2 * Copyright © 2015 Intel Corporation
3 * Copyright © 2014-2015 Broadcom
4 * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
22 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
27 #include "nir/nir_builder.h"
28 #include "glsl/list.h"
29 #include "main/imports.h"
30 #include "util/ralloc.h"
32 #include "prog_to_nir.h"
33 #include "prog_instruction.h"
34 #include "prog_parameter.h"
35 #include "prog_print.h"
40 * A translator from Mesa IR (prog_instruction.h) to NIR. This is primarily
41 * intended to support ARB_vertex_program, ARB_fragment_program, and fixed-function
42 * vertex processing. Full GLSL support should use glsl_to_nir instead.
46 const struct gl_program
*prog
;
50 nir_variable
*input_vars
[VARYING_SLOT_MAX
];
51 nir_variable
*output_vars
[VARYING_SLOT_MAX
];
52 nir_register
**output_regs
;
53 nir_register
**temp_regs
;
55 nir_register
*addr_reg
;
58 #define SWIZ(X, Y, Z, W) \
59 (unsigned[4]){ SWIZZLE_##X, SWIZZLE_##Y, SWIZZLE_##Z, SWIZZLE_##W }
60 #define ptn_swizzle(b, src, x, y, z, w) nir_swizzle(b, src, SWIZ(x, y, z, w), 4, true)
61 #define ptn_channel(b, src, ch) nir_swizzle(b, src, SWIZ(ch, ch, ch, ch), 1, true)
64 ptn_src_for_dest(struct ptn_compile
*c
, nir_alu_dest
*dest
)
66 nir_builder
*b
= &c
->build
;
69 memset(&src
, 0, sizeof(src
));
71 if (dest
->dest
.is_ssa
)
72 src
.src
= nir_src_for_ssa(&dest
->dest
.ssa
);
74 assert(!dest
->dest
.reg
.indirect
);
75 src
.src
= nir_src_for_reg(dest
->dest
.reg
.reg
);
76 src
.src
.reg
.base_offset
= dest
->dest
.reg
.base_offset
;
79 for (int i
= 0; i
< 4; i
++)
82 return nir_fmov_alu(b
, src
, 4);
86 ptn_get_dest(struct ptn_compile
*c
, const struct prog_dst_register
*prog_dst
)
90 memset(&dest
, 0, sizeof(dest
));
92 switch (prog_dst
->File
) {
93 case PROGRAM_TEMPORARY
:
94 dest
.dest
.reg
.reg
= c
->temp_regs
[prog_dst
->Index
];
97 dest
.dest
.reg
.reg
= c
->output_regs
[prog_dst
->Index
];
100 assert(prog_dst
->Index
== 0);
101 dest
.dest
.reg
.reg
= c
->addr_reg
;
103 case PROGRAM_UNDEFINED
:
107 dest
.write_mask
= prog_dst
->WriteMask
;
108 dest
.saturate
= false;
110 assert(!prog_dst
->RelAddr
);
116 * Multiply the contents of the ADDR register by 4 to convert from the number
117 * of vec4s to the number of floating point components.
120 ptn_addr_reg_value(struct ptn_compile
*c
)
122 nir_builder
*b
= &c
->build
;
124 memset(&src
, 0, sizeof(src
));
125 src
.src
= nir_src_for_reg(c
->addr_reg
);
127 return nir_imul(b
, nir_fmov_alu(b
, src
, 1), nir_imm_int(b
, 4));
131 ptn_get_src(struct ptn_compile
*c
, const struct prog_src_register
*prog_src
)
133 nir_builder
*b
= &c
->build
;
136 memset(&src
, 0, sizeof(src
));
138 switch (prog_src
->File
) {
139 case PROGRAM_UNDEFINED
:
140 return nir_imm_float(b
, 0.0);
141 case PROGRAM_TEMPORARY
:
142 assert(!prog_src
->RelAddr
&& prog_src
->Index
>= 0);
143 src
.src
.reg
.reg
= c
->temp_regs
[prog_src
->Index
];
145 case PROGRAM_INPUT
: {
146 /* ARB_vertex_program doesn't allow relative addressing on vertex
147 * attributes; ARB_fragment_program has no relative addressing at all.
149 assert(!prog_src
->RelAddr
);
151 assert(prog_src
->Index
>= 0 && prog_src
->Index
< VARYING_SLOT_MAX
);
153 nir_intrinsic_instr
*load
=
154 nir_intrinsic_instr_create(b
->shader
, nir_intrinsic_load_var
);
155 load
->num_components
= 4;
156 load
->variables
[0] = nir_deref_var_create(load
, c
->input_vars
[prog_src
->Index
]);
158 nir_ssa_dest_init(&load
->instr
, &load
->dest
, 4, NULL
);
159 nir_instr_insert_after_cf_list(b
->cf_node_list
, &load
->instr
);
161 src
.src
= nir_src_for_ssa(&load
->dest
.ssa
);
164 case PROGRAM_STATE_VAR
:
165 case PROGRAM_CONSTANT
: {
166 /* We actually want to look at the type in the Parameters list for this,
167 * because it lets us upload constant builtin uniforms as actual
170 struct gl_program_parameter_list
*plist
= c
->prog
->Parameters
;
171 gl_register_file file
= prog_src
->RelAddr
? prog_src
->File
:
172 plist
->Parameters
[prog_src
->Index
].Type
;
175 case PROGRAM_CONSTANT
:
176 if ((c
->prog
->IndirectRegisterFiles
& (1 << PROGRAM_CONSTANT
)) == 0) {
177 float *v
= (float *) plist
->ParameterValues
[prog_src
->Index
];
178 src
.src
= nir_src_for_ssa(nir_imm_vec4(b
, v
[0], v
[1], v
[2], v
[3]));
182 case PROGRAM_STATE_VAR
: {
183 nir_intrinsic_op load_op
=
184 prog_src
->RelAddr
? nir_intrinsic_load_uniform_indirect
:
185 nir_intrinsic_load_uniform
;
186 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(b
->shader
, load_op
);
187 nir_ssa_dest_init(&load
->instr
, &load
->dest
, 4, NULL
);
188 load
->num_components
= 4;
190 /* Multiply src->Index by 4 to scale from # of vec4s to components. */
191 load
->const_index
[0] = 4 * prog_src
->Index
;
192 load
->const_index
[1] = 1;
194 if (prog_src
->RelAddr
) {
195 nir_ssa_def
*reladdr
= ptn_addr_reg_value(c
);
196 if (prog_src
->Index
< 0) {
197 /* This is a negative offset which should be added to the address
200 reladdr
= nir_iadd(b
, reladdr
, nir_imm_int(b
, load
->const_index
[0]));
201 load
->const_index
[0] = 0;
203 load
->src
[0] = nir_src_for_ssa(reladdr
);
206 nir_instr_insert_after_cf_list(b
->cf_node_list
, &load
->instr
);
208 src
.src
= nir_src_for_ssa(&load
->dest
.ssa
);
212 fprintf(stderr
, "bad uniform src register file: %s (%d)\n",
213 _mesa_register_file_name(file
), file
);
219 fprintf(stderr
, "unknown src register file: %s (%d)\n",
220 _mesa_register_file_name(prog_src
->File
), prog_src
->File
);
225 if (!HAS_EXTENDED_SWIZZLE(prog_src
->Swizzle
) &&
226 (prog_src
->Negate
== NEGATE_NONE
|| prog_src
->Negate
== NEGATE_XYZW
)) {
227 /* The simple non-SWZ case. */
228 for (int i
= 0; i
< 4; i
++)
229 src
.swizzle
[i
] = GET_SWZ(prog_src
->Swizzle
, i
);
231 def
= nir_fmov_alu(b
, src
, 4);
234 def
= nir_fabs(b
, def
);
236 if (prog_src
->Negate
)
237 def
= nir_fneg(b
, def
);
239 /* The SWZ instruction allows per-component zero/one swizzles, and also
240 * per-component negation.
242 nir_ssa_def
*chans
[4];
243 for (int i
= 0; i
< 4; i
++) {
244 int swizzle
= GET_SWZ(prog_src
->Swizzle
, i
);
245 if (swizzle
== SWIZZLE_ZERO
) {
246 chans
[i
] = nir_imm_float(b
, 0.0);
247 } else if (swizzle
== SWIZZLE_ONE
) {
248 chans
[i
] = nir_imm_float(b
, 1.0);
250 assert(swizzle
!= SWIZZLE_NIL
);
251 nir_alu_instr
*mov
= nir_alu_instr_create(b
->shader
, nir_op_fmov
);
252 nir_ssa_dest_init(&mov
->instr
, &mov
->dest
.dest
, 1, NULL
);
253 mov
->dest
.write_mask
= 0x1;
255 mov
->src
[0].swizzle
[0] = swizzle
;
256 nir_instr_insert_after_cf_list(b
->cf_node_list
, &mov
->instr
);
258 chans
[i
] = &mov
->dest
.dest
.ssa
;
262 chans
[i
] = nir_fabs(b
, chans
[i
]);
264 if (prog_src
->Negate
& (1 << i
))
265 chans
[i
] = nir_fneg(b
, chans
[i
]);
267 def
= nir_vec4(b
, chans
[0], chans
[1], chans
[2], chans
[3]);
274 ptn_alu(nir_builder
*b
, nir_op op
, nir_alu_dest dest
, nir_ssa_def
**src
)
276 unsigned num_srcs
= nir_op_infos
[op
].num_inputs
;
277 nir_alu_instr
*instr
= nir_alu_instr_create(b
->shader
, op
);
280 for (i
= 0; i
< num_srcs
; i
++)
281 instr
->src
[i
].src
= nir_src_for_ssa(src
[i
]);
284 nir_instr_insert_after_cf_list(b
->cf_node_list
, &instr
->instr
);
288 ptn_move_dest_masked(nir_builder
*b
, nir_alu_dest dest
,
289 nir_ssa_def
*def
, unsigned write_mask
)
291 if (!(dest
.write_mask
& write_mask
))
294 nir_alu_instr
*mov
= nir_alu_instr_create(b
->shader
, nir_op_fmov
);
299 mov
->dest
.write_mask
&= write_mask
;
300 mov
->src
[0].src
= nir_src_for_ssa(def
);
301 for (unsigned i
= def
->num_components
; i
< 4; i
++)
302 mov
->src
[0].swizzle
[i
] = def
->num_components
- 1;
303 nir_instr_insert_after_cf_list(b
->cf_node_list
, &mov
->instr
);
307 ptn_move_dest(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
*def
)
309 ptn_move_dest_masked(b
, dest
, def
, WRITEMASK_XYZW
);
313 ptn_arl(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
315 ptn_move_dest(b
, dest
, nir_f2i(b
, nir_ffloor(b
, src
[0])));
318 /* EXP - Approximate Exponential Base 2
319 * dst.x = 2^{\lfloor src.x\rfloor}
320 * dst.y = src.x - \lfloor src.x\rfloor
325 ptn_exp(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
327 nir_ssa_def
*srcx
= ptn_channel(b
, src
[0], X
);
329 ptn_move_dest_masked(b
, dest
, nir_fexp2(b
, nir_ffloor(b
, srcx
)), WRITEMASK_X
);
330 ptn_move_dest_masked(b
, dest
, nir_fsub(b
, srcx
, nir_ffloor(b
, srcx
)), WRITEMASK_Y
);
331 ptn_move_dest_masked(b
, dest
, nir_fexp2(b
, srcx
), WRITEMASK_Z
);
332 ptn_move_dest_masked(b
, dest
, nir_imm_float(b
, 1.0), WRITEMASK_W
);
335 /* LOG - Approximate Logarithm Base 2
336 * dst.x = \lfloor\log_2{|src.x|}\rfloor
337 * dst.y = |src.x| * 2^{-\lfloor\log_2{|src.x|}\rfloor}}
338 * dst.z = \log_2{|src.x|}
342 ptn_log(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
344 nir_ssa_def
*abs_srcx
= nir_fabs(b
, ptn_channel(b
, src
[0], X
));
345 nir_ssa_def
*log2
= nir_flog2(b
, abs_srcx
);
346 nir_ssa_def
*floor_log2
= nir_ffloor(b
, log2
);
348 ptn_move_dest_masked(b
, dest
, floor_log2
, WRITEMASK_X
);
349 ptn_move_dest_masked(b
, dest
,
350 nir_fmul(b
, abs_srcx
,
351 nir_fexp2(b
, nir_fneg(b
, floor_log2
))),
353 ptn_move_dest_masked(b
, dest
, log2
, WRITEMASK_Z
);
354 ptn_move_dest_masked(b
, dest
, nir_imm_float(b
, 1.0), WRITEMASK_W
);
357 /* DST - Distance Vector
359 * dst.y = src0.y \times src1.y
364 ptn_dst(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
366 ptn_move_dest_masked(b
, dest
, nir_imm_float(b
, 1.0), WRITEMASK_X
);
367 ptn_move_dest_masked(b
, dest
, nir_fmul(b
, src
[0], src
[1]), WRITEMASK_Y
);
368 ptn_move_dest_masked(b
, dest
, nir_fmov(b
, src
[0]), WRITEMASK_Z
);
369 ptn_move_dest_masked(b
, dest
, nir_fmov(b
, src
[1]), WRITEMASK_W
);
372 /* LIT - Light Coefficients
374 * dst.y = max(src.x, 0.0)
375 * dst.z = (src.x > 0.0) ? max(src.y, 0.0)^{clamp(src.w, -128.0, 128.0))} : 0
379 ptn_lit(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
381 ptn_move_dest_masked(b
, dest
, nir_imm_float(b
, 1.0), WRITEMASK_XW
);
383 ptn_move_dest_masked(b
, dest
, nir_fmax(b
, ptn_channel(b
, src
[0], X
),
384 nir_imm_float(b
, 0.0)), WRITEMASK_Y
);
386 if (dest
.write_mask
& WRITEMASK_Z
) {
387 nir_ssa_def
*src0_y
= ptn_channel(b
, src
[0], Y
);
388 nir_ssa_def
*wclamp
= nir_fmax(b
, nir_fmin(b
, ptn_channel(b
, src
[0], W
),
389 nir_imm_float(b
, 128.0)),
390 nir_imm_float(b
, -128.0));
391 nir_ssa_def
*pow
= nir_fpow(b
, nir_fmax(b
, src0_y
, nir_imm_float(b
, 0.0)),
395 if (b
->shader
->options
->native_integers
) {
397 nir_fge(b
, nir_imm_float(b
, 0.0), ptn_channel(b
, src
[0], X
)),
398 nir_imm_float(b
, 0.0),
402 nir_sge(b
, nir_imm_float(b
, 0.0), ptn_channel(b
, src
[0], X
)),
403 nir_imm_float(b
, 0.0),
407 ptn_move_dest_masked(b
, dest
, z
, WRITEMASK_Z
);
412 * dst.x = \cos{src.x}
413 * dst.y = \sin{src.x}
418 ptn_scs(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
420 ptn_move_dest_masked(b
, dest
, nir_fcos(b
, ptn_channel(b
, src
[0], X
)),
422 ptn_move_dest_masked(b
, dest
, nir_fsin(b
, ptn_channel(b
, src
[0], X
)),
424 ptn_move_dest_masked(b
, dest
, nir_imm_float(b
, 0.0), WRITEMASK_Z
);
425 ptn_move_dest_masked(b
, dest
, nir_imm_float(b
, 1.0), WRITEMASK_W
);
429 * Emit SLT. For platforms with integers, prefer b2f(flt(...)).
432 ptn_slt(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
434 if (b
->shader
->options
->native_integers
) {
435 ptn_move_dest(b
, dest
, nir_b2f(b
, nir_flt(b
, src
[0], src
[1])));
437 ptn_move_dest(b
, dest
, nir_slt(b
, src
[0], src
[1]));
442 * Emit SGE. For platforms with integers, prefer b2f(fge(...)).
445 ptn_sge(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
447 if (b
->shader
->options
->native_integers
) {
448 ptn_move_dest(b
, dest
, nir_b2f(b
, nir_fge(b
, src
[0], src
[1])));
450 ptn_move_dest(b
, dest
, nir_sge(b
, src
[0], src
[1]));
455 ptn_sle(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
457 nir_ssa_def
*commuted
[] = { src
[1], src
[0] };
458 ptn_sge(b
, dest
, commuted
);
462 ptn_sgt(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
464 nir_ssa_def
*commuted
[] = { src
[1], src
[0] };
465 ptn_slt(b
, dest
, commuted
);
469 * Emit SEQ. For platforms with integers, prefer b2f(feq(...)).
472 ptn_seq(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
474 if (b
->shader
->options
->native_integers
) {
475 ptn_move_dest(b
, dest
, nir_b2f(b
, nir_feq(b
, src
[0], src
[1])));
477 ptn_move_dest(b
, dest
, nir_seq(b
, src
[0], src
[1]));
482 * Emit SNE. For platforms with integers, prefer b2f(fne(...)).
485 ptn_sne(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
487 if (b
->shader
->options
->native_integers
) {
488 ptn_move_dest(b
, dest
, nir_b2f(b
, nir_fne(b
, src
[0], src
[1])));
490 ptn_move_dest(b
, dest
, nir_sne(b
, src
[0], src
[1]));
495 ptn_xpd(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
497 ptn_move_dest_masked(b
, dest
,
500 ptn_swizzle(b
, src
[0], Y
, Z
, X
, X
),
501 ptn_swizzle(b
, src
[1], Z
, X
, Y
, X
)),
503 ptn_swizzle(b
, src
[1], Y
, Z
, X
, X
),
504 ptn_swizzle(b
, src
[0], Z
, X
, Y
, X
))),
506 ptn_move_dest_masked(b
, dest
, nir_imm_float(b
, 1.0), WRITEMASK_W
);
510 ptn_dp2(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
512 ptn_move_dest(b
, dest
, nir_fdot2(b
, src
[0], src
[1]));
516 ptn_dp3(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
518 ptn_move_dest(b
, dest
, nir_fdot3(b
, src
[0], src
[1]));
522 ptn_dp4(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
524 ptn_move_dest(b
, dest
, nir_fdot4(b
, src
[0], src
[1]));
528 ptn_dph(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
530 nir_ssa_def
*dp3
= nir_fdot3(b
, src
[0], src
[1]);
531 ptn_move_dest(b
, dest
, nir_fadd(b
, dp3
, ptn_channel(b
, src
[1], W
)));
535 ptn_cmp(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
537 if (b
->shader
->options
->native_integers
) {
538 ptn_move_dest(b
, dest
, nir_bcsel(b
,
539 nir_flt(b
, src
[0], nir_imm_float(b
, 0.0)),
542 ptn_move_dest(b
, dest
, nir_fcsel(b
,
543 nir_slt(b
, src
[0], nir_imm_float(b
, 0.0)),
549 ptn_lrp(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
551 ptn_move_dest(b
, dest
, nir_flrp(b
, src
[2], src
[1], src
[0]));
555 ptn_kil(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
)
557 nir_ssa_def
*cmp
= b
->shader
->options
->native_integers
?
558 nir_bany4(b
, nir_flt(b
, src
[0], nir_imm_float(b
, 0.0))) :
559 nir_fany4(b
, nir_slt(b
, src
[0], nir_imm_float(b
, 0.0)));
561 nir_intrinsic_instr
*discard
=
562 nir_intrinsic_instr_create(b
->shader
, nir_intrinsic_discard_if
);
563 discard
->src
[0] = nir_src_for_ssa(cmp
);
564 nir_instr_insert_after_cf_list(b
->cf_node_list
, &discard
->instr
);
568 ptn_tex(nir_builder
*b
, nir_alu_dest dest
, nir_ssa_def
**src
,
569 struct prog_instruction
*prog_inst
)
571 nir_tex_instr
*instr
;
575 switch (prog_inst
->Opcode
) {
597 assert(!"not handled");
602 fprintf(stderr
, "unknown tex op %d\n", prog_inst
->Opcode
);
606 if (prog_inst
->TexShadow
)
609 instr
= nir_tex_instr_create(b
->shader
, num_srcs
);
611 instr
->dest_type
= nir_type_float
;
612 instr
->is_shadow
= prog_inst
->TexShadow
;
613 instr
->sampler_index
= prog_inst
->TexSrcUnit
;
615 switch (prog_inst
->TexSrcTarget
) {
616 case TEXTURE_1D_INDEX
:
617 instr
->sampler_dim
= GLSL_SAMPLER_DIM_1D
;
619 case TEXTURE_2D_INDEX
:
620 instr
->sampler_dim
= GLSL_SAMPLER_DIM_2D
;
622 case TEXTURE_3D_INDEX
:
623 instr
->sampler_dim
= GLSL_SAMPLER_DIM_3D
;
625 case TEXTURE_CUBE_INDEX
:
626 instr
->sampler_dim
= GLSL_SAMPLER_DIM_CUBE
;
628 case TEXTURE_RECT_INDEX
:
629 instr
->sampler_dim
= GLSL_SAMPLER_DIM_RECT
;
632 fprintf(stderr
, "Unknown texture target %d\n", prog_inst
->TexSrcTarget
);
636 switch (instr
->sampler_dim
) {
637 case GLSL_SAMPLER_DIM_1D
:
638 case GLSL_SAMPLER_DIM_BUF
:
639 instr
->coord_components
= 1;
641 case GLSL_SAMPLER_DIM_2D
:
642 case GLSL_SAMPLER_DIM_RECT
:
643 case GLSL_SAMPLER_DIM_EXTERNAL
:
644 case GLSL_SAMPLER_DIM_MS
:
645 instr
->coord_components
= 2;
647 case GLSL_SAMPLER_DIM_3D
:
648 case GLSL_SAMPLER_DIM_CUBE
:
649 instr
->coord_components
= 3;
653 unsigned src_number
= 0;
655 instr
->src
[src_number
].src
=
656 nir_src_for_ssa(ptn_swizzle(b
, src
[0], X
, Y
, Z
, W
));
657 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
660 if (prog_inst
->Opcode
== OPCODE_TXP
) {
661 instr
->src
[src_number
].src
= nir_src_for_ssa(ptn_channel(b
, src
[0], W
));
662 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
666 if (prog_inst
->Opcode
== OPCODE_TXB
) {
667 instr
->src
[src_number
].src
= nir_src_for_ssa(ptn_channel(b
, src
[0], W
));
668 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
672 if (prog_inst
->Opcode
== OPCODE_TXL
) {
673 instr
->src
[src_number
].src
= nir_src_for_ssa(ptn_channel(b
, src
[0], W
));
674 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
678 if (instr
->is_shadow
) {
679 if (instr
->coord_components
< 3)
680 instr
->src
[src_number
].src
= nir_src_for_ssa(ptn_channel(b
, src
[0], Z
));
682 instr
->src
[src_number
].src
= nir_src_for_ssa(ptn_channel(b
, src
[0], W
));
684 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
688 assert(src_number
== num_srcs
);
690 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 4, NULL
);
691 nir_instr_insert_after_cf_list(b
->cf_node_list
, &instr
->instr
);
693 /* Resolve the writemask on the texture op. */
694 ptn_move_dest(b
, dest
, &instr
->dest
.ssa
);
697 static const nir_op op_trans
[MAX_OPCODE
] = {
699 [OPCODE_ABS
] = nir_op_fabs
,
700 [OPCODE_ADD
] = nir_op_fadd
,
703 [OPCODE_COS
] = nir_op_fcos
,
704 [OPCODE_DDX
] = nir_op_fddx
,
705 [OPCODE_DDY
] = nir_op_fddy
,
712 [OPCODE_EX2
] = nir_op_fexp2
,
714 [OPCODE_FLR
] = nir_op_ffloor
,
715 [OPCODE_FRC
] = nir_op_ffract
,
716 [OPCODE_LG2
] = nir_op_flog2
,
720 [OPCODE_MAD
] = nir_op_ffma
,
721 [OPCODE_MAX
] = nir_op_fmax
,
722 [OPCODE_MIN
] = nir_op_fmin
,
723 [OPCODE_MOV
] = nir_op_fmov
,
724 [OPCODE_MUL
] = nir_op_fmul
,
725 [OPCODE_POW
] = nir_op_fpow
,
726 [OPCODE_RCP
] = nir_op_frcp
,
728 [OPCODE_RSQ
] = nir_op_frsq
,
733 [OPCODE_SIN
] = nir_op_fsin
,
737 [OPCODE_SSG
] = nir_op_fsign
,
738 [OPCODE_SUB
] = nir_op_fsub
,
741 [OPCODE_TRUNC
] = nir_op_ftrunc
,
751 ptn_emit_instruction(struct ptn_compile
*c
, struct prog_instruction
*prog_inst
)
753 nir_builder
*b
= &c
->build
;
755 const unsigned op
= prog_inst
->Opcode
;
757 if (op
== OPCODE_END
)
761 for (i
= 0; i
< 3; i
++) {
762 src
[i
] = ptn_get_src(c
, &prog_inst
->SrcReg
[i
]);
764 nir_alu_dest dest
= ptn_get_dest(c
, &prog_inst
->DstReg
);
770 ptn_move_dest(b
, dest
, nir_frsq(b
, ptn_channel(b
, src
[0], X
)));
774 ptn_move_dest(b
, dest
, nir_frcp(b
, ptn_channel(b
, src
[0], X
)));
778 ptn_move_dest(b
, dest
, nir_fexp2(b
, ptn_channel(b
, src
[0], X
)));
782 ptn_move_dest(b
, dest
, nir_flog2(b
, ptn_channel(b
, src
[0], X
)));
786 ptn_move_dest(b
, dest
, nir_fpow(b
,
787 ptn_channel(b
, src
[0], X
),
788 ptn_channel(b
, src
[1], X
)));
792 ptn_move_dest(b
, dest
, nir_fcos(b
, ptn_channel(b
, src
[0], X
)));
796 ptn_move_dest(b
, dest
, nir_fsin(b
, ptn_channel(b
, src
[0], X
)));
800 ptn_arl(b
, dest
, src
);
804 ptn_exp(b
, dest
, src
);
808 ptn_log(b
, dest
, src
);
812 ptn_lrp(b
, dest
, src
);
816 ptn_dst(b
, dest
, src
);
820 ptn_lit(b
, dest
, src
);
824 ptn_xpd(b
, dest
, src
);
828 ptn_dp2(b
, dest
, src
);
832 ptn_dp3(b
, dest
, src
);
836 ptn_dp4(b
, dest
, src
);
840 ptn_dph(b
, dest
, src
);
844 ptn_kil(b
, dest
, src
);
848 ptn_cmp(b
, dest
, src
);
852 ptn_scs(b
, dest
, src
);
856 ptn_slt(b
, dest
, src
);
860 ptn_sgt(b
, dest
, src
);
864 ptn_sle(b
, dest
, src
);
868 ptn_sge(b
, dest
, src
);
872 ptn_seq(b
, dest
, src
);
876 ptn_sne(b
, dest
, src
);
885 ptn_tex(b
, dest
, src
, prog_inst
);
889 /* Extended swizzles were already handled in ptn_get_src(). */
890 ptn_alu(b
, nir_op_fmov
, dest
, src
);
897 if (op_trans
[op
] != 0 || op
== OPCODE_MOV
) {
898 ptn_alu(b
, op_trans
[op
], dest
, src
);
900 fprintf(stderr
, "unknown opcode: %s\n", _mesa_opcode_string(op
));
906 if (prog_inst
->SaturateMode
) {
907 assert(prog_inst
->SaturateMode
== SATURATE_ZERO_ONE
);
908 assert(!dest
.dest
.is_ssa
);
909 ptn_move_dest(b
, dest
, nir_fsat(b
, ptn_src_for_dest(c
, &dest
)));
914 * Puts a NIR intrinsic to store of each PROGRAM_OUTPUT value to the output
915 * variables at the end of the shader.
917 * We don't generate these incrementally as the PROGRAM_OUTPUT values are
918 * written, because there's no output load intrinsic, which means we couldn't
922 ptn_add_output_stores(struct ptn_compile
*c
)
924 nir_builder
*b
= &c
->build
;
926 foreach_list_typed(nir_variable
, var
, node
, &b
->shader
->outputs
) {
927 nir_intrinsic_instr
*store
=
928 nir_intrinsic_instr_create(b
->shader
, nir_intrinsic_store_var
);
929 store
->num_components
= 4;
930 store
->variables
[0] =
931 nir_deref_var_create(store
, c
->output_vars
[var
->data
.location
]);
932 store
->src
[0].reg
.reg
= c
->output_regs
[var
->data
.location
];
933 nir_instr_insert_after_cf_list(c
->build
.cf_node_list
, &store
->instr
);
938 setup_registers_and_variables(struct ptn_compile
*c
)
940 nir_builder
*b
= &c
->build
;
941 struct nir_shader
*shader
= b
->shader
;
943 /* Create input variables. */
944 const int num_inputs
= _mesa_flsll(c
->prog
->InputsRead
);
945 for (int i
= 0; i
< num_inputs
; i
++) {
946 if (!(c
->prog
->InputsRead
& BITFIELD64_BIT(i
)))
948 nir_variable
*var
= rzalloc(shader
, nir_variable
);
949 var
->type
= glsl_vec4_type();
950 var
->data
.read_only
= true;
951 var
->data
.mode
= nir_var_shader_in
;
952 var
->name
= ralloc_asprintf(var
, "in_%d", i
);
953 var
->data
.location
= i
;
956 if (c
->prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
) {
957 struct gl_fragment_program
*fp
=
958 (struct gl_fragment_program
*) c
->prog
;
960 var
->data
.interpolation
= fp
->InterpQualifier
[i
];
962 if (i
== VARYING_SLOT_POS
) {
963 var
->data
.origin_upper_left
= fp
->OriginUpperLeft
;
964 var
->data
.pixel_center_integer
= fp
->PixelCenterInteger
;
965 } else if (i
== VARYING_SLOT_FOGC
) {
966 /* fogcoord is defined as <f, 0.0, 0.0, 1.0>. Make the actual
967 * input variable a float, and create a local containing the
970 var
->type
= glsl_float_type();
972 nir_intrinsic_instr
*load_x
=
973 nir_intrinsic_instr_create(shader
, nir_intrinsic_load_var
);
974 load_x
->num_components
= 1;
975 load_x
->variables
[0] = nir_deref_var_create(load_x
, var
);
976 nir_ssa_dest_init(&load_x
->instr
, &load_x
->dest
, 1, NULL
);
977 nir_instr_insert_after_cf_list(b
->cf_node_list
, &load_x
->instr
);
979 nir_ssa_def
*f001
= nir_vec4(b
, &load_x
->dest
.ssa
, nir_imm_float(b
, 0.0),
980 nir_imm_float(b
, 0.0), nir_imm_float(b
, 1.0));
982 nir_variable
*fullvar
= rzalloc(shader
, nir_variable
);
983 fullvar
->type
= glsl_vec4_type();
984 fullvar
->data
.mode
= nir_var_local
;
985 fullvar
->name
= "fogcoord_tmp";
986 exec_list_push_tail(&b
->impl
->locals
, &fullvar
->node
);
988 nir_intrinsic_instr
*store
=
989 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
990 store
->num_components
= 4;
991 store
->variables
[0] = nir_deref_var_create(store
, fullvar
);
992 store
->src
[0] = nir_src_for_ssa(f001
);
993 nir_instr_insert_after_cf_list(b
->cf_node_list
, &store
->instr
);
995 /* Insert the real input into the list so the driver has real
996 * inputs, but set c->input_vars[i] to the temporary so we use
997 * the splatted value.
999 exec_list_push_tail(&shader
->inputs
, &var
->node
);
1000 c
->input_vars
[i
] = fullvar
;
1005 exec_list_push_tail(&shader
->inputs
, &var
->node
);
1006 c
->input_vars
[i
] = var
;
1009 /* Create output registers and variables. */
1010 int max_outputs
= _mesa_fls(c
->prog
->OutputsWritten
);
1011 c
->output_regs
= rzalloc_array(c
, nir_register
*, max_outputs
);
1013 for (int i
= 0; i
< max_outputs
; i
++) {
1014 if (!(c
->prog
->OutputsWritten
& BITFIELD64_BIT(i
)))
1017 /* Since we can't load from outputs in the IR, we make temporaries
1018 * for the outputs and emit stores to the real outputs at the end of
1021 nir_register
*reg
= nir_local_reg_create(b
->impl
);
1022 reg
->num_components
= 4;
1024 nir_variable
*var
= rzalloc(shader
, nir_variable
);
1025 var
->type
= glsl_vec4_type();
1026 var
->data
.mode
= nir_var_shader_out
;
1027 var
->name
= ralloc_asprintf(var
, "out_%d", i
);
1029 var
->data
.location
= i
;
1030 var
->data
.index
= 0;
1032 c
->output_regs
[i
] = reg
;
1034 exec_list_push_tail(&shader
->outputs
, &var
->node
);
1035 c
->output_vars
[i
] = var
;
1038 /* Create temporary registers. */
1039 c
->temp_regs
= rzalloc_array(c
, nir_register
*, c
->prog
->NumTemporaries
);
1042 for (int i
= 0; i
< c
->prog
->NumTemporaries
; i
++) {
1043 reg
= nir_local_reg_create(b
->impl
);
1048 reg
->num_components
= 4;
1049 c
->temp_regs
[i
] = reg
;
1052 /* Create the address register (for ARB_vertex_program). */
1053 reg
= nir_local_reg_create(b
->impl
);
1058 reg
->num_components
= 1;
1061 /* Set the number of uniforms */
1062 shader
->num_uniforms
= 4 * c
->prog
->Parameters
->NumParameters
;
1066 prog_to_nir(const struct gl_program
*prog
, const nir_shader_compiler_options
*options
)
1068 struct ptn_compile
*c
;
1069 struct nir_shader
*s
;
1071 c
= rzalloc(NULL
, struct ptn_compile
);
1074 s
= nir_shader_create(NULL
, options
);
1079 nir_function
*func
= nir_function_create(s
, "main");
1080 nir_function_overload
*overload
= nir_function_overload_create(func
);
1081 nir_function_impl
*impl
= nir_function_impl_create(overload
);
1083 c
->build
.shader
= s
;
1084 c
->build
.impl
= impl
;
1085 c
->build
.cf_node_list
= &impl
->body
;
1087 setup_registers_and_variables(c
);
1088 if (unlikely(c
->error
))
1091 for (unsigned int i
= 0; i
< prog
->NumInstructions
; i
++) {
1092 ptn_emit_instruction(c
, &prog
->Instructions
[i
]);
1094 if (unlikely(c
->error
))
1098 ptn_add_output_stores(c
);