2 /**************************************************************************
4 Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
5 Tungsten Graphics Inc., Austin, Texas.
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>
42 #include "array_cache/acache.h"
44 #include "tnl/t_pipeline.h"
46 #include "radeon_context.h"
47 #include "radeon_state.h"
48 #include "radeon_ioctl.h"
49 #include "radeon_tex.h"
50 #include "radeon_tcl.h"
51 #include "radeon_swtcl.h"
52 #include "radeon_maos.h"
57 * Render unclipped vertex buffers by emitting vertices directly to
58 * dma buffers. Use strip/fan hardware primitives where possible.
59 * Try to simulate missing primitives with indexed vertices.
63 #define HAVE_LINE_LOOP 0
64 #define HAVE_LINE_STRIPS 1
65 #define HAVE_TRIANGLES 1
66 #define HAVE_TRI_STRIPS 1
67 #define HAVE_TRI_STRIP_1 0
68 #define HAVE_TRI_FANS 1
70 #define HAVE_QUAD_STRIPS 0
71 #define HAVE_POLYGONS 1
75 #define HW_POINTS RADEON_CP_VC_CNTL_PRIM_TYPE_POINT
76 #define HW_LINES RADEON_CP_VC_CNTL_PRIM_TYPE_LINE
77 #define HW_LINE_LOOP 0
78 #define HW_LINE_STRIP RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP
79 #define HW_TRIANGLES RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST
80 #define HW_TRIANGLE_STRIP_0 RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP
81 #define HW_TRIANGLE_STRIP_1 0
82 #define HW_TRIANGLE_FAN RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN
84 #define HW_QUAD_STRIP 0
85 #define HW_POLYGON RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN
88 static GLboolean discrete_prim
[0x10] = {
96 1, /* 7 rect list (unused) */
97 1, /* 8 3vert point */
108 #define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx)
109 #define ELT_TYPE GLushort
111 #define ELT_INIT(prim, hw_prim) \
112 radeonTclPrimitive( ctx, prim, hw_prim | RADEON_CP_VC_CNTL_PRIM_WALK_IND )
114 #define GET_MESA_ELTS() rmesa->tcl.Elts
117 /* Don't really know how many elts will fit in what's left of cmdbuf,
118 * as there is state to emit, etc:
121 /* Testing on isosurf shows a maximum around here. Don't know if it's
122 * the card or driver or kernel module that is causing the behaviour.
124 #define GET_MAX_HW_ELTS() 300
127 #define RESET_STIPPLE() do { \
128 RADEON_STATECHANGE( rmesa, lin ); \
129 radeonEmitState( rmesa ); \
132 #define AUTO_STIPPLE( mode ) do { \
133 RADEON_STATECHANGE( rmesa, lin ); \
135 rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] |= \
136 RADEON_LINE_PATTERN_AUTO_RESET; \
138 rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &= \
139 ~RADEON_LINE_PATTERN_AUTO_RESET; \
140 radeonEmitState( rmesa ); \
145 #define ALLOC_ELTS(nr) radeonAllocElts( rmesa, nr )
147 static GLushort
*radeonAllocElts( radeonContextPtr rmesa
, GLuint nr
)
149 if (rmesa
->dma
.flush
)
150 rmesa
->dma
.flush( rmesa
);
152 radeonEnsureCmdBufSpace(rmesa
, AOS_BUFSZ(rmesa
->tcl
.nr_aos_components
) +
153 rmesa
->hw
.max_state_size
+ ELTS_BUFSZ(nr
));
155 radeonEmitAOS( rmesa
,
156 rmesa
->tcl
.aos_components
,
157 rmesa
->tcl
.nr_aos_components
, 0 );
159 return radeonAllocEltsOpenEnded( rmesa
,
160 rmesa
->tcl
.vertex_format
,
161 rmesa
->tcl
.hw_primitive
, nr
);
164 #define CLOSE_ELTS() RADEON_NEWPRIM( rmesa )
168 /* TODO: Try to extend existing primitive if both are identical,
169 * discrete and there are no intervening state changes. (Somewhat
170 * duplicates changes to DrawArrays code)
172 static void radeonEmitPrim( GLcontext
*ctx
,
178 radeonContextPtr rmesa
= RADEON_CONTEXT( ctx
);
179 radeonTclPrimitive( ctx
, prim
, hwprim
);
181 radeonEnsureCmdBufSpace( rmesa
, AOS_BUFSZ(rmesa
->tcl
.nr_aos_components
) +
182 rmesa
->hw
.max_state_size
+ VBUF_BUFSZ
);
184 radeonEmitAOS( rmesa
,
185 rmesa
->tcl
.aos_components
,
186 rmesa
->tcl
.nr_aos_components
,
189 /* Why couldn't this packet have taken an offset param?
191 radeonEmitVbufPrim( rmesa
,
192 rmesa
->tcl
.vertex_format
,
193 rmesa
->tcl
.hw_primitive
,
197 #define EMIT_PRIM( ctx, prim, hwprim, start, count ) do { \
198 radeonEmitPrim( ctx, prim, hwprim, start, count ); \
199 (void) rmesa; } while (0)
201 /* Try & join small primitives
204 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM ) 0
206 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM ) \
209 rmesa->tcl.hw_primitive == (PRIM| \
210 RADEON_CP_VC_CNTL_PRIM_WALK_IND| \
211 RADEON_CP_VC_CNTL_TCL_ENABLE)))
214 #ifdef MESA_BIG_ENDIAN
215 /* We could do without (most of) this ugliness if dest was always 32 bit word aligned... */
216 #define EMIT_ELT(dest, offset, x) do { \
217 int off = offset + ( ( (GLuint)dest & 0x2 ) >> 1 ); \
218 GLushort *des = (GLushort *)( (GLuint)dest & ~0x2 ); \
219 (des)[ off + 1 - 2 * ( off & 1 ) ] = (GLushort)(x); \
220 (void)rmesa; } while (0)
222 #define EMIT_ELT(dest, offset, x) do { \
223 (dest)[offset] = (GLushort) (x); \
224 (void)rmesa; } while (0)
227 #define EMIT_TWO_ELTS(dest, offset, x, y) *(GLuint *)(dest+offset) = ((y)<<16)|(x);
231 #define TAG(x) tcl_##x
232 #include "tnl_dd/t_dd_dmatmp2.h"
234 /**********************************************************************/
235 /* External entrypoints */
236 /**********************************************************************/
238 void radeonEmitPrimitive( GLcontext
*ctx
,
243 tcl_render_tab_verts
[flags
&PRIM_MODE_MASK
]( ctx
, first
, last
, flags
);
246 void radeonEmitEltPrimitive( GLcontext
*ctx
,
251 tcl_render_tab_elts
[flags
&PRIM_MODE_MASK
]( ctx
, first
, last
, flags
);
254 void radeonTclPrimitive( GLcontext
*ctx
,
258 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
260 GLuint newprim
= hw_prim
| RADEON_CP_VC_CNTL_TCL_ENABLE
;
262 if (newprim
!= rmesa
->tcl
.hw_primitive
||
263 !discrete_prim
[hw_prim
&0xf]) {
264 RADEON_NEWPRIM( rmesa
);
265 rmesa
->tcl
.hw_primitive
= newprim
;
268 se_cntl
= rmesa
->hw
.set
.cmd
[SET_SE_CNTL
];
269 se_cntl
&= ~RADEON_FLAT_SHADE_VTX_LAST
;
271 if (prim
== GL_POLYGON
&& (ctx
->_TriangleCaps
& DD_FLATSHADE
))
272 se_cntl
|= RADEON_FLAT_SHADE_VTX_0
;
274 se_cntl
|= RADEON_FLAT_SHADE_VTX_LAST
;
276 if (se_cntl
!= rmesa
->hw
.set
.cmd
[SET_SE_CNTL
]) {
277 RADEON_STATECHANGE( rmesa
, set
);
278 rmesa
->hw
.set
.cmd
[SET_SE_CNTL
] = se_cntl
;
283 /**********************************************************************/
284 /* Render pipeline stage */
285 /**********************************************************************/
290 static GLboolean
radeon_run_tcl_render( GLcontext
*ctx
,
291 struct tnl_pipeline_stage
*stage
)
293 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
294 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
295 struct vertex_buffer
*VB
= &tnl
->vb
;
296 GLuint inputs
= VERT_BIT_POS
| VERT_BIT_COLOR0
;
299 /* TODO: separate this from the swtnl pipeline
301 if (rmesa
->TclFallback
)
302 return GL_TRUE
; /* fallback to software t&l */
307 /* NOTE: inputs != tnl->render_inputs - these are the untransformed
310 if (ctx
->Light
.Enabled
) {
311 inputs
|= VERT_BIT_NORMAL
;
312 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
) {
313 inputs
|= VERT_BIT_COLOR1
;
317 if ( ctx
->Fog
.FogCoordinateSource
== GL_FOG_COORD
) {
318 inputs
|= VERT_BIT_FOG
;
321 for (i
= 0 ; i
< ctx
->Const
.MaxTextureUnits
; i
++) {
322 if (ctx
->Texture
.Unit
[i
]._ReallyEnabled
) {
323 /* TODO: probably should not emit texture coords when texgen is enabled */
324 if (rmesa
->TexGenNeedNormals
[i
]) {
325 inputs
|= VERT_BIT_NORMAL
;
327 inputs
|= VERT_BIT_TEX(i
);
331 radeonReleaseArrays( ctx
, ~0 );
332 radeonEmitArrays( ctx
, inputs
);
334 rmesa
->tcl
.Elts
= VB
->Elts
;
336 for (i
= 0 ; i
< VB
->PrimitiveCount
; i
++)
338 GLuint prim
= VB
->Primitive
[i
].mode
;
339 GLuint start
= VB
->Primitive
[i
].start
;
340 GLuint length
= VB
->Primitive
[i
].count
;
346 radeonEmitEltPrimitive( ctx
, start
, start
+length
, prim
);
348 radeonEmitPrimitive( ctx
, start
, start
+length
, prim
);
351 return GL_FALSE
; /* finished the pipe */
356 /* Initial state for tcl stage.
358 const struct tnl_pipeline_stage _radeon_tcl_stage
=
365 radeon_run_tcl_render
/* run */
370 /**********************************************************************/
371 /* Validate state at pipeline start */
372 /**********************************************************************/
375 /*-----------------------------------------------------------------------
376 * Manage TCL fallbacks
380 static void transition_to_swtnl( GLcontext
*ctx
)
382 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
383 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
386 RADEON_NEWPRIM( rmesa
);
387 rmesa
->swtcl
.vertex_format
= 0;
389 radeonChooseVertexState( ctx
);
390 radeonChooseRenderState( ctx
);
392 _mesa_validate_all_lighting_tables( ctx
);
394 tnl
->Driver
.NotifyMaterialChange
=
395 _mesa_validate_all_lighting_tables
;
397 radeonReleaseArrays( ctx
, ~0 );
399 se_cntl
= rmesa
->hw
.set
.cmd
[SET_SE_CNTL
];
400 se_cntl
|= RADEON_FLAT_SHADE_VTX_LAST
;
402 if (se_cntl
!= rmesa
->hw
.set
.cmd
[SET_SE_CNTL
]) {
403 RADEON_STATECHANGE( rmesa
, set
);
404 rmesa
->hw
.set
.cmd
[SET_SE_CNTL
] = se_cntl
;
409 static void transition_to_hwtnl( GLcontext
*ctx
)
411 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
412 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
413 GLuint se_coord_fmt
= (RADEON_VTX_W0_IS_NOT_1_OVER_W0
|
414 RADEON_TEX1_W_ROUTING_USE_Q1
);
416 if ( se_coord_fmt
!= rmesa
->hw
.set
.cmd
[SET_SE_COORDFMT
] ) {
417 RADEON_STATECHANGE( rmesa
, set
);
418 rmesa
->hw
.set
.cmd
[SET_SE_COORDFMT
] = se_coord_fmt
;
419 _tnl_need_projected_coords( ctx
, GL_FALSE
);
422 radeonUpdateMaterial( ctx
);
424 tnl
->Driver
.NotifyMaterialChange
= radeonUpdateMaterial
;
426 if ( rmesa
->dma
.flush
)
427 rmesa
->dma
.flush( rmesa
);
429 rmesa
->dma
.flush
= NULL
;
430 rmesa
->swtcl
.vertex_format
= 0;
432 if (rmesa
->swtcl
.indexed_verts
.buf
)
433 radeonReleaseDmaRegion( rmesa
, &rmesa
->swtcl
.indexed_verts
,
436 if (RADEON_DEBUG
& DEBUG_FALLBACKS
)
437 fprintf(stderr
, "Radeon end tcl fallback\n");
440 static char *fallbackStrings
[] = {
441 "Rasterization fallback",
442 "Unfilled triangles",
443 "Twosided lighting, differing materials",
444 "Materials in VB (maybe between begin/end)",
452 static char *getFallbackString(GLuint bit
)
459 return fallbackStrings
[i
];
464 void radeonTclFallback( GLcontext
*ctx
, GLuint bit
, GLboolean mode
)
466 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
467 GLuint oldfallback
= rmesa
->TclFallback
;
470 rmesa
->TclFallback
|= bit
;
471 if (oldfallback
== 0) {
472 if (RADEON_DEBUG
& DEBUG_FALLBACKS
)
473 fprintf(stderr
, "Radeon begin tcl fallback %s\n",
474 getFallbackString( bit
));
475 transition_to_swtnl( ctx
);
479 rmesa
->TclFallback
&= ~bit
;
480 if (oldfallback
== bit
) {
481 if (RADEON_DEBUG
& DEBUG_FALLBACKS
)
482 fprintf(stderr
, "Radeon end tcl fallback %s\n",
483 getFallbackString( bit
));
484 transition_to_hwtnl( ctx
);