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
,
81 bool invert
, float adjX
, float adjY
[2])
83 nir_builder
*b
= &state
->b
;
84 nir_variable
*fragcoord
= intr
->variables
[0]->var
;
85 nir_ssa_def
*wpostrans
, *wpos_temp
, *wpos_temp_y
, *wpos_input
;
87 assert(intr
->dest
.is_ssa
);
89 b
->cursor
= nir_before_instr(&intr
->instr
);
91 wpostrans
= get_transform(state
);
92 wpos_input
= nir_load_var(b
, fragcoord
);
94 /* First, apply the coordinate shift: */
95 if (adjX
|| adjY
[0] || adjY
[1]) {
96 if (adjY
[0] != adjY
[1]) {
97 /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
98 * depending on whether inversion is actually going to be applied
99 * or not, which is determined by testing against the inversion
100 * state variable used below, which will be either +1 or -1.
102 nir_ssa_def
*adj_temp
;
104 adj_temp
= nir_cmp(b
,
105 nir_channel(b
, wpostrans
, invert
? 2 : 0),
106 nir_imm_vec4(b
, adjX
, adjY
[0], 0.0f
, 0.0f
),
107 nir_imm_vec4(b
, adjX
, adjY
[1], 0.0f
, 0.0f
));
109 wpos_temp
= nir_fadd(b
, wpos_input
, adj_temp
);
111 wpos_temp
= nir_fadd(b
,
113 nir_imm_vec4(b
, adjX
, adjY
[0], 0.0f
, 0.0f
));
115 wpos_input
= wpos_temp
;
117 /* MOV wpos_temp, input[wpos]
119 wpos_temp
= wpos_input
;
122 /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
123 * inversion/identity, or the other way around if we're drawing to an FBO.
126 /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
128 wpos_temp_y
= nir_ffma(b
,
129 nir_channel(b
, wpos_temp
, 1),
130 nir_channel(b
, wpostrans
, 0),
131 nir_channel(b
, wpostrans
, 1));
133 /* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww
135 wpos_temp_y
= nir_ffma(b
,
136 nir_channel(b
, wpos_temp
, 1),
137 nir_channel(b
, wpostrans
, 2),
138 nir_channel(b
, wpostrans
, 3));
141 wpos_temp
= nir_vec4(b
,
142 nir_channel(b
, wpos_temp
, 0),
144 nir_channel(b
, wpos_temp
, 2),
145 nir_channel(b
, wpos_temp
, 3));
147 nir_ssa_def_rewrite_uses(&intr
->dest
.ssa
, nir_src_for_ssa(wpos_temp
));
151 lower_fragcoord(lower_wpos_ytransform_state
*state
, nir_intrinsic_instr
*intr
)
153 const nir_lower_wpos_ytransform_options
*options
= state
->options
;
154 nir_variable
*fragcoord
= intr
->variables
[0]->var
;
156 float adjY
[2] = { 0.0f
, 0.0f
};
159 /* Based on logic in emit_wpos():
161 * Query the pixel center conventions supported by the pipe driver and set
162 * adjX, adjY to help out if it cannot handle the requested one internally.
164 * The bias of the y-coordinate depends on whether y-inversion takes place
165 * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
166 * drawing to an FBO (causes additional inversion), and whether the the pipe
167 * driver origin and the requested origin differ (the latter condition is
168 * stored in the 'invert' variable).
170 * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
177 * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
178 * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
179 * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
180 * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
182 * inversion and center shift:
183 * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
184 * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
185 * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
186 * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
189 if (fragcoord
->data
.origin_upper_left
) {
190 /* Fragment shader wants origin in upper-left */
191 if (options
->fs_coord_origin_upper_left
) {
192 /* the driver supports upper-left origin */
193 } else if (options
->fs_coord_origin_lower_left
) {
194 /* the driver supports lower-left origin, need to invert Y */
197 unreachable("invalid options");
200 /* Fragment shader wants origin in lower-left */
201 if (options
->fs_coord_origin_lower_left
) {
202 /* the driver supports lower-left origin */
203 } else if (options
->fs_coord_origin_upper_left
) {
204 /* the driver supports upper-left origin, need to invert Y */
207 unreachable("invalid options");
211 if (fragcoord
->data
.pixel_center_integer
) {
212 /* Fragment shader wants pixel center integer */
213 if (options
->fs_coord_pixel_center_integer
) {
214 /* the driver supports pixel center integer */
216 } else if (options
->fs_coord_pixel_center_half_integer
) {
217 /* the driver supports pixel center half integer, need to bias X,Y */
222 unreachable("invalid options");
225 /* Fragment shader wants pixel center half integer */
226 if (options
->fs_coord_pixel_center_half_integer
) {
227 /* the driver supports pixel center half integer */
228 } else if (options
->fs_coord_pixel_center_integer
) {
229 /* the driver supports pixel center integer, need to bias X,Y */
230 adjX
= adjY
[0] = adjY
[1] = 0.5f
;
232 unreachable("invalid options");
236 emit_wpos_adjustment(state
, intr
, invert
, adjX
, adjY
);
239 /* turns 'fddy(p)' into 'fddy(fmul(p, transform.x))' */
241 lower_fddy(lower_wpos_ytransform_state
*state
, nir_alu_instr
*fddy
)
243 nir_builder
*b
= &state
->b
;
244 nir_ssa_def
*p
, *pt
, *trans
;
246 b
->cursor
= nir_before_instr(&fddy
->instr
);
248 p
= nir_ssa_for_alu_src(b
, fddy
, 0);
249 trans
= get_transform(state
);
250 pt
= nir_fmul(b
, p
, nir_channel(b
, trans
, 0));
252 nir_instr_rewrite_src(&fddy
->instr
,
254 nir_src_for_ssa(pt
));
256 for (unsigned i
= 0; i
< 4; i
++)
257 fddy
->src
[0].swizzle
[i
] = MIN2(i
, pt
->num_components
- 1);
261 lower_wpos_ytransform_block(lower_wpos_ytransform_state
*state
, nir_block
*block
)
263 nir_foreach_instr_safe(instr
, block
) {
264 if (instr
->type
== nir_instr_type_intrinsic
) {
265 nir_intrinsic_instr
*intr
= nir_instr_as_intrinsic(instr
);
266 if (intr
->intrinsic
== nir_intrinsic_load_var
) {
267 nir_deref_var
*dvar
= intr
->variables
[0];
268 nir_variable
*var
= dvar
->var
;
270 if (strcmp(var
->name
, "gl_FragCoord") == 0) {
271 /* gl_FragCoord should not have array/struct deref's: */
272 assert(dvar
->deref
.child
== NULL
);
273 lower_fragcoord(state
, intr
);
276 } else if (instr
->type
== nir_instr_type_alu
) {
277 nir_alu_instr
*alu
= nir_instr_as_alu(instr
);
278 if (alu
->op
== nir_op_fddy
)
279 lower_fddy(state
, alu
);
287 lower_wpos_ytransform_impl(lower_wpos_ytransform_state
*state
, nir_function_impl
*impl
)
289 nir_builder_init(&state
->b
, impl
);
291 nir_foreach_block(block
, impl
) {
292 lower_wpos_ytransform_block(state
, block
);
294 nir_metadata_preserve(impl
, nir_metadata_block_index
|
295 nir_metadata_dominance
);
299 nir_lower_wpos_ytransform(nir_shader
*shader
,
300 const nir_lower_wpos_ytransform_options
*options
)
302 lower_wpos_ytransform_state state
= {
307 assert(shader
->stage
== MESA_SHADER_FRAGMENT
);
309 nir_foreach_function(function
, shader
) {
311 lower_wpos_ytransform_impl(&state
, function
->impl
);
314 return state
.transform
!= NULL
;