2 * Copyright © 2015 Red Hat
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
25 #include "nir_builder.h"
26 #include "program/prog_instruction.h"
28 /* Lower gl_FragCoord (and fddy) to account for driver's requested coordinate-
29 * origin and pixel-center vs. shader. If transformation is required, a
30 * gl_FbWposYTransform uniform is inserted (with the specified state-slots)
31 * and additional instructions are inserted to transform gl_FragCoord (and
34 * This is based on the logic in emit_wpos()/emit_wpos_adjustment() in TGSI
37 * Run before nir_lower_io.
41 const nir_lower_wpos_ytransform_options
*options
;
44 nir_variable
*transform
;
45 } lower_wpos_ytransform_state
;
48 get_transform(lower_wpos_ytransform_state
*state
)
50 if (state
->transform
== NULL
) {
51 /* NOTE: name must be prefixed w/ "gl_" to trigger slot based
52 * special handling in uniform setup:
54 nir_variable
*var
= nir_variable_create(state
->shader
,
57 "gl_FbWposYTransform");
59 var
->num_state_slots
= 1;
60 var
->state_slots
= ralloc_array(var
, nir_state_slot
, 1);
61 var
->state_slots
[0].swizzle
= SWIZZLE_XYZW
;
62 memcpy(var
->state_slots
[0].tokens
, state
->options
->state_tokens
,
63 sizeof(var
->state_slots
[0].tokens
));
65 state
->transform
= var
;
67 return nir_load_var(&state
->b
, state
->transform
);
70 /* NIR equiv of TGSI CMP instruction: */
72 nir_cmp(nir_builder
*b
, nir_ssa_def
*src0
, nir_ssa_def
*src1
, nir_ssa_def
*src2
)
74 return nir_bcsel(b
, nir_flt(b
, src0
, nir_imm_float(b
, 0.0)), src1
, src2
);
77 /* see emit_wpos_adjustment() in st_mesa_to_tgsi.c */
79 emit_wpos_adjustment(lower_wpos_ytransform_state
*state
,
80 nir_intrinsic_instr
*intr
, nir_variable
*fragcoord
,
81 bool invert
, float adjX
, float adjY
[2])
83 nir_builder
*b
= &state
->b
;
84 nir_ssa_def
*wpostrans
, *wpos_temp
, *wpos_temp_y
, *wpos_input
;
86 assert(intr
->dest
.is_ssa
);
88 b
->cursor
= nir_before_instr(&intr
->instr
);
90 wpostrans
= get_transform(state
);
91 wpos_input
= nir_load_var(b
, fragcoord
);
93 /* First, apply the coordinate shift: */
94 if (adjX
|| adjY
[0] || adjY
[1]) {
95 if (adjY
[0] != adjY
[1]) {
96 /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
97 * depending on whether inversion is actually going to be applied
98 * or not, which is determined by testing against the inversion
99 * state variable used below, which will be either +1 or -1.
101 nir_ssa_def
*adj_temp
;
103 adj_temp
= nir_cmp(b
,
104 nir_channel(b
, wpostrans
, invert
? 2 : 0),
105 nir_imm_vec4(b
, adjX
, adjY
[0], 0.0f
, 0.0f
),
106 nir_imm_vec4(b
, adjX
, adjY
[1], 0.0f
, 0.0f
));
108 wpos_temp
= nir_fadd(b
, wpos_input
, adj_temp
);
110 wpos_temp
= nir_fadd(b
,
112 nir_imm_vec4(b
, adjX
, adjY
[0], 0.0f
, 0.0f
));
114 wpos_input
= wpos_temp
;
116 /* MOV wpos_temp, input[wpos]
118 wpos_temp
= wpos_input
;
121 /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
122 * inversion/identity, or the other way around if we're drawing to an FBO.
125 /* wpos_temp.y = wpos_input * wpostrans.xxxx + wpostrans.yyyy */
126 wpos_temp_y
= nir_fadd(b
, nir_fmul(b
, nir_channel(b
, wpos_temp
, 1),
127 nir_channel(b
, wpostrans
, 0)),
128 nir_channel(b
, wpostrans
, 1));
130 /* wpos_temp.y = wpos_input * wpostrans.zzzz + wpostrans.wwww */
131 wpos_temp_y
= nir_fadd(b
, nir_fmul(b
, nir_channel(b
, wpos_temp
, 1),
132 nir_channel(b
, wpostrans
, 2)),
133 nir_channel(b
, wpostrans
, 3));
136 wpos_temp
= nir_vec4(b
,
137 nir_channel(b
, wpos_temp
, 0),
139 nir_channel(b
, wpos_temp
, 2),
140 nir_channel(b
, wpos_temp
, 3));
142 nir_ssa_def_rewrite_uses(&intr
->dest
.ssa
, nir_src_for_ssa(wpos_temp
));
146 lower_fragcoord(lower_wpos_ytransform_state
*state
,
147 nir_intrinsic_instr
*intr
, nir_variable
*fragcoord
)
149 const nir_lower_wpos_ytransform_options
*options
= state
->options
;
151 float adjY
[2] = { 0.0f
, 0.0f
};
154 /* Based on logic in emit_wpos():
156 * Query the pixel center conventions supported by the pipe driver and set
157 * adjX, adjY to help out if it cannot handle the requested one internally.
159 * The bias of the y-coordinate depends on whether y-inversion takes place
160 * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
161 * drawing to an FBO (causes additional inversion), and whether the pipe
162 * driver origin and the requested origin differ (the latter condition is
163 * stored in the 'invert' variable).
165 * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
172 * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
173 * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
174 * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
175 * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
177 * inversion and center shift:
178 * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
179 * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
180 * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
181 * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
184 if (fragcoord
->data
.origin_upper_left
) {
185 /* Fragment shader wants origin in upper-left */
186 if (options
->fs_coord_origin_upper_left
) {
187 /* the driver supports upper-left origin */
188 } else if (options
->fs_coord_origin_lower_left
) {
189 /* the driver supports lower-left origin, need to invert Y */
192 unreachable("invalid options");
195 /* Fragment shader wants origin in lower-left */
196 if (options
->fs_coord_origin_lower_left
) {
197 /* the driver supports lower-left origin */
198 } else if (options
->fs_coord_origin_upper_left
) {
199 /* the driver supports upper-left origin, need to invert Y */
202 unreachable("invalid options");
206 if (fragcoord
->data
.pixel_center_integer
) {
207 /* Fragment shader wants pixel center integer */
208 if (options
->fs_coord_pixel_center_integer
) {
209 /* the driver supports pixel center integer */
211 } else if (options
->fs_coord_pixel_center_half_integer
) {
212 /* the driver supports pixel center half integer, need to bias X,Y */
217 unreachable("invalid options");
220 /* Fragment shader wants pixel center half integer */
221 if (options
->fs_coord_pixel_center_half_integer
) {
222 /* the driver supports pixel center half integer */
223 } else if (options
->fs_coord_pixel_center_integer
) {
224 /* the driver supports pixel center integer, need to bias X,Y */
225 adjX
= adjY
[0] = adjY
[1] = 0.5f
;
227 unreachable("invalid options");
231 emit_wpos_adjustment(state
, intr
, fragcoord
, invert
, adjX
, adjY
);
235 lower_load_pointcoord(lower_wpos_ytransform_state
*state
,
236 nir_intrinsic_instr
*intr
)
238 nir_builder
*b
= &state
->b
;
239 b
->cursor
= nir_after_instr(&intr
->instr
);
241 nir_ssa_def
*pntc
= &intr
->dest
.ssa
;
242 nir_ssa_def
*transform
= get_transform(state
);
243 nir_ssa_def
*y
= nir_channel(b
, pntc
, 1);
244 /* The offset is 1 if we're flipping, 0 otherwise. */
245 nir_ssa_def
*offset
= nir_fmax(b
, nir_channel(b
, transform
, 2),
246 nir_imm_float(b
, 0.0));
247 /* Flip the sign of y if we're flipping. */
248 nir_ssa_def
*scaled
= nir_fmul(b
, y
, nir_channel(b
, transform
, 0));
250 /* Reassemble the vector. */
251 nir_ssa_def
*flipped_pntc
= nir_vec2(b
,
252 nir_channel(b
, pntc
, 0),
253 nir_fadd(b
, offset
, scaled
));
255 nir_ssa_def_rewrite_uses_after(&intr
->dest
.ssa
, nir_src_for_ssa(flipped_pntc
),
256 flipped_pntc
->parent_instr
);
259 /* turns 'fddy(p)' into 'fddy(fmul(p, transform.x))' */
261 lower_fddy(lower_wpos_ytransform_state
*state
, nir_alu_instr
*fddy
)
263 nir_builder
*b
= &state
->b
;
264 nir_ssa_def
*p
, *pt
, *trans
;
266 b
->cursor
= nir_before_instr(&fddy
->instr
);
268 p
= nir_ssa_for_alu_src(b
, fddy
, 0);
269 trans
= get_transform(state
);
270 pt
= nir_fmul(b
, p
, nir_channel(b
, trans
, 0));
272 nir_instr_rewrite_src(&fddy
->instr
,
274 nir_src_for_ssa(pt
));
276 for (unsigned i
= 0; i
< 4; i
++)
277 fddy
->src
[0].swizzle
[i
] = MIN2(i
, pt
->num_components
- 1);
280 /* Multiply interp_deref_at_offset's offset by transform.x to flip it. */
282 lower_interp_deref_at_offset(lower_wpos_ytransform_state
*state
,
283 nir_intrinsic_instr
*interp
)
285 nir_builder
*b
= &state
->b
;
289 b
->cursor
= nir_before_instr(&interp
->instr
);
291 offset
= nir_ssa_for_src(b
, interp
->src
[1], 2);
292 flip_y
= nir_fmul(b
, nir_channel(b
, offset
, 1),
293 nir_channel(b
, get_transform(state
), 0));
294 nir_instr_rewrite_src(&interp
->instr
, &interp
->src
[1],
295 nir_src_for_ssa(nir_vec2(b
, nir_channel(b
, offset
, 0),
300 lower_load_sample_pos(lower_wpos_ytransform_state
*state
,
301 nir_intrinsic_instr
*intr
)
303 nir_builder
*b
= &state
->b
;
304 b
->cursor
= nir_after_instr(&intr
->instr
);
306 nir_ssa_def
*pos
= &intr
->dest
.ssa
;
307 nir_ssa_def
*scale
= nir_channel(b
, get_transform(state
), 0);
308 nir_ssa_def
*neg_scale
= nir_channel(b
, get_transform(state
), 2);
309 /* Either y or 1-y for scale equal to 1 or -1 respectively. */
310 nir_ssa_def
*flipped_y
=
311 nir_fadd(b
, nir_fmax(b
, neg_scale
, nir_imm_float(b
, 0.0)),
312 nir_fmul(b
, nir_channel(b
, pos
, 1), scale
));
313 nir_ssa_def
*flipped_pos
= nir_vec2(b
, nir_channel(b
, pos
, 0), flipped_y
);
315 nir_ssa_def_rewrite_uses_after(&intr
->dest
.ssa
, nir_src_for_ssa(flipped_pos
),
316 flipped_pos
->parent_instr
);
320 lower_wpos_ytransform_block(lower_wpos_ytransform_state
*state
, nir_block
*block
)
322 nir_foreach_instr_safe(instr
, block
) {
323 if (instr
->type
== nir_instr_type_intrinsic
) {
324 nir_intrinsic_instr
*intr
= nir_instr_as_intrinsic(instr
);
325 if (intr
->intrinsic
== nir_intrinsic_load_deref
) {
326 nir_deref_instr
*deref
= nir_src_as_deref(intr
->src
[0]);
327 nir_variable
*var
= nir_deref_instr_get_variable(deref
);
329 if ((var
->data
.mode
== nir_var_shader_in
&&
330 var
->data
.location
== VARYING_SLOT_POS
) ||
331 (var
->data
.mode
== nir_var_system_value
&&
332 var
->data
.location
== SYSTEM_VALUE_FRAG_COORD
)) {
333 /* gl_FragCoord should not have array/struct derefs: */
334 lower_fragcoord(state
, intr
, var
);
335 } else if (var
->data
.mode
== nir_var_system_value
&&
336 var
->data
.location
== SYSTEM_VALUE_SAMPLE_POS
) {
337 lower_load_sample_pos(state
, intr
);
338 } else if (var
->data
.mode
== nir_var_shader_in
&&
339 var
->data
.location
== VARYING_SLOT_PNTC
&&
340 state
->shader
->options
->lower_wpos_pntc
) {
341 lower_load_pointcoord(state
, intr
);
343 } else if (intr
->intrinsic
== nir_intrinsic_load_frag_coord
) {
344 lower_fragcoord(state
, intr
, NULL
);
345 } else if (intr
->intrinsic
== nir_intrinsic_load_sample_pos
) {
346 lower_load_sample_pos(state
, intr
);
347 } else if (intr
->intrinsic
== nir_intrinsic_interp_deref_at_offset
) {
348 lower_interp_deref_at_offset(state
, intr
);
350 } else if (instr
->type
== nir_instr_type_alu
) {
351 nir_alu_instr
*alu
= nir_instr_as_alu(instr
);
352 if (alu
->op
== nir_op_fddy
||
353 alu
->op
== nir_op_fddy_fine
||
354 alu
->op
== nir_op_fddy_coarse
)
355 lower_fddy(state
, alu
);
361 lower_wpos_ytransform_impl(lower_wpos_ytransform_state
*state
, nir_function_impl
*impl
)
363 nir_builder_init(&state
->b
, impl
);
365 nir_foreach_block(block
, impl
) {
366 lower_wpos_ytransform_block(state
, block
);
368 nir_metadata_preserve(impl
, nir_metadata_block_index
|
369 nir_metadata_dominance
);
373 nir_lower_wpos_ytransform(nir_shader
*shader
,
374 const nir_lower_wpos_ytransform_options
*options
)
376 lower_wpos_ytransform_state state
= {
381 assert(shader
->info
.stage
== MESA_SHADER_FRAGMENT
);
383 nir_foreach_function(function
, shader
) {
385 lower_wpos_ytransform_impl(&state
, function
->impl
);
388 return state
.transform
!= NULL
;