1 /* $XFree86: xc/lib/GL/mesa/src/drv/r200/r200_tcl.c,v 1.2 2002/12/16 16:18:55 dawes Exp $ */
3 Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved.
5 The Weather Channel (TM) funded Tungsten Graphics to develop the
6 initial release of the Radeon 8500 driver under the XFree86 license.
7 This notice must be preserved.
9 Permission is hereby granted, free of charge, to any person obtaining
10 a copy of this software and associated documentation files (the
11 "Software"), to deal in the Software without restriction, including
12 without limitation the rights to use, copy, modify, merge, publish,
13 distribute, sublicense, and/or sell copies of the Software, and to
14 permit persons to whom the Software is furnished to do so, subject to
15 the following conditions:
17 The above copyright notice and this permission notice (including the
18 next paragraph) shall be included in all copies or substantial
19 portions of the Software.
21 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
22 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
23 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
24 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
25 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
26 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
27 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
29 **************************************************************************/
33 * Keith Whitwell <keith@tungstengraphics.com>
43 #include "array_cache/acache.h"
45 #include "tnl/t_pipeline.h"
47 #include "r200_context.h"
48 #include "r200_state.h"
49 #include "r200_ioctl.h"
52 #include "r200_swtcl.h"
53 #include "r200_maos.h"
59 #define HAVE_LINE_LOOP 0
60 #define HAVE_LINE_STRIPS 1
61 #define HAVE_TRIANGLES 1
62 #define HAVE_TRI_STRIPS 1
63 #define HAVE_TRI_STRIP_1 0
64 #define HAVE_TRI_FANS 1
66 #define HAVE_QUAD_STRIPS 1
67 #define HAVE_POLYGONS 1
71 #define HW_POINTS ((ctx->_TriangleCaps & DD_POINT_SIZE) ? \
72 R200_VF_PRIM_POINT_SPRITES : R200_VF_PRIM_POINTS)
73 #define HW_LINES R200_VF_PRIM_LINES
74 #define HW_LINE_LOOP 0
75 #define HW_LINE_STRIP R200_VF_PRIM_LINE_STRIP
76 #define HW_TRIANGLES R200_VF_PRIM_TRIANGLES
77 #define HW_TRIANGLE_STRIP_0 R200_VF_PRIM_TRIANGLE_STRIP
78 #define HW_TRIANGLE_STRIP_1 0
79 #define HW_TRIANGLE_FAN R200_VF_PRIM_TRIANGLE_FAN
80 #define HW_QUADS R200_VF_PRIM_QUADS
81 #define HW_QUAD_STRIP R200_VF_PRIM_QUAD_STRIP
82 #define HW_POLYGON R200_VF_PRIM_POLYGON
85 static GLboolean discrete_prim
[0x10] = {
93 0, /* 7 tri_w_flags */
94 1, /* 8 rect list (unused) */
95 1, /* 9 3vert point */
97 0, /* b point sprite */
100 0, /* e quad strip */
105 #define LOCAL_VARS r200ContextPtr rmesa = R200_CONTEXT(ctx)
106 #define ELT_TYPE GLushort
108 #define ELT_INIT(prim, hw_prim) \
109 r200TclPrimitive( ctx, prim, hw_prim | R200_VF_PRIM_WALK_IND )
111 #define GET_MESA_ELTS() rmesa->tcl.Elts
114 /* Don't really know how many elts will fit in what's left of cmdbuf,
115 * as there is state to emit, etc:
118 /* Testing on isosurf shows a maximum around here. Don't know if it's
119 * the card or driver or kernel module that is causing the behaviour.
121 #define GET_MAX_HW_ELTS() 300
123 #define RESET_STIPPLE() do { \
124 R200_STATECHANGE( rmesa, lin ); \
125 r200EmitState( rmesa ); \
128 #define AUTO_STIPPLE( mode ) do { \
129 R200_STATECHANGE( rmesa, lin ); \
131 rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] |= \
132 R200_LINE_PATTERN_AUTO_RESET; \
134 rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &= \
135 ~R200_LINE_PATTERN_AUTO_RESET; \
136 r200EmitState( rmesa ); \
140 #define ALLOC_ELTS(nr) r200AllocElts( rmesa, nr )
142 static GLushort
*r200AllocElts( r200ContextPtr rmesa
, GLuint nr
)
144 if (rmesa
->dma
.flush
== r200FlushElts
&&
145 rmesa
->store
.cmd_used
+ nr
*2 < R200_CMD_BUF_SZ
) {
147 GLushort
*dest
= (GLushort
*)(rmesa
->store
.cmd_buf
+
148 rmesa
->store
.cmd_used
);
150 rmesa
->store
.cmd_used
+= nr
*2;
155 if (rmesa
->dma
.flush
)
156 rmesa
->dma
.flush( rmesa
);
158 r200EnsureCmdBufSpace( rmesa
, AOS_BUFSZ(rmesa
->tcl
.nr_aos_components
) +
159 rmesa
->hw
.max_state_size
+ ELTS_BUFSZ(nr
) );
162 rmesa
->tcl
.aos_components
,
163 rmesa
->tcl
.nr_aos_components
, 0 );
165 return r200AllocEltsOpenEnded( rmesa
, rmesa
->tcl
.hw_primitive
, nr
);
170 #define CLOSE_ELTS() \
172 if (0) R200_NEWPRIM( rmesa ); \
177 /* TODO: Try to extend existing primitive if both are identical,
178 * discrete and there are no intervening state changes. (Somewhat
179 * duplicates changes to DrawArrays code)
181 static void r200EmitPrim( GLcontext
*ctx
,
187 r200ContextPtr rmesa
= R200_CONTEXT( ctx
);
188 r200TclPrimitive( ctx
, prim
, hwprim
);
190 r200EnsureCmdBufSpace( rmesa
, AOS_BUFSZ(rmesa
->tcl
.nr_aos_components
) +
191 rmesa
->hw
.max_state_size
+ VBUF_BUFSZ
);
194 rmesa
->tcl
.aos_components
,
195 rmesa
->tcl
.nr_aos_components
,
198 /* Why couldn't this packet have taken an offset param?
200 r200EmitVbufPrim( rmesa
,
201 rmesa
->tcl
.hw_primitive
,
205 #define EMIT_PRIM(ctx, prim, hwprim, start, count) do { \
206 r200EmitPrim( ctx, prim, hwprim, start, count ); \
207 (void) rmesa; } while (0)
209 /* Try & join small primitives
212 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM ) 0
214 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM ) \
217 rmesa->tcl.hw_primitive == (PRIM| \
218 R200_VF_TCL_OUTPUT_VTX_ENABLE| \
219 R200_VF_PRIM_WALK_IND)))
222 #ifdef MESA_BIG_ENDIAN
223 /* We could do without (most of) this ugliness if dest was always 32 bit word aligned... */
224 #define EMIT_ELT(dest, offset, x) do { \
225 int off = offset + ( ( (GLuint)dest & 0x2 ) >> 1 ); \
226 GLushort *des = (GLushort *)( (GLuint)dest & ~0x2 ); \
227 (des)[ off + 1 - 2 * ( off & 1 ) ] = (GLushort)(x); \
228 (void)rmesa; } while (0)
230 #define EMIT_ELT(dest, offset, x) do { \
231 (dest)[offset] = (GLushort) (x); \
232 (void)rmesa; } while (0)
235 #define EMIT_TWO_ELTS(dest, offset, x, y) *(GLuint *)((dest)+offset) = ((y)<<16)|(x);
239 #define TAG(x) tcl_##x
240 #include "tnl_dd/t_dd_dmatmp2.h"
242 /**********************************************************************/
243 /* External entrypoints */
244 /**********************************************************************/
246 void r200EmitPrimitive( GLcontext
*ctx
,
251 tcl_render_tab_verts
[flags
&PRIM_MODE_MASK
]( ctx
, first
, last
, flags
);
254 void r200EmitEltPrimitive( GLcontext
*ctx
,
259 tcl_render_tab_elts
[flags
&PRIM_MODE_MASK
]( ctx
, first
, last
, flags
);
262 void r200TclPrimitive( GLcontext
*ctx
,
266 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
267 GLuint newprim
= hw_prim
| R200_VF_TCL_OUTPUT_VTX_ENABLE
;
269 if (newprim
!= rmesa
->tcl
.hw_primitive
||
270 !discrete_prim
[hw_prim
&0xf]) {
271 R200_NEWPRIM( rmesa
);
272 rmesa
->tcl
.hw_primitive
= newprim
;
277 /**********************************************************************/
278 /* Fog blend factor computation for hw tcl */
279 /* same calculation used as in t_vb_fog.c */
280 /**********************************************************************/
282 #define FOG_EXP_TABLE_SIZE 256
283 #define FOG_MAX (10.0)
284 #define EXP_FOG_MAX .0006595
285 #define FOG_INCR (FOG_MAX/FOG_EXP_TABLE_SIZE)
286 static GLfloat exp_table
[FOG_EXP_TABLE_SIZE
];
289 #define NEG_EXP( result, narg ) \
291 GLfloat f = (GLfloat) (narg * (1.0/FOG_INCR)); \
292 GLint k = (GLint) f; \
293 if (k > FOG_EXP_TABLE_SIZE-2) \
294 result = (GLfloat) EXP_FOG_MAX; \
296 result = exp_table[k] + (f-k)*(exp_table[k+1]-exp_table[k]); \
299 #define NEG_EXP( result, narg ) \
301 result = exp(-narg); \
307 * Initialize the exp_table[] lookup table for approximating exp().
310 r200InitStaticFogData( void )
314 for ( ; i
< FOG_EXP_TABLE_SIZE
; i
++, f
+= FOG_INCR
) {
315 exp_table
[i
] = (GLfloat
) exp(-f
);
321 * Compute per-vertex fog blend factors from fog coordinates by
322 * evaluating the GL_LINEAR, GL_EXP or GL_EXP2 fog function.
323 * Fog coordinates are distances from the eye (typically between the
324 * near and far clip plane distances).
325 * Note the fog (eye Z) coords may be negative so we use ABS(z) below.
326 * Fog blend factors are in the range [0,1].
329 r200ComputeFogBlendFactor( GLcontext
*ctx
, GLfloat fogcoord
)
331 GLfloat end
= ctx
->Fog
.End
;
333 const GLfloat z
= FABSF(fogcoord
);
335 switch (ctx
->Fog
.Mode
) {
337 if (ctx
->Fog
.Start
== ctx
->Fog
.End
)
340 d
= 1.0F
/ (ctx
->Fog
.End
- ctx
->Fog
.Start
);
341 temp
= (end
- z
) * d
;
342 return CLAMP(temp
, 0.0F
, 1.0F
);
345 d
= ctx
->Fog
.Density
;
346 NEG_EXP( temp
, d
* z
);
350 d
= ctx
->Fog
.Density
*ctx
->Fog
.Density
;
351 NEG_EXP( temp
, d
* z
* z
);
355 _mesa_problem(ctx
, "Bad fog mode in make_fog_coord");
361 /**********************************************************************/
362 /* Render pipeline stage */
363 /**********************************************************************/
368 static GLboolean
r200_run_tcl_render( GLcontext
*ctx
,
369 struct tnl_pipeline_stage
*stage
)
371 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
372 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
373 struct vertex_buffer
*VB
= &tnl
->vb
;
374 GLuint inputs
= VERT_BIT_POS
| VERT_BIT_COLOR0
;
377 /* TODO: separate this from the swtnl pipeline
379 if (rmesa
->TclFallback
)
380 return GL_TRUE
; /* fallback to software t&l */
382 if (R200_DEBUG
& DEBUG_PRIMS
)
383 fprintf(stderr
, "%s\n", __FUNCTION__
);
390 if (rmesa
->NewGLState
)
391 r200ValidateState( ctx
);
393 if (!ctx
->VertexProgram
._Enabled
) {
394 /* NOTE: inputs != tnl->render_inputs - these are the untransformed
397 if (ctx
->Light
.Enabled
) {
398 inputs
|= VERT_BIT_NORMAL
;
401 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
) {
402 inputs
|= VERT_BIT_COLOR1
;
405 if ( (ctx
->Fog
.FogCoordinateSource
== GL_FOG_COORD
) && ctx
->Fog
.Enabled
) {
406 inputs
|= VERT_BIT_FOG
;
409 for (i
= 0 ; i
< ctx
->Const
.MaxTextureUnits
; i
++) {
410 if (ctx
->Texture
.Unit
[i
]._ReallyEnabled
) {
411 if (rmesa
->TexGenNeedNormals
[i
]) {
412 inputs
|= VERT_BIT_NORMAL
;
414 inputs
|= VERT_BIT_TEX(i
);
419 GLuint out_vtxfmt0
= 0;
420 GLuint out_vtxfmt1
= 0;
421 GLuint out_compsel
= 0;
422 GLuint vp_out
= rmesa
->curr_vp_hw
->mesa_program
.Base
.OutputsWritten
;
423 /* can't handle other inputs, generic attribs etc. currently - should never arrive here */
424 assert ((rmesa
->curr_vp_hw
->mesa_program
.Base
.InputsRead
&
425 ~(VERT_BIT_POS
| VERT_BIT_NORMAL
| VERT_BIT_COLOR0
| VERT_BIT_COLOR1
|
426 VERT_BIT_FOG
| VERT_BIT_TEX0
| VERT_BIT_TEX1
| VERT_BIT_TEX2
|
427 VERT_BIT_TEX3
| VERT_BIT_TEX4
| VERT_BIT_TEX5
)) == 0);
428 inputs
|= rmesa
->curr_vp_hw
->mesa_program
.Base
.InputsRead
;
429 /* FIXME: this is a mess. Not really sure how to set up TCL_OUTPUT_VTXFMT
430 in "undefined" cases (e.g. output needed later but not written by vertex program or vice versa)
431 - however misconfiguration here will almost certainly lock up the chip.
432 I think at the very least we need to enable tcl outputs which we write to. Maybe even need to
433 fix up a vertex program so an output needed later always gets written?
434 For now just set the compsel and output_vtxfmt to the outputs written.
435 However, for simplicity we assume always all 4 values are written which may not be correct
436 (but I don't know if it could lead to lockups). */
437 assert(vp_out
& (1 << VERT_RESULT_HPOS
));
438 out_vtxfmt0
= R200_VTX_XY
| R200_VTX_Z0
| R200_VTX_W0
;
439 /* FIXME: need to always enable color_0 otherwise doom3's shadow vp (?) will lock up (?) */
440 out_vtxfmt0
|= R200_VTX_FP_RGBA
<< R200_VTX_COLOR_0_SHIFT
;
441 out_compsel
= R200_OUTPUT_XYZW
;
442 if (vp_out
& (1 << VERT_RESULT_COL0
)) {
443 out_vtxfmt0
|= R200_VTX_FP_RGBA
<< R200_VTX_COLOR_0_SHIFT
;
444 out_compsel
|= R200_OUTPUT_COLOR_0
;
446 if (vp_out
& (1 << VERT_RESULT_COL1
)) {
447 out_vtxfmt0
|= R200_VTX_FP_RGBA
<< R200_VTX_COLOR_1_SHIFT
;
448 out_compsel
|= R200_OUTPUT_COLOR_1
;
450 /* FIXME: probably not everything is set up for fogc and psiz to work correctly */
451 if (vp_out
& (1 << VERT_RESULT_FOGC
)) {
452 out_vtxfmt0
|= R200_VTX_DISCRETE_FOG
;
453 out_compsel
|= R200_OUTPUT_DISCRETE_FOG
;
455 if (vp_out
& (1 << VERT_RESULT_PSIZ
)) {
456 out_vtxfmt0
|= R200_VTX_POINT_SIZE
;
457 out_compsel
|= R200_OUTPUT_PT_SIZE
;
459 for (i
= VERT_RESULT_TEX0
; i
< VERT_RESULT_TEX6
; i
++) {
460 if (vp_out
& (1 << i
)) {
461 out_vtxfmt1
|= 4 << ((i
- VERT_RESULT_TEX0
) * 3);
462 out_compsel
|= R200_OUTPUT_TEX_0
<< (i
- VERT_RESULT_TEX0
);
465 if ((rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_VTXFMT_0
] != out_vtxfmt0
) ||
466 (rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_VTXFMT_1
] != out_vtxfmt1
) ||
467 (rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_COMPSEL
] != out_compsel
)) {
468 R200_STATECHANGE( rmesa
, vtx
);
469 rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_VTXFMT_0
] = out_vtxfmt0
;
470 rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_VTXFMT_1
] = out_vtxfmt1
;
471 rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_COMPSEL
] = out_compsel
;
472 /* FIXME: should restore this when disabling vertex programs maybe? */
476 /* Do the actual work:
478 r200ReleaseArrays( ctx
, ~0 /* stage->changed_inputs */ );
479 r200EmitArrays( ctx
, inputs
);
481 rmesa
->tcl
.Elts
= VB
->Elts
;
483 for (i
= 0 ; i
< VB
->PrimitiveCount
; i
++)
485 GLuint prim
= VB
->Primitive
[i
].mode
;
486 GLuint start
= VB
->Primitive
[i
].start
;
487 GLuint length
= VB
->Primitive
[i
].count
;
493 r200EmitEltPrimitive( ctx
, start
, start
+length
, prim
);
495 r200EmitPrimitive( ctx
, start
, start
+length
, prim
);
498 return GL_FALSE
; /* finished the pipe */
503 /* Initial state for tcl stage.
505 const struct tnl_pipeline_stage _r200_tcl_stage
=
512 r200_run_tcl_render
/* run */
517 /**********************************************************************/
518 /* Validate state at pipeline start */
519 /**********************************************************************/
522 /*-----------------------------------------------------------------------
523 * Manage TCL fallbacks
527 static void transition_to_swtnl( GLcontext
*ctx
)
529 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
530 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
532 R200_NEWPRIM( rmesa
);
534 r200ChooseVertexState( ctx
);
535 r200ChooseRenderState( ctx
);
537 _mesa_validate_all_lighting_tables( ctx
);
539 tnl
->Driver
.NotifyMaterialChange
=
540 _mesa_validate_all_lighting_tables
;
542 r200ReleaseArrays( ctx
, ~0 );
544 /* Still using the D3D based hardware-rasterizer from the radeon;
545 * need to put the card into D3D mode to make it work:
547 R200_STATECHANGE( rmesa
, vap
);
548 /* not sure if it's strictly necessary to disable VAP_PROG_VTX_SHADER_ENABLE in addition to VAP_TCL_ENABLE) */
549 rmesa
->hw
.vap
.cmd
[VAP_SE_VAP_CNTL
] &= ~(R200_VAP_TCL_ENABLE
|R200_VAP_PROG_VTX_SHADER_ENABLE
);
552 static void transition_to_hwtnl( GLcontext
*ctx
)
554 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
555 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
557 _tnl_need_projected_coords( ctx
, GL_FALSE
);
559 r200UpdateMaterial( ctx
);
561 tnl
->Driver
.NotifyMaterialChange
= r200UpdateMaterial
;
563 if ( rmesa
->dma
.flush
)
564 rmesa
->dma
.flush( rmesa
);
566 rmesa
->dma
.flush
= NULL
;
568 if (rmesa
->swtcl
.indexed_verts
.buf
)
569 r200ReleaseDmaRegion( rmesa
, &rmesa
->swtcl
.indexed_verts
,
572 R200_STATECHANGE( rmesa
, vap
);
573 rmesa
->hw
.vap
.cmd
[VAP_SE_VAP_CNTL
] |= R200_VAP_TCL_ENABLE
;
574 rmesa
->hw
.vap
.cmd
[VAP_SE_VAP_CNTL
] &= ~R200_VAP_FORCE_W_TO_ONE
;
576 if (ctx
->VertexProgram
._Enabled
) {
577 rmesa
->hw
.vap
.cmd
[VAP_SE_VAP_CNTL
] |= R200_VAP_PROG_VTX_SHADER_ENABLE
;
580 if ( ((rmesa
->hw
.ctx
.cmd
[CTX_PP_FOG_COLOR
] & R200_FOG_USE_MASK
)
581 == R200_FOG_USE_SPEC_ALPHA
) &&
582 (ctx
->Fog
.FogCoordinateSource
== GL_FOG_COORD
)) {
583 R200_STATECHANGE( rmesa
, ctx
);
584 rmesa
->hw
.ctx
.cmd
[CTX_PP_FOG_COLOR
] &= ~R200_FOG_USE_MASK
;
585 rmesa
->hw
.ctx
.cmd
[CTX_PP_FOG_COLOR
] |= R200_FOG_USE_VTX_FOG
;
588 R200_STATECHANGE( rmesa
, vte
);
589 rmesa
->hw
.vte
.cmd
[VTE_SE_VTE_CNTL
] &= ~(R200_VTX_XY_FMT
|R200_VTX_Z_FMT
);
590 rmesa
->hw
.vte
.cmd
[VTE_SE_VTE_CNTL
] |= R200_VTX_W0_FMT
;
592 if (R200_DEBUG
& DEBUG_FALLBACKS
)
593 fprintf(stderr
, "R200 end tcl fallback\n");
597 static char *fallbackStrings
[] = {
598 "Rasterization fallback",
599 "Unfilled triangles",
600 "Twosided lighting, differing materials",
601 "Materials in VB (maybe between begin/end)",
614 static char *getFallbackString(GLuint bit
)
621 return fallbackStrings
[i
];
626 void r200TclFallback( GLcontext
*ctx
, GLuint bit
, GLboolean mode
)
628 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
629 GLuint oldfallback
= rmesa
->TclFallback
;
632 rmesa
->TclFallback
|= bit
;
633 if (oldfallback
== 0) {
634 if (R200_DEBUG
& DEBUG_FALLBACKS
)
635 fprintf(stderr
, "R200 begin tcl fallback %s\n",
636 getFallbackString( bit
));
637 transition_to_swtnl( ctx
);
641 rmesa
->TclFallback
&= ~bit
;
642 if (oldfallback
== bit
) {
643 if (R200_DEBUG
& DEBUG_FALLBACKS
)
644 fprintf(stderr
, "R200 end tcl fallback %s\n",
645 getFallbackString( bit
));
646 transition_to_hwtnl( ctx
);