2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
38 #include "program/prog_instruction.h"
39 #include "brw_context.h"
42 #define SATURATE (1<<5)
44 /* A big lookup table is used to figure out which and how many
45 * additional regs will inserted before the main payload in the WM
46 * program execution. These mainly relate to depth and stencil
47 * processing and the early-depth-test optimization.
49 #define IZ_PS_KILL_ALPHATEST_BIT 0x1
50 #define IZ_PS_COMPUTES_DEPTH_BIT 0x2
51 #define IZ_DEPTH_WRITE_ENABLE_BIT 0x4
52 #define IZ_DEPTH_TEST_ENABLE_BIT 0x8
53 #define IZ_STENCIL_WRITE_ENABLE_BIT 0x10
54 #define IZ_STENCIL_TEST_ENABLE_BIT 0x20
55 #define IZ_BIT_MAX 0x40
58 #define AA_SOMETIMES 1
61 struct brw_wm_prog_key
{
65 GLuint nr_color_regions
:5;
66 GLuint render_to_fbo
:1;
68 GLuint clamp_fragment_color
:1;
72 * Per-sampler comparison functions:
74 * If comparison mode is GL_COMPARE_R_TO_TEXTURE, then this is set to one
75 * of GL_NEVER, GL_LESS, GL_EQUAL, GL_LEQUAL, GL_GREATER, GL_NOTEQUAL,
76 * GL_GEQUAL, or GL_ALWAYS. Otherwise (comparison mode is GL_NONE), this
77 * field is irrelevant so it's left as GL_NONE (0).
79 * While this is a GLenum, all possible values fit in 16-bits.
81 uint16_t compare_funcs
[BRW_MAX_TEX_UNIT
];
83 GLbitfield proj_attrib_mask
; /**< one bit per fragment program attribute */
84 GLuint yuvtex_mask
:16;
85 GLuint yuvtex_swap_mask
:16; /* UV swaped */
86 uint16_t gl_clamp_mask
[3];
88 GLushort tex_swizzles
[BRW_MAX_TEX_UNIT
];
89 GLushort drawable_height
;
90 GLbitfield64 vp_outputs_written
;
91 GLuint program_string_id
:32;
95 /* A bit of a glossary:
97 * brw_wm_value: A computed value or program input. Values are
98 * constant, they are created once and are never modified. When a
99 * fragment program register is written or overwritten, new values are
100 * created fresh, preserving the rule that values are constant.
102 * brw_wm_ref: A reference to a value. Wherever a value used is by an
103 * instruction or as a program output, that is tracked with an
104 * instance of this struct. All references to a value occur after it
105 * is created. After the last reference, a value is dead and can be
108 * brw_wm_grf: Represents a physical hardware register. May be either
109 * empty or hold a value. Register allocation is the process of
110 * assigning values to grf registers. This occurs in pass2 and the
111 * brw_wm_grf struct is not used before that.
113 * Fragment program registers: These are time-varying constructs that
114 * are hard to reason about and which we translate away in pass0. A
115 * single fragment program register element (eg. temp[0].x) will be
116 * translated to one or more brw_wm_value structs, one for each time
117 * that temp[0].x is written to during the program.
122 /* Used in pass2 to track register allocation.
125 struct brw_wm_value
*value
;
129 struct brw_wm_value
{
130 struct brw_reg hw_reg
; /* emitted to this reg, may not always be there */
131 struct brw_wm_ref
*lastuse
;
132 struct brw_wm_grf
*resident
;
133 GLuint contributes_to_output
:1;
134 GLuint spill_slot
:16; /* if non-zero, spill immediately after calculation */
138 struct brw_reg hw_reg
; /* nr filled in in pass2, everything else, pass0 */
139 struct brw_wm_value
*value
;
140 struct brw_wm_ref
*prevuse
;
141 GLuint unspill_reg
:7; /* unspill to reg */
146 struct brw_wm_constref
{
147 const struct brw_wm_ref
*ref
;
152 struct brw_wm_instruction
{
153 struct brw_wm_value
*dst
[4];
154 struct brw_wm_ref
*src
[3][4];
158 GLuint tex_unit
:4; /* texture unit for TEX, TXD, TXP instructions */
159 GLuint tex_idx
:3; /* TEXTURE_1D,2D,3D,CUBE,RECT_INDEX source target */
160 GLuint tex_shadow
:1; /* do shadow comparison? */
161 GLuint eot
:1; /* End of thread indicator for FB_WRITE*/
162 GLuint target
:10; /* target binding table index for FB_WRITE*/
166 #define BRW_WM_MAX_INSN (MAX_PROGRAM_INSTRUCTIONS*3 + FRAG_ATTRIB_MAX + 3)
167 #define BRW_WM_MAX_GRF 128 /* hardware limit */
168 #define BRW_WM_MAX_VREG (BRW_WM_MAX_INSN * 4)
169 #define BRW_WM_MAX_REF (BRW_WM_MAX_INSN * 12)
170 #define BRW_WM_MAX_PARAM 256
171 #define BRW_WM_MAX_CONST 256
172 #define BRW_WM_MAX_SUBROUTINE 16
174 /* used in masks next to WRITEMASK_*. */
175 #define SATURATE (1<<5)
178 /* New opcodes to track internal operations required for WM unit.
179 * These are added early so that the registers used can be tracked,
180 * freed and reused like those of other instructions.
182 #define WM_PIXELXY (MAX_OPCODE)
183 #define WM_DELTAXY (MAX_OPCODE + 1)
184 #define WM_PIXELW (MAX_OPCODE + 2)
185 #define WM_LINTERP (MAX_OPCODE + 3)
186 #define WM_PINTERP (MAX_OPCODE + 4)
187 #define WM_CINTERP (MAX_OPCODE + 5)
188 #define WM_WPOSXY (MAX_OPCODE + 6)
189 #define WM_FB_WRITE (MAX_OPCODE + 7)
190 #define WM_FRONTFACING (MAX_OPCODE + 8)
191 #define MAX_WM_OPCODE (MAX_OPCODE + 9)
193 #define PROGRAM_PAYLOAD (PROGRAM_FILE_MAX)
194 #define NUM_FILES (PROGRAM_PAYLOAD + 1)
196 #define PAYLOAD_DEPTH (FRAG_ATTRIB_MAX)
197 #define PAYLOAD_W (FRAG_ATTRIB_MAX + 1)
198 #define PAYLOAD_FP_REG_MAX (FRAG_ATTRIB_MAX + 2)
200 struct brw_wm_compile
{
201 struct brw_compile func
;
202 struct brw_wm_prog_key key
;
203 struct brw_wm_prog_data prog_data
;
205 struct brw_fragment_program
*fp
;
207 GLfloat (*env_param
)[4];
214 uint8_t source_depth_reg
;
215 uint8_t source_w_reg
;
216 uint8_t aa_dest_stencil_reg
;
217 uint8_t dest_depth_reg
;
218 uint8_t nr_payload_regs
;
219 GLuint computes_depth
:1; /* could be derived from program string */
220 GLuint source_depth_to_render_target
:1;
221 GLuint runtime_check_aads_emit
:1;
223 /* Initial pass - translate fp instructions to fp instructions,
224 * simplifying and adding instructions for interpolation and
225 * framebuffer writes.
227 struct prog_instruction
*prog_instructions
;
230 GLuint fp_interp_emitted
;
232 struct prog_src_register pixel_xy
;
233 struct prog_src_register delta_xy
;
234 struct prog_src_register pixel_w
;
237 struct brw_wm_value
*vreg
;
240 struct brw_wm_value creg
[BRW_WM_MAX_PARAM
];
244 struct brw_wm_value depth
[4]; /* includes r0/r1 */
245 struct brw_wm_value input_interp
[FRAG_ATTRIB_MAX
];
249 const struct brw_wm_ref
*pass0_fp_reg
[NUM_FILES
][256][4];
251 struct brw_wm_ref undef_ref
;
252 struct brw_wm_value undef_value
;
254 struct brw_wm_ref
*refs
;
257 struct brw_wm_instruction
*instruction
;
260 struct brw_wm_constref constref
[BRW_WM_MAX_CONST
];
263 struct brw_wm_grf pass2_grf
[BRW_WM_MAX_GRF
/2];
269 GLuint cur_inst
; /**< index of current instruction */
271 GLboolean out_of_regs
; /**< ran out of GRF registers? */
273 /** Mapping from Mesa registers to hardware registers */
277 } wm_regs
[NUM_FILES
][256][4];
279 GLboolean used_grf
[BRW_WM_MAX_GRF
];
280 GLuint first_free_grf
;
281 struct brw_reg stack
;
282 struct brw_reg emit_mask_reg
;
283 GLuint tmp_regs
[BRW_WM_MAX_GRF
];
286 GLuint subroutines
[BRW_WM_MAX_SUBROUTINE
];
287 GLuint dispatch_width
;
289 /** we may need up to 3 constants per instruction (if use_const_buffer) */
297 /** Bits for prog_instruction::Aux field */
298 #define INST_AUX_EOT 0x1
299 #define INST_AUX_TARGET(T) (T << 1)
300 #define INST_AUX_GET_TARGET(AUX) ((AUX) >> 1)
303 GLuint
brw_wm_nr_args( GLuint opcode
);
304 GLuint
brw_wm_is_scalar_result( GLuint opcode
);
306 void brw_wm_pass_fp( struct brw_wm_compile
*c
);
307 void brw_wm_pass0( struct brw_wm_compile
*c
);
308 void brw_wm_pass1( struct brw_wm_compile
*c
);
309 void brw_wm_pass2( struct brw_wm_compile
*c
);
310 void brw_wm_emit( struct brw_wm_compile
*c
);
311 GLboolean
brw_wm_arg_can_be_immediate(enum prog_opcode
, int arg
);
312 void brw_wm_print_value( struct brw_wm_compile
*c
,
313 struct brw_wm_value
*value
);
315 void brw_wm_print_ref( struct brw_wm_compile
*c
,
316 struct brw_wm_ref
*ref
);
318 void brw_wm_print_insn( struct brw_wm_compile
*c
,
319 struct brw_wm_instruction
*inst
);
321 void brw_wm_print_program( struct brw_wm_compile
*c
,
324 void brw_wm_lookup_iz(struct intel_context
*intel
,
325 struct brw_wm_compile
*c
);
327 bool brw_wm_fs_emit(struct brw_context
*brw
, struct brw_wm_compile
*c
,
328 struct gl_shader_program
*prog
);
331 void emit_alu1(struct brw_compile
*p
,
332 struct brw_instruction
*(*func
)(struct brw_compile
*,
335 const struct brw_reg
*dst
,
337 const struct brw_reg
*arg0
);
338 void emit_alu2(struct brw_compile
*p
,
339 struct brw_instruction
*(*func
)(struct brw_compile
*,
343 const struct brw_reg
*dst
,
345 const struct brw_reg
*arg0
,
346 const struct brw_reg
*arg1
);
347 void emit_cinterp(struct brw_compile
*p
,
348 const struct brw_reg
*dst
,
350 const struct brw_reg
*arg0
);
351 void emit_cmp(struct brw_compile
*p
,
352 const struct brw_reg
*dst
,
354 const struct brw_reg
*arg0
,
355 const struct brw_reg
*arg1
,
356 const struct brw_reg
*arg2
);
357 void emit_ddxy(struct brw_compile
*p
,
358 const struct brw_reg
*dst
,
361 const struct brw_reg
*arg0
);
362 void emit_delta_xy(struct brw_compile
*p
,
363 const struct brw_reg
*dst
,
365 const struct brw_reg
*arg0
);
366 void emit_dp2(struct brw_compile
*p
,
367 const struct brw_reg
*dst
,
369 const struct brw_reg
*arg0
,
370 const struct brw_reg
*arg1
);
371 void emit_dp3(struct brw_compile
*p
,
372 const struct brw_reg
*dst
,
374 const struct brw_reg
*arg0
,
375 const struct brw_reg
*arg1
);
376 void emit_dp4(struct brw_compile
*p
,
377 const struct brw_reg
*dst
,
379 const struct brw_reg
*arg0
,
380 const struct brw_reg
*arg1
);
381 void emit_dph(struct brw_compile
*p
,
382 const struct brw_reg
*dst
,
384 const struct brw_reg
*arg0
,
385 const struct brw_reg
*arg1
);
386 void emit_fb_write(struct brw_wm_compile
*c
,
387 struct brw_reg
*arg0
,
388 struct brw_reg
*arg1
,
389 struct brw_reg
*arg2
,
392 void emit_frontfacing(struct brw_compile
*p
,
393 const struct brw_reg
*dst
,
395 void emit_linterp(struct brw_compile
*p
,
396 const struct brw_reg
*dst
,
398 const struct brw_reg
*arg0
,
399 const struct brw_reg
*deltas
);
400 void emit_lrp(struct brw_compile
*p
,
401 const struct brw_reg
*dst
,
403 const struct brw_reg
*arg0
,
404 const struct brw_reg
*arg1
,
405 const struct brw_reg
*arg2
);
406 void emit_mad(struct brw_compile
*p
,
407 const struct brw_reg
*dst
,
409 const struct brw_reg
*arg0
,
410 const struct brw_reg
*arg1
,
411 const struct brw_reg
*arg2
);
412 void emit_math1(struct brw_wm_compile
*c
,
414 const struct brw_reg
*dst
,
416 const struct brw_reg
*arg0
);
417 void emit_math2(struct brw_wm_compile
*c
,
419 const struct brw_reg
*dst
,
421 const struct brw_reg
*arg0
,
422 const struct brw_reg
*arg1
);
423 void emit_min(struct brw_compile
*p
,
424 const struct brw_reg
*dst
,
426 const struct brw_reg
*arg0
,
427 const struct brw_reg
*arg1
);
428 void emit_max(struct brw_compile
*p
,
429 const struct brw_reg
*dst
,
431 const struct brw_reg
*arg0
,
432 const struct brw_reg
*arg1
);
433 void emit_pinterp(struct brw_compile
*p
,
434 const struct brw_reg
*dst
,
436 const struct brw_reg
*arg0
,
437 const struct brw_reg
*deltas
,
438 const struct brw_reg
*w
);
439 void emit_pixel_xy(struct brw_wm_compile
*c
,
440 const struct brw_reg
*dst
,
442 void emit_pixel_w(struct brw_wm_compile
*c
,
443 const struct brw_reg
*dst
,
445 const struct brw_reg
*arg0
,
446 const struct brw_reg
*deltas
);
447 void emit_sop(struct brw_compile
*p
,
448 const struct brw_reg
*dst
,
451 const struct brw_reg
*arg0
,
452 const struct brw_reg
*arg1
);
453 void emit_sign(struct brw_compile
*p
,
454 const struct brw_reg
*dst
,
456 const struct brw_reg
*arg0
);
457 void emit_tex(struct brw_wm_compile
*c
,
461 struct brw_reg depth_payload
,
465 void emit_txb(struct brw_wm_compile
*c
,
469 struct brw_reg depth_payload
,
472 void emit_wpos_xy(struct brw_wm_compile
*c
,
473 const struct brw_reg
*dst
,
475 const struct brw_reg
*arg0
);
476 void emit_xpd(struct brw_compile
*p
,
477 const struct brw_reg
*dst
,
479 const struct brw_reg
*arg0
,
480 const struct brw_reg
*arg1
);
482 GLboolean
brw_link_shader(struct gl_context
*ctx
, struct gl_shader_program
*prog
);
483 struct gl_shader
*brw_new_shader(struct gl_context
*ctx
, GLuint name
, GLuint type
);
484 struct gl_shader_program
*brw_new_shader_program(struct gl_context
*ctx
, GLuint name
);
486 bool brw_color_buffer_write_enabled(struct brw_context
*brw
);
487 bool brw_render_target_supported(gl_format format
);
488 void brw_wm_payload_setup(struct brw_context
*brw
,
489 struct brw_wm_compile
*c
);
490 bool do_wm_prog(struct brw_context
*brw
,
491 struct gl_shader_program
*prog
,
492 struct brw_fragment_program
*fp
,
493 struct brw_wm_prog_key
*key
);