GLint );
GLboolean (*emit_const_chan)( struct tnl_clipspace_codegen *,
GLint, GLchan );
GLboolean (*emit_float_to_ubyte)( struct tnl_clipspace_codegen *,
GLint, GLint );
GLboolean (*emit_const_ubyte)( struct tnl_clipspace_codegen *,
GLint, GLubyte );
tnl_emit_func (*emit_store_func)( struct tnl_clipspace_codegen * );
struct _tnl_dynfn codegen_list;
char *buf;
int buf_size;
int buf_used;
int out_offset;
};
typedef void (*tnl_points_func)( GLcontext *ctx, GLuint first, GLuint last );
@@ -595,6 +563,20 @@ typedef void (*tnl_setup_func)( GLcontext *ctx,
GLuint new_inputs);
struct tnl_clipspace_fastpath {
GLuint vertex_size;
GLuint attr_count;
GLboolean match_strides;
struct {
GLuint format;
GLuint stride;
} *attr;
tnl_emit_func func;
struct tnl_clipspace_fastpath *next;
};
/**
* Used to describe conversion of vertex arrays to vertex structures.
* I.e. Structure of arrays to arrays of structs.
@@ -616,7 +598,17 @@ struct tnl_clipspace
tnl_interp_func interp;
tnl_copy_pv_func copy_pv;
struct tnl_clipspace_codegen codegen;
/* Parameters and constants for codegen:
*/
GLboolean need_viewport;
GLfloat vp_scale[4];
GLfloat vp_xlate[4];
GLfloat chan_scale[4];
GLfloat identity[4];
struct tnl_clipspace_fastpath *fastpath;
void (*codegen_emit)( GLcontext *ctx );
};

16
src/mesa/tnl/t_pipeline.c
View File

@@ -38,6 +38,7 @@
#include "t_context.h"
#include "t_pipeline.h"
#include "t_vp_build.h"
#include "t_vertex.h"
void _tnl_install_pipeline( GLcontext *ctx,
const struct tnl_pipeline_stage **stages )
@@ -93,7 +94,7 @@ static GLuint check_input_changes( GLcontext *ctx )
}
static void check_output_changes( GLcontext *ctx )
static GLuint check_output_changes( GLcontext *ctx )
{
#if 0
TNLcontext *tnl = TNL_CONTEXT(ctx);
@@ -109,6 +110,10 @@ static void check_output_changes( GLcontext *ctx )
if (tnl->pipeline.output_changes)
tnl->Driver.NotifyOutputChanges( ctx, tnl->pipeline.output_changes );
return tnl->pipeline.output_changes;
#else
return ~0;
#endif
}
@@ -138,10 +143,15 @@ void _tnl_run_pipeline( GLcontext *ctx )
tnl->pipeline.new_state = 0;
tnl->pipeline.input_changes = 0;
check_output_changes( ctx );
/* Pipeline can only change its output in response to either a
* statechange or an input size/stride change. No other changes
* are allowed.
*/
if (check_output_changes( ctx ))
_tnl_notify_pipeline_output_change( ctx );
}
START_FAST_MATH(__tmp);
for (i = 0; i < tnl->pipeline.nr_stages ; i++) {

1356
src/mesa/tnl/t_vertex.c
View File

File diff suppressed because it is too large Load Diff

68
src/mesa/tnl/t_vertex.h
View File

@@ -65,7 +65,15 @@ struct tnl_attr_map {
enum tnl_attr_format format;
GLuint offset;
};
struct tnl_format_info {
const char *name;
tnl_extract_func extract;
tnl_insert_func insert[4];
const GLuint attrsize;
};
extern const struct tnl_format_info _tnl_format_info[EMIT_MAX];
/* Interpolate between two vertices to produce a third:
@@ -95,18 +103,11 @@ extern void _tnl_set_attr( GLcontext *ctx, void *vout, GLenum attrib,
extern void *_tnl_get_vertex( GLcontext *ctx, GLuint nr );
/*
*/
extern GLuint _tnl_install_attrs( GLcontext *ctx,
const struct tnl_attr_map *map,
GLuint nr, const GLfloat *vp,
GLuint unpacked_size );
extern void _tnl_free_vertices( GLcontext *ctx );
extern void _tnl_init_vertices( GLcontext *ctx,
@@ -127,25 +128,42 @@ extern void _tnl_invalidate_vertices( GLcontext *ctx, GLuint newinputs );
extern void _tnl_invalidate_vertex_state( GLcontext *ctx, GLuint new_state );
extern tnl_emit_func _tnl_codegen_emit( GLcontext *ctx );
extern void _tnl_notify_pipeline_output_change( GLcontext *ctx );
#define REG_IN (0<<16)
#define REG_OUT (1<<16)
#define REG_VP (2<<16)
#define REG_TMP (3<<16)
#define REG_MASK (3<<16)
#define REG_OFFSET_MASK 0xffff
#define in( offset ) (REG_IN | (offset))
#define out( offset ) (REG_OUT | (offset))
#define vp( offset ) (REG_VP | (offset))
#define tmp( offset ) (REG_TMP | (offset))
extern void _tnl_init_c_codegen( struct tnl_clipspace_codegen *p );
extern void _tnl_free_c_codegen( struct tnl_clipspace_codegen *p );
#define GET_VERTEX_STATE(ctx) &(TNL_CONTEXT(ctx)->clipspace)
/* Internal function:
*/
void _tnl_register_fastpath( struct tnl_clipspace *vtx,
GLboolean match_strides );
/* t_vertex_generic.c -- Internal functions for t_vertex.c
*/
void _tnl_generic_copy_pv_extras( GLcontext *ctx,
GLuint dst, GLuint src );
void _tnl_generic_interp_extras( GLcontext *ctx,
GLfloat t,
GLuint dst, GLuint out, GLuint in,
GLboolean force_boundary );
void _tnl_generic_copy_pv( GLcontext *ctx, GLuint edst, GLuint esrc );
void _tnl_generic_interp( GLcontext *ctx,
GLfloat t,
GLuint edst, GLuint eout, GLuint ein,
GLboolean force_boundary );
void _tnl_generic_emit( GLcontext *ctx,
GLuint count,
GLubyte *v );
void _tnl_generate_hardwired_emit( GLcontext *ctx );
/* t_vertex_sse.c -- Internal functions for t_vertex.c
*/
void _tnl_generate_sse_emit( GLcontext *ctx );
#endif

267
src/mesa/tnl/t_vertex_c.c
View File

@@ -1,267 +0,0 @@
/*
* Copyright 2003 Tungsten Graphics, inc.
* 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
* on the rights to use, copy, modify, merge, publish, distribute, sub
* license, 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 (including the next
* paragraph) 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 NON-INFRINGEMENT. IN NO EVENT SHALL
* TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS 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.
*
* Authors:
* Keith Whitwell <keithw@tungstengraphics.com>
*/
#include "glheader.h"
#include "context.h"
#include "colormac.h"
#include "t_context.h"
#include "t_vertex.h"
#include "simple_list.h"
#ifdef WIN32
#define vsnprintf _vsnprintf
#endif
/* A version of code generation for t_clipspace_codegen.c which prints
* out 'c' code to implement the generated function. A useful
* debugging tool, and in concert with something like tcc or a
* library-ized gcc, could do the whole job.
*/
static GLboolean emit( struct tnl_clipspace_codegen *p,
const char *fmt,
... )
{
if (p->buf_used < p->buf_size) {
va_list ap;
va_start( ap, fmt );
p->buf_used += vsnprintf( p->buf + p->buf_used,
p->buf_size - p->buf_used,
fmt, ap );
va_end( ap );
}
return p->buf_used < p->buf_size;
}
static GLboolean print_header( struct tnl_clipspace_codegen *p,
struct tnl_clipspace *vtx )
{
p->buf_used = 0;
p->out_offset = 0;
return
emit(p,
"struct tnl_clipspace_attr\n"
"{\n"
" unsigned int attrib; \n"
" unsigned int format;\n"
" unsigned int vertoffset; \n"
" unsigned int vertattrsize; \n"
" char *inputptr;\n"
" unsigned int inputstride;\n"
" void *insert;\n"
" void *emit;\n"
" void * extract;\n"
" const float *vp; \n"
"};\n"
"\n"
) &&
emit(p,
"void emit_vertices( int start, int end, char *dest, \n"
" struct tnl_clipspace_attr *a) \n"
"{\n"
" int i;"
" for (i = start ; i < end ; i++, dest += %d) {\n",
vtx->vertex_size);
}
static GLboolean print_footer( struct tnl_clipspace_codegen *p )
{
return
emit(p,
" }\n"
"}\n"
);
}
static GLboolean emit_reg( struct tnl_clipspace_codegen *p, GLint reg )
{
int idx = reg & REG_OFFSET_MASK;
switch (reg & REG_MASK) {
case REG_IN: return emit(p, "in[%d]", idx);
case REG_VP: return emit(p, "vp[%d]", idx);
case REG_TMP: return emit(p, "temp[%d]", idx); /* not used? */
case REG_OUT: return emit(p, "out[%d]", idx);
}
return GL_FALSE;
}
static GLboolean print_mov( struct tnl_clipspace_codegen *p, GLint dest, GLint src )
{
return
emit(p, " ") &&
emit_reg(p, dest) &&
emit(p, " = ") &&
emit_reg(p, src) &&
emit(p, ";\n");
}
static GLboolean print_const( struct tnl_clipspace_codegen *p,
GLint dest, GLfloat c )
{
return
emit(p, " ") &&
emit_reg(p, dest) &&
emit(p, " = %g;\n", c);
}
static GLboolean print_const_chan( struct tnl_clipspace_codegen *p,
GLint dest, GLchan c )
{
return
emit(p, " ") &&
emit_reg(p, dest) &&
emit(p, " = ") &&
#if CHAN_TYPE == GL_FLOAT
emit(p, "%f", c) &&
#else
emit(p, "%d", c) &&
#endif
emit(p, ";\n");
}
static GLboolean print_const_ubyte( struct tnl_clipspace_codegen *p,
GLint dest, GLubyte c )
{
return
emit(p, " ") &&
emit_reg(p, dest) &&
emit(p, " = %x;\n", c);
}
static GLboolean print_mad( struct tnl_clipspace_codegen *p,
GLint dest, GLint src0, GLint src1, GLint src2 )
{
return
emit(p, " ") &&
emit_reg(p, dest) &&
emit(p, " = ") &&
emit_reg(p, src0) &&
emit(p, " * ") &&
emit_reg(p, src1) &&
emit(p, " + ") &&
emit_reg(p, src2) &&
emit(p, ";\n");
}
static GLboolean print_float_to_ubyte( struct tnl_clipspace_codegen *p,
GLint dest, GLint src )
{
return
emit(p, " ") &&
emit(p, "UNCLAMPED_FLOAT_TO_UBYTE(") &&
emit_reg(p, dest) &&
emit(p, ", ") &&
emit_reg(p, src) &&
emit(p, ");\n");
}
static GLboolean print_float_to_chan( struct tnl_clipspace_codegen *p,
GLint dest, GLint src )
{
return
emit(p, " ") &&
emit(p, "UNCLAMPED_FLOAT_TO_CHAN(") &&
emit_reg(p, dest) &&
emit(p, ", ") &&
emit_reg(p, src) &&
emit(p, ");\n");
}
static GLboolean print_attr_header( struct tnl_clipspace_codegen *p,
struct tnl_clipspace_attr *a,
GLint j,
GLenum out_type,
GLboolean need_vp)
{
char *out_type_str = "void";
switch(out_type) {
case GL_FLOAT: out_type_str = "float"; break;
case GL_UNSIGNED_BYTE: out_type_str = "unsigned char"; break;
case GL_UNSIGNED_SHORT: out_type_str = "unsigned short"; break;
}
return
emit(p, " {\n") &&
(need_vp ? emit(p, " const float *vp = a[%d].vp;\n", j) : 1) &&
emit(p, " %s *out = (%s *)(dest + %d);\n",
out_type_str, out_type_str, a[j].vertoffset) &&
emit(p, " const float *in = (const float *)a[%d].inputptr;\n",
j) &&
emit(p, " a[%d].inputptr += a[%d].inputstride;\n", j, j);
}
static GLboolean print_attr_footer( struct tnl_clipspace_codegen *p )
{
return emit(p, " }\n");
}
static tnl_emit_func print_store_func( struct tnl_clipspace_codegen *p )
{
fprintf(stderr, "print_store_func: emitted:\n%s\n", p->buf);
return NULL;
}
void _tnl_init_c_codegen( struct tnl_clipspace_codegen *p )
{
p->emit_header = print_header;
p->emit_footer = print_footer;
p->emit_attr_header = print_attr_header;
p->emit_attr_footer = print_attr_footer;
p->emit_mov = print_mov;
p->emit_const = print_const;
p->emit_mad = print_mad;
p->emit_float_to_chan = print_float_to_chan;
p->emit_const_chan = print_const_chan;
p->emit_float_to_ubyte = print_float_to_ubyte;
p->emit_const_ubyte = print_const_ubyte;
p->emit_store_func = print_store_func;
make_empty_list(&p->codegen_list);
p->buf_size = 2048;
p->buf = (char *) MALLOC(p->buf_size);
}
void _tnl_free_c_codegen( struct tnl_clipspace_codegen *p )
{
if (p->buf) {
FREE(p->buf);
p->buf = NULL;
}
}

510
src/mesa/tnl/t_vertex_codegen.c
View File

@@ -1,510 +0,0 @@
/*
* Copyright 2003 Tungsten Graphics, inc.
* 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
* on the rights to use, copy, modify, merge, publish, distribute, sub
* license, 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 (including the next
* paragraph) 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 NON-INFRINGEMENT. IN NO EVENT SHALL
* TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS 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.
*
* Authors:
* Keith Whitwell <keithw@tungstengraphics.com>
*/
#include "glheader.h"
#include "context.h"
#include "colormac.h"
#include "t_context.h"
#include "t_vertex.h"
#include "simple_list.h"
/* Another codegen scheme, hopefully portable to a few different
* architectures without too much work.
*/
static GLboolean emit_4f_viewport_4( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
p->emit_mad(p, out(2), vp(10), in(2), vp(14)) &&
p->emit_mov(p, out(3), in(3));
}
static GLboolean emit_4f_viewport_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
p->emit_mad(p, out(2), vp(10), in(2), vp(14)) &&
p->emit_const(p, out(3), 1.0);
}
static GLboolean emit_4f_viewport_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
p->emit_mov(p, out(2), vp(14)) &&
p->emit_const(p, out(3), 1.0);
}
static GLboolean emit_4f_viewport_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mov(p, out(1), vp(13)) &&
p->emit_mov(p, out(2), vp(14)) &&
p->emit_const(p, out(3), 1.0);
}
static GLboolean emit_3f_viewport_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
p->emit_mad(p, out(2), vp(10), in(2), vp(14));
}
static GLboolean emit_3f_viewport_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
p->emit_mov(p, out(2), vp(14));
}
static GLboolean emit_3f_viewport_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mov(p, out(1), vp(13)) &&
p->emit_mov(p, out(2), vp(14));
}
static GLboolean emit_2f_viewport_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
p->emit_mad(p, out(1), vp(5), in(1), vp(13));
}
static GLboolean emit_2f_viewport_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_mad(p, out(0), vp(0), in(0), vp(12));
}
static GLboolean emit_4f_4( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_mov(p, out(1), in(1)) &&
p->emit_mov(p, out(2), in(2)) &&
p->emit_mov(p, out(3), in(3));
}
static GLboolean emit_4f_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_mov(p, out(1), in(1)) &&
p->emit_mov(p, out(2), in(2)) &&
p->emit_const(p, out(3), 1.0);
}
static GLboolean emit_4f_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_mov(p, out(1), in(1)) &&
p->emit_const(p, out(2), 0.0) &&
p->emit_const(p, out(3), 1.0);
}
static GLboolean emit_4f_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_const(p, out(1), 0.0) &&
p->emit_const(p, out(2), 0.0) &&
p->emit_const(p, out(3), 1.0);
}
static GLboolean emit_3f_xyw_4( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_mov(p, out(1), in(1)) &&
p->emit_mov(p, out(2), in(3));
}
static GLboolean emit_3f_xyw_err( struct tnl_clipspace_codegen *p )
{
(void) p;
assert(0);
return GL_FALSE;
}
static GLboolean emit_3f_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_mov(p, out(1), in(1)) &&
p->emit_mov(p, out(2), in(2));
}
static GLboolean emit_3f_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_mov(p, out(1), in(1)) &&
p->emit_const(p, out(2), 0.0);
}
static GLboolean emit_3f_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_const(p, out(1), 0.0) &&
p->emit_const(p, out(2), 0.0);
}
static GLboolean emit_2f_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_mov(p, out(1), in(1));
}
static GLboolean emit_2f_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0)) &&
p->emit_const(p, out(1), 0.0);
}
static GLboolean emit_1f_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_mov(p, out(0), in(0));
}
static GLboolean emit_4chan_4f_rgba_4( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_chan(p, out(0), in(0)) &&
p->emit_float_to_chan(p, out(1), in(1)) &&
p->emit_float_to_chan(p, out(2), in(2)) &&
p->emit_float_to_chan(p, out(3), in(3));
}
static GLboolean emit_4chan_4f_rgba_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_chan(p, out(0), in(0)) &&
p->emit_float_to_chan(p, out(1), in(1)) &&
p->emit_float_to_chan(p, out(2), in(2)) &&
p->emit_const_chan(p, out(3), CHAN_MAX);
}
static GLboolean emit_4chan_4f_rgba_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_chan(p, out(0), in(0)) &&
p->emit_float_to_chan(p, out(1), in(1)) &&
p->emit_const_chan(p, out(2), 0) &&
p->emit_const_chan(p, out(3), CHAN_MAX);
}
static GLboolean emit_4chan_4f_rgba_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_chan(p, out(0), in(0)) &&
p->emit_const_chan(p, out(1), 0) &&
p->emit_const_chan(p, out(2), 0) &&
p->emit_const_chan(p, out(3), CHAN_MAX);
}
static GLboolean emit_4ub_4f_rgba_4( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_float_to_ubyte(p, out(2), in(2)) &&
p->emit_float_to_ubyte(p, out(3), in(3));
}
static GLboolean emit_4ub_4f_rgba_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_float_to_ubyte(p, out(2), in(2)) &&
p->emit_const_ubyte(p, out(3), 0xff);
}
static GLboolean emit_4ub_4f_rgba_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_const_ubyte(p, out(2), 0) &&
p->emit_const_ubyte(p, out(3), 0xff);
}
static GLboolean emit_4ub_4f_rgba_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0)) &&
p->emit_const_ubyte(p, out(1), 0) &&
p->emit_const_ubyte(p, out(2), 0) &&
p->emit_const_ubyte(p, out(3), 0xff);
}
static GLboolean emit_4ub_4f_bgra_4( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(2), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_float_to_ubyte(p, out(0), in(2)) &&
p->emit_float_to_ubyte(p, out(3), in(3));
}
static GLboolean emit_4ub_4f_bgra_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(2), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_float_to_ubyte(p, out(0), in(2)) &&
p->emit_const_ubyte(p, out(3), 0xff);
}
static GLboolean emit_4ub_4f_bgra_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(2), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_const_ubyte(p, out(0), 0) &&
p->emit_const_ubyte(p, out(3), 0xff);
}
static GLboolean emit_4ub_4f_bgra_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(2), in(0)) &&
p->emit_const_ubyte(p, out(1), 0) &&
p->emit_const_ubyte(p, out(0), 0) &&
p->emit_const_ubyte(p, out(3), 0xff);
}
static GLboolean emit_3ub_3f_rgb_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_float_to_ubyte(p, out(2), in(2));
}
static GLboolean emit_3ub_3f_rgb_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_const_ubyte(p, out(2), 0);
}
static GLboolean emit_3ub_3f_rgb_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0)) &&
p->emit_const_ubyte(p, out(1), 0) &&
p->emit_const_ubyte(p, out(2), 0);
}
static GLboolean emit_3ub_3f_bgr_3( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(2), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_float_to_ubyte(p, out(0), in(2));
}
static GLboolean emit_3ub_3f_bgr_2( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(2), in(0)) &&
p->emit_float_to_ubyte(p, out(1), in(1)) &&
p->emit_const_ubyte(p, out(0), 0);
}
static GLboolean emit_3ub_3f_bgr_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(2), in(0)) &&
p->emit_const_ubyte(p, out(1), 0) &&
p->emit_const_ubyte(p, out(0), 0);
}
static GLboolean emit_1ub_1f_1( struct tnl_clipspace_codegen *p )
{
return
p->emit_float_to_ubyte(p, out(0), in(0));
}
static struct {
const char *name;
GLenum out_type;
GLboolean need_vp;
GLboolean (*emit[4])( struct tnl_clipspace_codegen * );
} emit_info[EMIT_MAX] = {
{ "1f", GL_FLOAT, GL_FALSE,
{ emit_1f_1, emit_1f_1, emit_1f_1, emit_1f_1 } },
{ "2f", GL_FLOAT, GL_FALSE,
{ emit_2f_1, emit_2f_2, emit_2f_2, emit_2f_2 } },
{ "3f", GL_FLOAT, GL_FALSE,
{ emit_3f_1, emit_3f_2, emit_3f_3, emit_3f_3 } },
{ "4f", GL_FLOAT, GL_FALSE,
{ emit_4f_1, emit_4f_2, emit_4f_3, emit_4f_4 } },
{ "2f_viewport", GL_FLOAT, GL_TRUE,
{ emit_2f_viewport_1, emit_2f_viewport_2, emit_2f_viewport_2,
emit_2f_viewport_2 } },
{ "3f_viewport", GL_FLOAT, GL_TRUE,
{ emit_3f_viewport_1, emit_3f_viewport_2, emit_3f_viewport_3,
emit_3f_viewport_3 } },
{ "4f_viewport", GL_FLOAT, GL_TRUE,
{ emit_4f_viewport_1, emit_4f_viewport_2, emit_4f_viewport_3,
emit_4f_viewport_4 } },
{ "3f_xyw", GL_FLOAT, GL_FALSE,
{ emit_3f_xyw_err, emit_3f_xyw_err, emit_3f_xyw_err,
emit_3f_xyw_4 } },
{ "1ub_1f", GL_UNSIGNED_BYTE, GL_FALSE,
{ emit_1ub_1f_1, emit_1ub_1f_1, emit_1ub_1f_1, emit_1ub_1f_1 } },
{ "3ub_3f_rgb", GL_UNSIGNED_BYTE, GL_FALSE,
{ emit_3ub_3f_rgb_1, emit_3ub_3f_rgb_2, emit_3ub_3f_rgb_3,
emit_3ub_3f_rgb_3 } },
{ "3ub_3f_bgr", GL_UNSIGNED_BYTE, GL_FALSE,
{ emit_3ub_3f_bgr_1, emit_3ub_3f_bgr_2, emit_3ub_3f_bgr_3,
emit_3ub_3f_bgr_3 } },
{ "4ub_4f_rgba", GL_UNSIGNED_BYTE, GL_FALSE,
{ emit_4ub_4f_rgba_1, emit_4ub_4f_rgba_2, emit_4ub_4f_rgba_3,
emit_4ub_4f_rgba_4 } },
{ "4ub_4f_bgra", GL_UNSIGNED_BYTE, GL_FALSE,
{ emit_4ub_4f_bgra_1, emit_4ub_4f_bgra_2, emit_4ub_4f_bgra_3,
emit_4ub_4f_bgra_4 } },
{ "4chan_4f_rgba", CHAN_TYPE, GL_FALSE,
{ emit_4chan_4f_rgba_1, emit_4chan_4f_rgba_2, emit_4chan_4f_rgba_3,
emit_4chan_4f_rgba_4 } },
{ "pad", 0, 0,
{ NULL, NULL, NULL, NULL } }
};
/***********************************************************************
* list(attrib, size) --> function
*
* Because of the dependence of size, this all has to take place after
* the pipeline has been run.
*/
tnl_emit_func _tnl_codegen_emit( GLcontext *ctx )
{
struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
struct tnl_clipspace_attr *a = vtx->attr;
struct tnl_clipspace_codegen *p = &vtx->codegen;
const GLuint count = vtx->attr_count;
GLuint j;
/* Need a faster lookup, or is this linear scan of an MRU list good
* enough? MRU chosen based on the guess that consecutive VB's are
* likely to be of the same format. A hash of attributes and sizes
* might be a better technique.
*
* With the vtx code now in place, it should be possible to track
* changes to the sizes of input arrays (and state, of course) and
* only invalidate this function when those sizes have changed.
*/
#if 0
foreach (l, p->codegen_list) {
if (l->attr_count != count)
continue;
/* Assumptions:
* a[j].vp will not change for a given attrib
* a[j].vertex_offset will not change nothing else has changed.
*/
for (j = 0; j < count; j++)
if (a[j].attrib != l->a[j].attrib ||
a[j].sz != l->a[j].sz)
break;
if (j == count) {
move_to_head(l, p->codegen_list);
return l->func;
}
}
#endif
p->emit_header( p, vtx );
for (j = 0; j < count; j++) {
GLuint sz = VB->AttribPtr[a[j].attrib]->size - 1;
p->emit_attr_header( p, a, j,
emit_info[a[j].format].out_type,
emit_info[a[j].format].need_vp );
if (!emit_info[a[j].format].emit[sz]( p )) {
fprintf(stderr, "codegen failed\n");
return NULL;
}
p->emit_attr_footer( p );
}
p->emit_footer( p );
return p->emit_store_func( p );
}

1119
src/mesa/tnl/t_vertex_generic.c Normal file
View File

File diff suppressed because it is too large Load Diff

9
src/mesa/tnl/t_vertex_sse.c
View File

@@ -603,10 +603,8 @@ static void emit_load3f_3( struct x86_program *p,
struct x86_reg dest,
struct x86_reg arg0 )
{
/* Over-reads by 1 dword - potential SEGV... Deal with in
* array_cache by treating size-3 arrays specially, copying to
* temporary storage if last element (how can you tell?) falls on a
* 4k boundary.
/* Over-reads by 1 dword - potential SEGV if input is a vertex
* array.
*/
if (p->inputs_safe) {
emit_movups(p, dest, arg0);
@@ -1003,6 +1001,9 @@ void _tnl_generate_sse_emit( GLcontext *ctx )
_mesa_printf("disassemble 0x%x 0x%x\n", p.store, p.csr);
}
else {
/* Note the failure:
*/
_tnl_register_fastpath( vtx, GL_FALSE );
FREE(p.store);
}