1 /**************************************************************************
3 Copyright (C) 2004 Nicolai Haehnle.
7 Permission is hereby granted, free of charge, to any person obtaining a
8 copy of this software and associated documentation files (the "Software"),
9 to deal in the Software without restriction, including without limitation
10 on the rights to use, copy, modify, merge, publish, distribute, sub
11 license, and/or sell copies of the Software, and to permit persons to whom
12 the Software is furnished to do so, subject to the following conditions:
14 The above copyright notice and this permission notice (including the next
15 paragraph) shall be included in all copies or substantial portions of the
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Nicolai Haehnle <prefect_@gmx.net>
40 #include "simple_list.h"
42 #include "api_arrayelt.h"
43 #include "swrast/swrast.h"
44 #include "swrast_setup/swrast_setup.h"
45 #include "array_cache/acache.h"
48 #include "radeon_reg.h"
49 #include "radeon_macros.h"
50 #include "radeon_ioctl.h"
51 #include "radeon_state.h"
52 #include "r300_context.h"
53 #include "r300_ioctl.h"
54 #include "r300_state.h"
56 #include "r300_program.h"
59 #include "r300_emit.h"
61 /**********************************************************************
62 * Hardware rasterization
64 * When we fell back to software TCL, we still try to use the
65 * rasterization hardware for rendering.
66 **********************************************************************/
68 static int r300_get_primitive_type(r300ContextPtr rmesa
,
74 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
75 struct vertex_buffer
*VB
= &tnl
->vb
;
77 int type
=-1, min_vertices
=0;
80 if(end
<=start
)return -1; /* do we need to watch for this ? */
82 switch (prim
& PRIM_MODE_MASK
) {
85 type
=R300_VAP_VF_CNTL__PRIM_POINTS
;
90 type
=R300_VAP_VF_CNTL__PRIM_LINES
;
95 type
=R300_VAP_VF_CNTL__PRIM_LINE_STRIP
;
105 type
=R300_VAP_VF_CNTL__PRIM_TRIANGLES
;
108 case GL_TRIANGLE_STRIP
:
110 type
=R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP
;
113 case GL_TRIANGLE_FAN
:
115 type
=R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN
;
120 type
=R300_VAP_VF_CNTL__PRIM_QUADS
;
125 type
=R300_VAP_VF_CNTL__PRIM_QUAD_STRIP
;
130 type
=R300_VAP_VF_CNTL__PRIM_POLYGON
;
134 fprintf(stderr
, "%s:%s Do not know how to handle primitive %02x - help me !\n",
135 __FILE__
, __FUNCTION__
,
136 prim
& PRIM_MODE_MASK
);
141 fprintf(stderr
, "[%d-%d]%s ", start
, end
, name
);
143 if(start
+min_vertices
>end
){
144 static int warn_once
=1;
146 fprintf(stderr
, "%s:%s ***WARN_ONCE*** Not enough vertices to draw primitive %02x - help me !\n",
147 __FILE__
, __FUNCTION__
,
148 prim
& PRIM_MODE_MASK
);
156 /* This function compiles GL context into state registers that
157 describe data routing inside of R300 pipeline.
159 In particular, it programs input_route, output_vtx_fmt, texture
160 unit configuration and gb_output_vtx_fmt
162 This function encompasses setup_AOS() from r300_lib.c
168 /* Immediate implementation - vertex data is sent via command stream */
170 static GLfloat default_vector
[4]={0.0, 0.0, 0.0, 1.0};
172 #define output_vector(v, i) \
175 for(_i=0;_i<v->size;_i++){ \
176 efloat(VEC_ELT(v, GLfloat, i)[_i]); \
178 for(_i=v->size;_i<4;_i++){ \
179 efloat(default_vector[_i]); \
183 /* Immediate implementation - vertex data is sent via command stream */
185 static void r300_render_immediate_primitive(r300ContextPtr rmesa
,
191 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
192 struct vertex_buffer
*VB
= &tnl
->vb
;
197 type
=r300_get_primitive_type(rmesa
, ctx
, start
, end
, prim
);
200 fprintf(stderr
,"ObjPtr: size=%d stride=%d\n",
201 VB
->ObjPtr
->size
, VB
->ObjPtr
->stride
);
202 fprintf(stderr
,"ColorPtr[0]: size=%d stride=%d\n",
203 VB
->ColorPtr
[0]->size
, VB
->ColorPtr
[0]->stride
);
204 fprintf(stderr
,"TexCoordPtr[0]: size=%d stride=%d\n",
205 VB
->TexCoordPtr
[0]->size
, VB
->TexCoordPtr
[0]->stride
);
211 WARN_ONCE("FIXME: Don't know how to handle GL_ARB_vertex_buffer_object correctly\n");
214 /* A packet cannot have more than 16383 data words.. */
215 if(((end
-start
)*4*rmesa
->state
.aos_count
)>16380){
216 WARN_ONCE("Too many vertices to paint. Fix me !\n");
220 //fprintf(stderr, "aos_count=%d start=%d end=%d\n", rmesa->state.aos_count, start, end);
222 if(rmesa
->state
.aos_count
==0){
223 WARN_ONCE("Aeiee ! aos_count==0, while it shouldn't. Skipping rendering\n");
227 start_immediate_packet(end
-start
, type
, 4*rmesa
->state
.aos_count
);
229 for(i
=start
;i
<end
;i
++){
231 fprintf(stderr
, "* (%f %f %f %f) (%f %f %f %f)\n",
232 VEC_ELT(VB
->ObjPtr
, GLfloat
, i
)[0],
233 VEC_ELT(VB
->ObjPtr
, GLfloat
, i
)[1],
234 VEC_ELT(VB
->ObjPtr
, GLfloat
, i
)[2],
235 VEC_ELT(VB
->ObjPtr
, GLfloat
, i
)[3],
237 VEC_ELT(VB
->ColorPtr
[0], GLfloat
, i
)[0],
238 VEC_ELT(VB
->ColorPtr
[0], GLfloat
, i
)[1],
239 VEC_ELT(VB
->ColorPtr
[0], GLfloat
, i
)[2],
240 VEC_ELT(VB
->ColorPtr
[0], GLfloat
, i
)[3]
246 if(tnl
->render_inputs
& _TNL_BIT_POS
)
247 output_vector(VB
->ObjPtr
, i
);
248 if(tnl
->render_inputs
& _TNL_BIT_NORMAL
)
249 output_vector(VB
->NormalPtr
, i
);
251 /* color components */
252 if(tnl
->render_inputs
& _TNL_BIT_COLOR0
)
253 output_vector(VB
->ColorPtr
[0], i
);
254 if(tnl
->render_inputs
& _TNL_BIT_COLOR1
)
255 output_vector(VB
->SecondaryColorPtr
[0], i
);
257 if(tnl
->render_inputs
& _TNL_BIT_FOG
)
258 output_vector(VB
->FogCoordPtr
, i
);
260 /* texture coordinates */
261 for(k
=0;k
< ctx
->Const
.MaxTextureUnits
;k
++)
262 if(tnl
->render_inputs
& (_TNL_BIT_TEX0
<<k
))
263 output_vector(VB
->TexCoordPtr
[k
], i
);
265 if(tnl
->render_inputs
& _TNL_BIT_INDEX
)
266 output_vector(VB
->IndexPtr
[0], i
);
267 if(tnl
->render_inputs
& _TNL_BIT_POINTSIZE
)
268 output_vector(VB
->PointSizePtr
, i
);
274 static GLboolean
r300_run_immediate_render(GLcontext
*ctx
,
275 struct tnl_pipeline_stage
*stage
)
277 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
278 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
279 struct vertex_buffer
*VB
= &tnl
->vb
;
281 /* Only do 2d textures */
282 struct gl_texture_object
*to
=ctx
->Texture
.Unit
[0].Current2D
;
283 r300TexObjPtr t
=to
->DriverData
;
287 /* Update texture state - needs to be done only when actually changed..
288 All the time for now.. */
291 if (RADEON_DEBUG
== DEBUG_PRIMS
)
292 fprintf(stderr
, "%s\n", __FUNCTION__
);
294 #if 1 /* we need this, somehow */
295 /* Flush state - make sure command buffer is nice and large */
297 /* Make sure we have enough space */
299 /* Count is very imprecize, but should be good upper bound */
300 r300EnsureCmdBufSpace(rmesa
, rmesa
->hw
.max_state_size
+ 4+2+30
301 +VB
->PrimitiveCount
*(1+8)+VB
->Count
*4*rmesa
->state
.texture
.tc_count
+4, __FUNCTION__
);
304 /* needed before starting 3d operation .. */
305 reg_start(R300_RB3D_DSTCACHE_CTLSTAT
,0);
312 #if 0 /* looks like the Z offset issue got fixed */
313 rmesa
->hw
.vte
.cmd
[1] = R300_VPORT_X_SCALE_ENA
314 | R300_VPORT_X_OFFSET_ENA
315 | R300_VPORT_Y_SCALE_ENA
316 | R300_VPORT_Y_OFFSET_ENA
318 R300_STATECHANGE(rmesa
, vte
);
323 /* Magic register - note it is right after 20b0 */
326 if(rmesa
->state
.texture
.tc_count
>0){
332 r300EmitState(rmesa
);
335 reg_start(R300_RB3D_COLORMASK
, 0);
338 vsf_start_fragment(0x406, 4);
344 vsf_start_fragment(0x400, 4);
351 /* We need LOAD_VBPNTR to setup AOS_ATTR fields.. the offsets are irrelevant */
352 r300EmitLOAD_VBPNTR(rmesa
, 0);
354 for(i
=0; i
< VB
->PrimitiveCount
; i
++){
355 GLuint prim
= VB
->Primitive
[i
].mode
;
356 GLuint start
= VB
->Primitive
[i
].start
;
357 GLuint length
= VB
->Primitive
[i
].count
;
358 r300_render_immediate_primitive(rmesa
, ctx
, start
, start
+ length
, prim
);
361 /* This sequence is required after any 3d drawing packet
362 I suspect it work arounds a bug (or deficiency) in hardware */
364 reg_start(R300_RB3D_DSTCACHE_CTLSTAT
,0);
374 /* vertex buffer implementation */
376 /* We use the start part of GART texture buffer for vertices */
379 static void upload_vertex_buffer(r300ContextPtr rmesa
, GLcontext
*ctx
)
381 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
382 struct vertex_buffer
*VB
= &tnl
->vb
;
385 radeonScreenPtr rsp
=rmesa
->radeon
.radeonScreen
;
387 /* A hack - we don't want to overwrite vertex buffers, so we
388 just use AGP space for them.. Fix me ! */
390 if(offset
>2*1024*1024){
391 //fprintf(stderr, "Wrapping agp vertex buffer offset\n");
394 /* Not the most efficient implementation, but, for now, I just want something that
396 /* to do - make single memcpy per column (is it possible ?) */
397 /* to do - use dirty flags to avoid redundant copies */
398 #define UPLOAD_VECTOR(v)\
400 /* Is the data dirty ? */ \
401 if (v->flags & ((1<<v->size)-1)) { \
402 /* fprintf(stderr, "size=%d vs stride=%d\n", v->size, v->stride); */ \
403 if(v->size*4==v->stride){\
405 memcpy(rsp->gartTextures.map+offset, v->data, v->stride*VB->Count); \
407 for(i=0;i<VB->Count;i++){ \
408 /* copy one vertex at a time*/ \
409 memcpy(rsp->gartTextures.map+offset+i*v->size*4, VEC_ELT(v, GLfloat, i), v->size*4); \
412 /* v->flags &= ~((1<<v->size)-1);*/ \
414 rmesa->state.aos[idx].offset=rsp->gartTextures.handle+offset; \
415 offset+=v->size*4*VB->Count; \
419 UPLOAD_VECTOR(VB
->ObjPtr
);
420 UPLOAD_VECTOR(VB
->ColorPtr
[0]);
421 /* texture coordinates */
422 for(k
=0;k
< ctx
->Const
.MaxTextureUnits
;k
++)
423 if(ctx
->Texture
.Unit
[k
].Enabled
)
424 UPLOAD_VECTOR(VB
->TexCoordPtr
[k
]);
426 if(idx
>=R300_MAX_AOS_ARRAYS
){
427 fprintf(stderr
, "Aieee ! Maximum AOS arrays count exceeded.. \n");
432 static void r300_render_vb_primitive(r300ContextPtr rmesa
,
441 if(end
<=start
)return; /* do we need to watch for this ? */
443 type
=r300_get_primitive_type(rmesa
, ctx
, start
, end
, prim
);
446 fire_AOS(PASS_PREFIX end
-start
, type
);
449 static GLboolean
r300_run_vb_render(GLcontext
*ctx
,
450 struct tnl_pipeline_stage
*stage
)
452 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
453 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
454 struct vertex_buffer
*VB
= &tnl
->vb
;
458 if (RADEON_DEBUG
== DEBUG_PRIMS
)
459 fprintf(stderr
, "%s\n", __FUNCTION__
);
462 reg_start(R300_RB3D_DSTCACHE_CTLSTAT
,0);
468 r300_setup_routing(ctx
, GL_FALSE
);
470 r300EmitState(rmesa
);
472 /* setup array of structures data */
473 LOCK_HARDWARE(&(rmesa
->radeon
));
475 upload_vertex_buffer(rmesa
, ctx
);
476 //fprintf(stderr, "Using %d AOS arrays\n", n_arrays);
478 for(i
=0; i
< VB
->PrimitiveCount
; i
++){
479 GLuint prim
= VB
->Primitive
[i
].mode
;
480 GLuint start
= VB
->Primitive
[i
].start
;
481 GLuint length
= VB
->Primitive
[i
].count
;
483 /* We need LOAD_VBPNTR to setup AOS_ATTR fields.. */
484 r300EmitLOAD_VBPNTR(rmesa
, start
);
486 r300_render_vb_primitive(rmesa
, ctx
, start
, start
+ length
, prim
);
489 /* This sequence is required after any 3d drawing packet
490 I suspect it works around a bug (or deficiency) in hardware */
492 reg_start(R300_RB3D_DSTCACHE_CTLSTAT
,0);
498 end_3d(PASS_PREFIX_VOID
);
500 /* Flush state - we are done drawing.. */
501 r300FlushCmdBufLocked(ctx
, __FUNCTION__
);
502 radeonWaitForIdleLocked(&(rmesa
->radeon
));
504 UNLOCK_HARDWARE(&(rmesa
->radeon
));
510 * Called by the pipeline manager to render a batch of primitives.
511 * We can return true to pass on to the next stage (i.e. software
512 * rasterization) or false to indicate that the pipeline has finished
513 * after we render something.
515 static GLboolean
r300_run_render(GLcontext
*ctx
,
516 struct tnl_pipeline_stage
*stage
)
518 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
519 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
520 struct vertex_buffer
*VB
= &tnl
->vb
;
523 if (RADEON_DEBUG
== DEBUG_PRIMS
)
524 fprintf(stderr
, "%s\n", __FUNCTION__
);
530 return r300_run_immediate_render(ctx
, stage
);
532 return r300_run_vb_render(ctx
, stage
);
539 mgaContextPtr mmesa
= MGA_CONTEXT(ctx
);
540 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
541 struct vertex_buffer
*VB
= &tnl
->vb
;
544 /* Don't handle clipping or indexed vertices or vertex manipulations.
546 if (mmesa
->RenderIndex
!= 0 ||
547 !mga_validate_render( ctx
, VB
)) {
551 tnl
->Driver
.Render
.Start( ctx
);
552 mmesa
->SetupNewInputs
= ~0;
554 for (i
= 0 ; i
< VB
->PrimitiveCount
; i
++)
556 GLuint prim
= VB
->Primitive
[i
].mode
;
557 GLuint start
= VB
->Primitive
[i
].start
;
558 GLuint length
= VB
->Primitive
[i
].count
;
563 mga_render_tab_verts
[prim
& PRIM_MODE_MASK
]( ctx
, start
, start
+ length
,
567 tnl
->Driver
.Render
.Finish( ctx
);
569 return GL_FALSE
; /* finished the pipe */
575 * Called by the pipeline manager once before rendering.
576 * We check the GL state here to
577 * a) decide whether we can do the current state in hardware and
578 * b) update hardware registers
580 #define FALLBACK_IF(expr) \
583 if (1 || RADEON_DEBUG & DEBUG_FALLBACKS) \
584 fprintf(stderr, "%s: fallback:%s\n", \
585 __FUNCTION__, #expr); \
586 stage->active = GL_FALSE; \
591 static void r300_check_render(GLcontext
*ctx
, struct tnl_pipeline_stage
*stage
)
593 r300ContextPtr r300
= R300_CONTEXT(ctx
);
596 if (RADEON_DEBUG
& DEBUG_STATE
)
597 fprintf(stderr
, "%s\n", __FUNCTION__
);
599 /* We only support rendering in hardware for now */
600 if (ctx
->RenderMode
!= GL_RENDER
) {
601 stage
->active
= GL_FALSE
;
605 #if 0 // selecting VERT_BIT_NORMAL still doesnt give us normals but why?
606 stage
->inputs
|= VERT_BIT_POS
;
607 stage
->inputs
|= VERT_BIT_NORMAL
;
608 stage
->inputs
|= VERT_BIT_COLOR0
;
611 // I failed to figure out how dither works in hardware,
612 // let's just ignore it for now
613 //FALLBACK_IF(ctx->Color.DitherFlag);
615 /* I'm almost certain I forgot something here */
616 #if 0 /* This should work now.. */
617 FALLBACK_IF(ctx
->Color
.AlphaEnabled
); // GL_ALPHA_TEST
618 FALLBACK_IF(ctx
->Color
.BlendEnabled
); // GL_BLEND
620 FALLBACK_IF(ctx
->Fog
.Enabled
); // GL_FOG
621 FALLBACK_IF(ctx
->Line
.SmoothFlag
); // GL_LINE_SMOOTH
622 FALLBACK_IF(ctx
->Line
.StippleFlag
); // GL_LINE_STIPPLE
623 FALLBACK_IF(ctx
->Point
.SmoothFlag
); // GL_POINT_SMOOTH
624 if (ctx
->Extensions
.NV_point_sprite
|| ctx
->Extensions
.ARB_point_sprite
)
625 FALLBACK_IF(ctx
->Point
.PointSprite
); // GL_POINT_SPRITE_NV
626 FALLBACK_IF(ctx
->Polygon
.OffsetPoint
); // GL_POLYGON_OFFSET_POINT
627 FALLBACK_IF(ctx
->Polygon
.OffsetLine
); // GL_POLYGON_OFFSET_LINE
628 //FALLBACK_IF(ctx->Polygon.OffsetFill); // GL_POLYGON_OFFSET_FILL
629 //if(ctx->Polygon.OffsetFill)WARN_ONCE("Polygon.OffsetFill not implemented, ignoring\n");
630 FALLBACK_IF(ctx
->Polygon
.SmoothFlag
); // GL_POLYGON_SMOOTH
631 FALLBACK_IF(ctx
->Polygon
.StippleFlag
); // GL_POLYGON_STIPPLE
632 //FALLBACK_IF(ctx->Stencil.Enabled); // GL_STENCIL_TEST
633 FALLBACK_IF(ctx
->Multisample
.Enabled
); // GL_MULTISAMPLE_ARB
635 /* One step at a time - let one texture pass.. */
636 for (i
= 1; i
< ctx
->Const
.MaxTextureUnits
; i
++)
637 FALLBACK_IF(ctx
->Texture
.Unit
[i
].Enabled
);
639 /* let r300_run_render do its job */
641 stage
->active
= GL_FALSE
;
646 static void dtr(struct tnl_pipeline_stage
*stage
)
651 const struct tnl_pipeline_stage _r300_render_stage
= {
653 _NEW_ALL
, /* re-check (always re-check for now) */
654 0, /* re-run (always runs) */
655 GL_TRUE
, /* active */
656 0, 0, /* inputs (set in check_render), outputs */
657 0, 0, /* changed_inputs, private */
658 dtr
, /* destructor */
659 r300_check_render
, /* check */
660 r300_run_render
/* run */