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>
33 #include "main/glheader.h"
34 #include "main/macros.h"
35 #include "main/enums.h"
36 #include "brw_context.h"
40 #include "program/prog_parameter.h"
41 #include "program/prog_print.h"
42 #include "program/prog_statevars.h"
45 /** An invalid texture target */
46 #define TEX_TARGET_NONE NUM_TEXTURE_TARGETS
48 /** An invalid texture unit */
49 #define TEX_UNIT_NONE BRW_MAX_TEX_UNIT
51 #define FIRST_INTERNAL_TEMP MAX_NV_FRAGMENT_PROGRAM_TEMPS
59 static const char *wm_opcode_strings
[] = {
72 static const char *wm_file_strings
[] = {
78 /***********************************************************************
82 static struct prog_src_register
src_reg(GLuint file
, GLuint idx
)
84 struct prog_src_register reg
;
87 reg
.Swizzle
= SWIZZLE_NOOP
;
89 reg
.Negate
= NEGATE_NONE
;
97 static struct prog_src_register
src_reg_from_dst(struct prog_dst_register dst
)
99 return src_reg(dst
.File
, dst
.Index
);
102 static struct prog_src_register
src_undef( void )
104 return src_reg(PROGRAM_UNDEFINED
, 0);
107 static GLboolean
src_is_undef(struct prog_src_register src
)
109 return src
.File
== PROGRAM_UNDEFINED
;
112 static struct prog_src_register
src_swizzle( struct prog_src_register reg
, int x
, int y
, int z
, int w
)
114 reg
.Swizzle
= MAKE_SWIZZLE4(x
,y
,z
,w
);
118 static struct prog_src_register
src_swizzle1( struct prog_src_register reg
, int x
)
120 return src_swizzle(reg
, x
, x
, x
, x
);
123 static struct prog_src_register
src_swizzle4( struct prog_src_register reg
, uint swizzle
)
125 reg
.Swizzle
= swizzle
;
130 /***********************************************************************
134 static struct prog_dst_register
dst_reg(GLuint file
, GLuint idx
)
136 struct prog_dst_register reg
;
139 reg
.WriteMask
= WRITEMASK_XYZW
;
141 reg
.CondMask
= COND_TR
;
147 static struct prog_dst_register
dst_mask( struct prog_dst_register reg
, int mask
)
149 reg
.WriteMask
&= mask
;
153 static struct prog_dst_register
dst_undef( void )
155 return dst_reg(PROGRAM_UNDEFINED
, 0);
160 static struct prog_dst_register
get_temp( struct brw_wm_compile
*c
)
162 int bit
= _mesa_ffs( ~c
->fp_temp
);
165 printf("%s: out of temporaries\n", __FILE__
);
169 c
->fp_temp
|= 1<<(bit
-1);
170 return dst_reg(PROGRAM_TEMPORARY
, FIRST_INTERNAL_TEMP
+(bit
-1));
174 static void release_temp( struct brw_wm_compile
*c
, struct prog_dst_register temp
)
176 c
->fp_temp
&= ~(1 << (temp
.Index
- FIRST_INTERNAL_TEMP
));
180 /***********************************************************************
184 static struct prog_instruction
*get_fp_inst(struct brw_wm_compile
*c
)
186 assert(c
->nr_fp_insns
< BRW_WM_MAX_INSN
);
187 memset(&c
->prog_instructions
[c
->nr_fp_insns
], 0,
188 sizeof(*c
->prog_instructions
));
189 return &c
->prog_instructions
[c
->nr_fp_insns
++];
192 static struct prog_instruction
*emit_insn(struct brw_wm_compile
*c
,
193 const struct prog_instruction
*inst0
)
195 struct prog_instruction
*inst
= get_fp_inst(c
);
200 static struct prog_instruction
* emit_tex_op(struct brw_wm_compile
*c
,
202 struct prog_dst_register dest
,
205 GLuint tex_src_target
,
207 struct prog_src_register src0
,
208 struct prog_src_register src1
,
209 struct prog_src_register src2
)
211 struct prog_instruction
*inst
= get_fp_inst(c
);
213 assert(tex_src_unit
< BRW_MAX_TEX_UNIT
||
214 tex_src_unit
== TEX_UNIT_NONE
);
215 assert(tex_src_target
< NUM_TEXTURE_TARGETS
||
216 tex_src_target
== TEX_TARGET_NONE
);
218 /* update mask of which texture units are referenced by this program */
219 if (tex_src_unit
!= TEX_UNIT_NONE
)
220 c
->fp
->tex_units_used
|= (1 << tex_src_unit
);
222 memset(inst
, 0, sizeof(*inst
));
226 inst
->SaturateMode
= saturate
;
227 inst
->TexSrcUnit
= tex_src_unit
;
228 inst
->TexSrcTarget
= tex_src_target
;
229 inst
->TexShadow
= tex_shadow
;
230 inst
->SrcReg
[0] = src0
;
231 inst
->SrcReg
[1] = src1
;
232 inst
->SrcReg
[2] = src2
;
237 static struct prog_instruction
* emit_op(struct brw_wm_compile
*c
,
239 struct prog_dst_register dest
,
241 struct prog_src_register src0
,
242 struct prog_src_register src1
,
243 struct prog_src_register src2
)
245 return emit_tex_op(c
, op
, dest
, saturate
,
246 TEX_UNIT_NONE
, TEX_TARGET_NONE
, 0, /* unit, tgt, shadow */
251 /* Many Mesa opcodes produce the same value across all the result channels.
252 * We'd rather not have to support that splatting in the opcode implementations,
253 * and brw_wm_pass*.c wants to optimize them out by shuffling references around
254 * anyway. We can easily get both by emitting the opcode to one channel, and
255 * then MOVing it to the others, which brw_wm_pass*.c already understands.
257 static struct prog_instruction
*emit_scalar_insn(struct brw_wm_compile
*c
,
258 const struct prog_instruction
*inst0
)
260 struct prog_instruction
*inst
;
261 unsigned int dst_chan
;
262 unsigned int other_channel_mask
;
264 if (inst0
->DstReg
.WriteMask
== 0)
267 dst_chan
= _mesa_ffs(inst0
->DstReg
.WriteMask
) - 1;
268 inst
= get_fp_inst(c
);
270 inst
->DstReg
.WriteMask
= 1 << dst_chan
;
272 other_channel_mask
= inst0
->DstReg
.WriteMask
& ~(1 << dst_chan
);
273 if (other_channel_mask
!= 0) {
276 dst_mask(inst0
->DstReg
, other_channel_mask
),
278 src_swizzle1(src_reg_from_dst(inst0
->DstReg
), dst_chan
),
286 /***********************************************************************
287 * Special instructions for interpolation and other tasks
290 static struct prog_src_register
get_pixel_xy( struct brw_wm_compile
*c
)
292 if (src_is_undef(c
->pixel_xy
)) {
293 struct prog_dst_register pixel_xy
= get_temp(c
);
294 struct prog_src_register payload_r0_depth
= src_reg(PROGRAM_PAYLOAD
, PAYLOAD_DEPTH
);
297 /* Emit the out calculations, and hold onto the results. Use
298 * two instructions as a temporary is required.
300 /* pixel_xy.xy = PIXELXY payload[0];
304 dst_mask(pixel_xy
, WRITEMASK_XY
),
310 c
->pixel_xy
= src_reg_from_dst(pixel_xy
);
316 static struct prog_src_register
get_delta_xy( struct brw_wm_compile
*c
)
318 if (src_is_undef(c
->delta_xy
)) {
319 struct prog_dst_register delta_xy
= get_temp(c
);
320 struct prog_src_register pixel_xy
= get_pixel_xy(c
);
321 struct prog_src_register payload_r0_depth
= src_reg(PROGRAM_PAYLOAD
, PAYLOAD_DEPTH
);
323 /* deltas.xy = DELTAXY pixel_xy, payload[0]
327 dst_mask(delta_xy
, WRITEMASK_XY
),
333 c
->delta_xy
= src_reg_from_dst(delta_xy
);
339 static struct prog_src_register
get_pixel_w( struct brw_wm_compile
*c
)
341 /* This is only called for producing 1/w in pre-gen6 interp. for
342 * gen6, the interp opcodes don't use this argument.
344 if (c
->func
.brw
->intel
.gen
>= 6)
347 if (src_is_undef(c
->pixel_w
)) {
348 struct prog_dst_register pixel_w
= get_temp(c
);
349 struct prog_src_register deltas
= get_delta_xy(c
);
350 struct prog_src_register interp_wpos
= src_reg(PROGRAM_PAYLOAD
, FRAG_ATTRIB_WPOS
);
352 /* deltas.xyw = DELTAS2 deltas.xy, payload.interp_wpos.x
356 dst_mask(pixel_w
, WRITEMASK_W
),
363 c
->pixel_w
= src_reg_from_dst(pixel_w
);
369 static void emit_interp( struct brw_wm_compile
*c
,
372 struct prog_dst_register dst
= dst_reg(PROGRAM_INPUT
, idx
);
373 struct prog_src_register interp
= src_reg(PROGRAM_PAYLOAD
, idx
);
374 struct prog_src_register deltas
;
376 if (c
->func
.brw
->intel
.gen
< 6) {
377 deltas
= get_delta_xy(c
);
379 deltas
= src_undef();
382 /* Need to use PINTERP on attributes which have been
383 * multiplied by 1/W in the SF program, and LINTERP on those
387 case FRAG_ATTRIB_WPOS
:
388 /* Have to treat wpos.xy specially:
392 dst_mask(dst
, WRITEMASK_XY
),
398 dst
= dst_mask(dst
, WRITEMASK_ZW
);
400 /* PROGRAM_INPUT.attr.xyzw = INTERP payload.interp[attr].x, deltas.xyw
410 case FRAG_ATTRIB_COL0
:
411 case FRAG_ATTRIB_COL1
:
412 if (c
->key
.flat_shade
) {
422 if (c
->key
.linear_color
) {
432 /* perspective-corrected color interpolation */
443 case FRAG_ATTRIB_FOGC
:
444 /* Interpolate the fog coordinate */
447 dst_mask(dst
, WRITEMASK_X
),
455 dst_mask(dst
, WRITEMASK_YZW
),
466 case FRAG_ATTRIB_FACE
:
469 dst_mask(dst
, WRITEMASK_X
),
476 case FRAG_ATTRIB_PNTC
:
477 /* XXX review/test this case */
480 dst_mask(dst
, WRITEMASK_XY
),
488 dst_mask(dst
, WRITEMASK_ZW
),
510 c
->fp_interp_emitted
|= 1<<idx
;
513 /***********************************************************************
514 * Hacks to extend the program parameter and constant lists.
517 /* Add the fog parameters to the parameter list of the original
518 * program, rather than creating a new list. Doesn't really do any
519 * harm and it's not as if the parameter handling isn't a big hack
522 static struct prog_src_register
search_or_add_param5(struct brw_wm_compile
*c
,
529 struct gl_program_parameter_list
*paramList
= c
->fp
->program
.Base
.Parameters
;
530 gl_state_index tokens
[STATE_LENGTH
];
538 idx
= _mesa_add_state_reference( paramList
, tokens
);
540 return src_reg(PROGRAM_STATE_VAR
, idx
);
544 static struct prog_src_register
search_or_add_const4f( struct brw_wm_compile
*c
,
550 struct gl_program_parameter_list
*paramList
= c
->fp
->program
.Base
.Parameters
;
554 struct prog_src_register reg
;
561 idx
= _mesa_add_unnamed_constant( paramList
, values
, 4, &swizzle
);
562 reg
= src_reg(PROGRAM_STATE_VAR
, idx
);
563 reg
.Swizzle
= swizzle
;
570 /***********************************************************************
571 * Expand various instructions here to simpler forms.
573 static void precalc_dst( struct brw_wm_compile
*c
,
574 const struct prog_instruction
*inst
)
576 struct prog_src_register src0
= inst
->SrcReg
[0];
577 struct prog_src_register src1
= inst
->SrcReg
[1];
578 struct prog_dst_register dst
= inst
->DstReg
;
580 if (dst
.WriteMask
& WRITEMASK_Y
) {
581 /* dst.y = mul src0.y, src1.y
585 dst_mask(dst
, WRITEMASK_Y
),
592 if (dst
.WriteMask
& WRITEMASK_XZ
) {
593 struct prog_instruction
*swz
;
594 GLuint z
= GET_SWZ(src0
.Swizzle
, Z
);
596 /* dst.xz = swz src0.1zzz
600 dst_mask(dst
, WRITEMASK_XZ
),
602 src_swizzle(src0
, SWIZZLE_ONE
, z
, z
, z
),
605 /* Avoid letting negation flag of src0 affect our 1 constant. */
606 swz
->SrcReg
[0].Negate
&= ~NEGATE_X
;
608 if (dst
.WriteMask
& WRITEMASK_W
) {
609 /* dst.w = mov src1.w
613 dst_mask(dst
, WRITEMASK_W
),
622 static void precalc_lit( struct brw_wm_compile
*c
,
623 const struct prog_instruction
*inst
)
625 struct prog_src_register src0
= inst
->SrcReg
[0];
626 struct prog_dst_register dst
= inst
->DstReg
;
628 if (dst
.WriteMask
& WRITEMASK_XW
) {
629 struct prog_instruction
*swz
;
631 /* dst.xw = swz src0.1111
635 dst_mask(dst
, WRITEMASK_XW
),
637 src_swizzle1(src0
, SWIZZLE_ONE
),
640 /* Avoid letting the negation flag of src0 affect our 1 constant. */
641 swz
->SrcReg
[0].Negate
= NEGATE_NONE
;
644 if (dst
.WriteMask
& WRITEMASK_YZ
) {
647 dst_mask(dst
, WRITEMASK_YZ
),
657 * Some TEX instructions require extra code, cube map coordinate
658 * normalization, or coordinate scaling for RECT textures, etc.
659 * This function emits those extra instructions and the TEX
660 * instruction itself.
662 static void precalc_tex( struct brw_wm_compile
*c
,
663 const struct prog_instruction
*inst
)
665 struct prog_src_register coord
;
666 struct prog_dst_register tmpcoord
= { 0 };
667 const GLuint unit
= c
->fp
->program
.Base
.SamplerUnits
[inst
->TexSrcUnit
];
669 assert(unit
< BRW_MAX_TEX_UNIT
);
671 if (inst
->TexSrcTarget
== TEXTURE_CUBE_INDEX
) {
672 struct prog_instruction
*out
;
673 struct prog_dst_register tmp0
= get_temp(c
);
674 struct prog_src_register tmp0src
= src_reg_from_dst(tmp0
);
675 struct prog_dst_register tmp1
= get_temp(c
);
676 struct prog_src_register tmp1src
= src_reg_from_dst(tmp1
);
677 struct prog_src_register src0
= inst
->SrcReg
[0];
679 /* find longest component of coord vector and normalize it */
680 tmpcoord
= get_temp(c
);
681 coord
= src_reg_from_dst(tmpcoord
);
683 /* tmpcoord = src0 (i.e.: coord = src0) */
684 out
= emit_op(c
, OPCODE_MOV
,
690 out
->SrcReg
[0].Negate
= NEGATE_NONE
;
691 out
->SrcReg
[0].Abs
= 1;
693 /* tmp0 = MAX(coord.X, coord.Y) */
694 emit_op(c
, OPCODE_MAX
,
697 src_swizzle1(coord
, X
),
698 src_swizzle1(coord
, Y
),
701 /* tmp1 = MAX(tmp0, coord.Z) */
702 emit_op(c
, OPCODE_MAX
,
706 src_swizzle1(coord
, Z
),
709 /* tmp0 = 1 / tmp1 */
710 emit_op(c
, OPCODE_RCP
,
711 dst_mask(tmp0
, WRITEMASK_X
),
717 /* tmpCoord = src0 * tmp0 */
718 emit_op(c
, OPCODE_MUL
,
722 src_swizzle1(tmp0src
, SWIZZLE_X
),
725 release_temp(c
, tmp0
);
726 release_temp(c
, tmp1
);
728 else if (inst
->TexSrcTarget
== TEXTURE_RECT_INDEX
) {
729 struct prog_src_register scale
=
730 search_or_add_param5( c
,
736 tmpcoord
= get_temp(c
);
738 /* coord.xy = MUL inst->SrcReg[0], { 1/width, 1/height }
752 coord
= src_reg_from_dst(tmpcoord
);
755 coord
= inst
->SrcReg
[0];
758 /* Need to emit YUV texture conversions by hand. Probably need to
759 * do this here - the alternative is in brw_wm_emit.c, but the
760 * conversion requires allocating a temporary variable which we
761 * don't have the facility to do that late in the compilation.
763 if (c
->key
.yuvtex_mask
& (1 << unit
)) {
764 /* convert ycbcr to RGBA */
765 GLboolean swap_uv
= c
->key
.yuvtex_swap_mask
& (1<<unit
);
768 CONST C0 = { -.5, -.0625, -.5, 1.164 }
769 CONST C1 = { 1.596, -0.813, 2.018, -.391 }
771 UYV.xyz = ADD UYV, C0
772 UYV.y = MUL UYV.y, C0.w
774 RGB.xyz = MAD UYV.zzx, C1, UYV.y
776 RGB.xyz = MAD UYV.xxz, C1, UYV.y
777 RGB.y = MAD UYV.z, C1.w, RGB.y
779 struct prog_dst_register dst
= inst
->DstReg
;
780 struct prog_dst_register tmp
= get_temp(c
);
781 struct prog_src_register tmpsrc
= src_reg_from_dst(tmp
);
782 struct prog_src_register C0
= search_or_add_const4f( c
, -.5, -.0625, -.5, 1.164 );
783 struct prog_src_register C1
= search_or_add_const4f( c
, 1.596, -0.813, 2.018, -.391 );
798 /* tmp.xyz = ADD TMP, C0
802 dst_mask(tmp
, WRITEMASK_XYZ
),
808 /* YUV.y = MUL YUV.y, C0.w
813 dst_mask(tmp
, WRITEMASK_Y
),
821 * RGB.xyz = MAD YUV.zzx, C1, YUV.y
823 * RGB.xyz = MAD YUV.xxz, C1, YUV.y
828 dst_mask(dst
, WRITEMASK_XYZ
),
830 swap_uv
?src_swizzle(tmpsrc
, Z
,Z
,X
,X
):src_swizzle(tmpsrc
, X
,X
,Z
,Z
),
832 src_swizzle1(tmpsrc
, Y
));
834 /* RGB.y = MAD YUV.z, C1.w, RGB.y
838 dst_mask(dst
, WRITEMASK_Y
),
840 src_swizzle1(tmpsrc
, Z
),
842 src_swizzle1(src_reg_from_dst(dst
), Y
));
844 release_temp(c
, tmp
);
847 /* ordinary RGBA tex instruction */
860 /* For GL_EXT_texture_swizzle: */
861 if (c
->key
.tex_swizzles
[unit
] != SWIZZLE_NOOP
) {
862 /* swizzle the result of the TEX instruction */
863 struct prog_src_register tmpsrc
= src_reg_from_dst(inst
->DstReg
);
864 emit_op(c
, OPCODE_SWZ
,
866 SATURATE_OFF
, /* saturate already done above */
867 src_swizzle4(tmpsrc
, c
->key
.tex_swizzles
[unit
]),
872 if ((inst
->TexSrcTarget
== TEXTURE_RECT_INDEX
) ||
873 (inst
->TexSrcTarget
== TEXTURE_CUBE_INDEX
))
874 release_temp(c
, tmpcoord
);
879 * Check if the given TXP instruction really needs the divide-by-W step.
881 static GLboolean
projtex( struct brw_wm_compile
*c
,
882 const struct prog_instruction
*inst
)
884 const struct prog_src_register src
= inst
->SrcReg
[0];
887 assert(inst
->Opcode
== OPCODE_TXP
);
889 /* Only try to detect the simplest cases. Could detect (later)
890 * cases where we are trying to emit code like RCP {1.0}, MUL x,
893 * More complex cases than this typically only arise from
894 * user-provided fragment programs anyway:
896 if (inst
->TexSrcTarget
== TEXTURE_CUBE_INDEX
)
897 retVal
= GL_FALSE
; /* ut2004 gun rendering !?! */
898 else if (src
.File
== PROGRAM_INPUT
&&
899 GET_SWZ(src
.Swizzle
, W
) == W
&&
900 (c
->key
.proj_attrib_mask
& (1 << src
.Index
)) == 0)
912 static void precalc_txp( struct brw_wm_compile
*c
,
913 const struct prog_instruction
*inst
)
915 struct prog_src_register src0
= inst
->SrcReg
[0];
917 if (projtex(c
, inst
)) {
918 struct prog_dst_register tmp
= get_temp(c
);
919 struct prog_instruction tmp_inst
;
921 /* tmp0.w = RCP inst.arg[0][3]
925 dst_mask(tmp
, WRITEMASK_W
),
927 src_swizzle1(src0
, GET_SWZ(src0
.Swizzle
, W
)),
931 /* tmp0.xyz = MUL inst.arg[0], tmp0.wwww
935 dst_mask(tmp
, WRITEMASK_XYZ
),
938 src_swizzle1(src_reg_from_dst(tmp
), W
),
941 /* dst = precalc(TEX tmp0)
944 tmp_inst
.SrcReg
[0] = src_reg_from_dst(tmp
);
945 precalc_tex(c
, &tmp_inst
);
947 release_temp(c
, tmp
);
951 /* dst = precalc(TEX src0)
953 precalc_tex(c
, inst
);
959 static void emit_render_target_writes( struct brw_wm_compile
*c
)
961 struct prog_src_register payload_r0_depth
= src_reg(PROGRAM_PAYLOAD
, PAYLOAD_DEPTH
);
962 struct prog_src_register outdepth
= src_reg(PROGRAM_OUTPUT
, FRAG_RESULT_DEPTH
);
963 struct prog_src_register outcolor
;
966 struct prog_instruction
*inst
, *last_inst
= NULL
;
968 /* The inst->Aux field is used for FB write target and the EOT marker */
970 if (c
->key
.nr_color_regions
> 1) {
971 for (i
= 0 ; i
< c
->key
.nr_color_regions
; i
++) {
972 outcolor
= src_reg(PROGRAM_OUTPUT
, FRAG_RESULT_DATA0
+ i
);
973 last_inst
= inst
= emit_op(c
, WM_FB_WRITE
, dst_mask(dst_undef(), 0),
974 0, outcolor
, payload_r0_depth
, outdepth
);
975 inst
->Aux
= INST_AUX_TARGET(i
);
976 if (c
->fp_fragcolor_emitted
) {
977 outcolor
= src_reg(PROGRAM_OUTPUT
, FRAG_RESULT_COLOR
);
978 last_inst
= inst
= emit_op(c
, WM_FB_WRITE
, dst_mask(dst_undef(), 0),
979 0, outcolor
, payload_r0_depth
, outdepth
);
980 inst
->Aux
= INST_AUX_TARGET(i
);
983 last_inst
->Aux
|= INST_AUX_EOT
;
986 /* if gl_FragData[0] is written, use it, else use gl_FragColor */
987 if (c
->fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DATA0
))
988 outcolor
= src_reg(PROGRAM_OUTPUT
, FRAG_RESULT_DATA0
);
990 outcolor
= src_reg(PROGRAM_OUTPUT
, FRAG_RESULT_COLOR
);
992 inst
= emit_op(c
, WM_FB_WRITE
, dst_mask(dst_undef(),0),
993 0, outcolor
, payload_r0_depth
, outdepth
);
994 inst
->Aux
= INST_AUX_EOT
| INST_AUX_TARGET(0);
1001 /***********************************************************************
1002 * Emit INTERP instructions ahead of first use of each attrib.
1005 static void validate_src_regs( struct brw_wm_compile
*c
,
1006 const struct prog_instruction
*inst
)
1008 GLuint nr_args
= brw_wm_nr_args( inst
->Opcode
);
1011 for (i
= 0; i
< nr_args
; i
++) {
1012 if (inst
->SrcReg
[i
].File
== PROGRAM_INPUT
) {
1013 GLuint idx
= inst
->SrcReg
[i
].Index
;
1014 if (!(c
->fp_interp_emitted
& (1<<idx
))) {
1015 emit_interp(c
, idx
);
1021 static void validate_dst_regs( struct brw_wm_compile
*c
,
1022 const struct prog_instruction
*inst
)
1024 if (inst
->DstReg
.File
== PROGRAM_OUTPUT
) {
1025 GLuint idx
= inst
->DstReg
.Index
;
1026 if (idx
== FRAG_RESULT_COLOR
)
1027 c
->fp_fragcolor_emitted
= 1;
1031 static void print_insns( const struct prog_instruction
*insn
,
1035 for (i
= 0; i
< nr
; i
++, insn
++) {
1037 if (insn
->Opcode
< MAX_OPCODE
)
1038 _mesa_fprint_instruction_opt(stdout
, insn
, 0, PROG_PRINT_DEBUG
, NULL
);
1039 else if (insn
->Opcode
< MAX_WM_OPCODE
) {
1040 GLuint idx
= insn
->Opcode
- MAX_OPCODE
;
1042 _mesa_fprint_alu_instruction(stdout
, insn
, wm_opcode_strings
[idx
],
1043 3, PROG_PRINT_DEBUG
, NULL
);
1046 printf("965 Opcode %d\n", insn
->Opcode
);
1052 * Initial pass for fragment program code generation.
1053 * This function is used by both the GLSL and non-GLSL paths.
1055 void brw_wm_pass_fp( struct brw_wm_compile
*c
)
1057 struct intel_context
*intel
= &c
->func
.brw
->intel
;
1058 struct brw_fragment_program
*fp
= c
->fp
;
1061 if (unlikely(INTEL_DEBUG
& DEBUG_WM
)) {
1062 printf("pre-fp:\n");
1063 _mesa_fprint_program_opt(stdout
, &fp
->program
.Base
, PROG_PRINT_DEBUG
,
1068 c
->pixel_xy
= src_undef();
1069 if (intel
->gen
>= 6) {
1070 /* The interpolation deltas come in as the perspective pixel
1071 * location barycentric params.
1073 c
->delta_xy
= src_reg(PROGRAM_PAYLOAD
, PAYLOAD_DEPTH
);
1075 c
->delta_xy
= src_undef();
1077 c
->pixel_w
= src_undef();
1079 c
->fp
->tex_units_used
= 0x0;
1081 /* Emit preamble instructions. This is where special instructions such as
1082 * WM_CINTERP, WM_LINTERP, WM_PINTERP and WM_WPOSXY are emitted to
1083 * compute shader inputs from varying vars.
1085 for (insn
= 0; insn
< fp
->program
.Base
.NumInstructions
; insn
++) {
1086 const struct prog_instruction
*inst
= &fp
->program
.Base
.Instructions
[insn
];
1087 validate_src_regs(c
, inst
);
1088 validate_dst_regs(c
, inst
);
1091 /* Loop over all instructions doing assorted simplifications and
1094 for (insn
= 0; insn
< fp
->program
.Base
.NumInstructions
; insn
++) {
1095 const struct prog_instruction
*inst
= &fp
->program
.Base
.Instructions
[insn
];
1096 struct prog_instruction
*out
;
1098 /* Check for INPUT values, emit INTERP instructions where
1102 switch (inst
->Opcode
) {
1104 out
= emit_insn(c
, inst
);
1105 out
->Opcode
= OPCODE_MOV
;
1109 out
= emit_insn(c
, inst
);
1110 out
->Opcode
= OPCODE_MOV
;
1111 out
->SrcReg
[0].Negate
= NEGATE_NONE
;
1112 out
->SrcReg
[0].Abs
= 1;
1116 out
= emit_insn(c
, inst
);
1117 out
->Opcode
= OPCODE_ADD
;
1118 out
->SrcReg
[1].Negate
^= NEGATE_XYZW
;
1122 out
= emit_insn(c
, inst
);
1123 /* This should probably be done in the parser.
1125 out
->DstReg
.WriteMask
&= WRITEMASK_XY
;
1129 precalc_dst(c
, inst
);
1133 precalc_lit(c
, inst
);
1137 precalc_tex(c
, inst
);
1141 precalc_txp(c
, inst
);
1145 out
= emit_insn(c
, inst
);
1146 out
->TexSrcUnit
= fp
->program
.Base
.SamplerUnits
[inst
->TexSrcUnit
];
1147 assert(out
->TexSrcUnit
< BRW_MAX_TEX_UNIT
);
1151 out
= emit_insn(c
, inst
);
1152 /* This should probably be done in the parser.
1154 out
->DstReg
.WriteMask
&= WRITEMASK_XYZ
;
1158 out
= emit_insn(c
, inst
);
1159 /* This should probably be done in the parser.
1161 out
->DstReg
.WriteMask
= 0;
1164 emit_render_target_writes(c
);
1169 if (brw_wm_is_scalar_result(inst
->Opcode
))
1170 emit_scalar_insn(c
, inst
);
1177 if (unlikely(INTEL_DEBUG
& DEBUG_WM
)) {
1178 printf("pass_fp:\n");
1179 print_insns( c
->prog_instructions
, c
->nr_fp_insns
);