2 * Copyright © 2017 Intel Corporation
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 #include "nir_builder.h"
26 #include "util/format_rgb9e5.h"
28 static inline nir_ssa_def
*
29 nir_shift(nir_builder
*b
, nir_ssa_def
*value
, int left_shift
)
32 return nir_ishl(b
, value
, nir_imm_int(b
, left_shift
));
33 else if (left_shift
< 0)
34 return nir_ushr(b
, value
, nir_imm_int(b
, -left_shift
));
39 static inline nir_ssa_def
*
40 nir_mask_shift(struct nir_builder
*b
, nir_ssa_def
*src
,
41 uint32_t mask
, int left_shift
)
43 return nir_shift(b
, nir_iand(b
, src
, nir_imm_int(b
, mask
)), left_shift
);
46 static inline nir_ssa_def
*
47 nir_mask_shift_or(struct nir_builder
*b
, nir_ssa_def
*dst
, nir_ssa_def
*src
,
48 uint32_t src_mask
, int src_left_shift
)
50 return nir_ior(b
, nir_mask_shift(b
, src
, src_mask
, src_left_shift
), dst
);
53 static inline nir_ssa_def
*
54 nir_format_unpack_int(nir_builder
*b
, nir_ssa_def
*packed
,
55 const unsigned *bits
, unsigned num_components
,
58 assert(num_components
>= 1 && num_components
<= 4);
59 const unsigned bit_size
= packed
->bit_size
;
60 nir_ssa_def
*comps
[4];
62 if (bits
[0] >= bit_size
) {
63 assert(bits
[0] == bit_size
);
64 assert(num_components
== 1);
69 for (unsigned i
= 0; i
< num_components
; i
++) {
70 assert(bits
[i
] < bit_size
);
71 assert(offset
+ bits
[i
] <= bit_size
);
72 nir_ssa_def
*lshift
= nir_imm_int(b
, bit_size
- (offset
+ bits
[i
]));
73 nir_ssa_def
*rshift
= nir_imm_int(b
, bit_size
- bits
[i
]);
75 comps
[i
] = nir_ishr(b
, nir_ishl(b
, packed
, lshift
), rshift
);
77 comps
[i
] = nir_ushr(b
, nir_ishl(b
, packed
, lshift
), rshift
);
80 assert(offset
<= bit_size
);
82 return nir_vec(b
, comps
, num_components
);
85 static inline nir_ssa_def
*
86 nir_format_unpack_uint(nir_builder
*b
, nir_ssa_def
*packed
,
87 const unsigned *bits
, unsigned num_components
)
89 return nir_format_unpack_int(b
, packed
, bits
, num_components
, false);
92 static inline nir_ssa_def
*
93 nir_format_unpack_sint(nir_builder
*b
, nir_ssa_def
*packed
,
94 const unsigned *bits
, unsigned num_components
)
96 return nir_format_unpack_int(b
, packed
, bits
, num_components
, true);
99 static inline nir_ssa_def
*
100 nir_format_pack_uint_unmasked(nir_builder
*b
, nir_ssa_def
*color
,
101 const unsigned *bits
, unsigned num_components
)
103 assert(num_components
>= 1 && num_components
<= 4);
104 nir_ssa_def
*packed
= nir_imm_int(b
, 0);
106 for (unsigned i
= 0; i
< num_components
; i
++) {
107 packed
= nir_ior(b
, packed
, nir_shift(b
, nir_channel(b
, color
, i
),
111 assert(offset
<= packed
->bit_size
);
116 static inline nir_ssa_def
*
117 nir_format_pack_uint(nir_builder
*b
, nir_ssa_def
*color
,
118 const unsigned *bits
, unsigned num_components
)
120 nir_const_value mask
;
121 for (unsigned i
= 0; i
< num_components
; i
++) {
122 assert(bits
[i
] < 32);
123 mask
.u32
[i
] = (1u << bits
[i
]) - 1;
125 nir_ssa_def
*mask_imm
= nir_build_imm(b
, num_components
, 32, mask
);
127 return nir_format_pack_uint_unmasked(b
, nir_iand(b
, color
, mask_imm
),
128 bits
, num_components
);
131 static inline nir_ssa_def
*
132 nir_format_bitcast_uint_vec_unmasked(nir_builder
*b
, nir_ssa_def
*src
,
133 unsigned src_bits
, unsigned dst_bits
)
135 assert(src_bits
== 8 || src_bits
== 16 || src_bits
== 32);
136 assert(dst_bits
== 8 || dst_bits
== 16 || dst_bits
== 32);
138 if (src_bits
== dst_bits
)
141 const unsigned dst_components
=
142 DIV_ROUND_UP(src
->num_components
* src_bits
, dst_bits
);
143 assert(dst_components
<= 4);
145 nir_ssa_def
*dst_chan
[4] = {0};
146 if (dst_bits
> src_bits
) {
148 unsigned dst_idx
= 0;
149 for (unsigned i
= 0; i
< src
->num_components
; i
++) {
150 nir_ssa_def
*shifted
= nir_ishl(b
, nir_channel(b
, src
, i
),
151 nir_imm_int(b
, shift
));
153 dst_chan
[dst_idx
] = shifted
;
155 dst_chan
[dst_idx
] = nir_ior(b
, dst_chan
[dst_idx
], shifted
);
159 if (shift
>= dst_bits
) {
165 nir_ssa_def
*mask
= nir_imm_int(b
, ~0u >> (32 - dst_bits
));
167 unsigned src_idx
= 0;
169 for (unsigned i
= 0; i
< dst_components
; i
++) {
170 dst_chan
[i
] = nir_iand(b
, nir_ushr(b
, nir_channel(b
, src
, src_idx
),
171 nir_imm_int(b
, shift
)),
174 if (shift
>= src_bits
) {
181 return nir_vec(b
, dst_chan
, dst_components
);
184 static inline nir_ssa_def
*
185 nir_format_linear_to_srgb(nir_builder
*b
, nir_ssa_def
*c
)
187 nir_ssa_def
*linear
= nir_fmul(b
, c
, nir_imm_float(b
, 12.92f
));
188 nir_ssa_def
*curved
=
189 nir_fsub(b
, nir_fmul(b
, nir_imm_float(b
, 1.055f
),
190 nir_fpow(b
, c
, nir_imm_float(b
, 1.0 / 2.4))),
191 nir_imm_float(b
, 0.055f
));
193 return nir_fsat(b
, nir_bcsel(b
, nir_flt(b
, c
, nir_imm_float(b
, 0.0031308f
)),
197 static inline nir_ssa_def
*
198 nir_format_srgb_to_linear(nir_builder
*b
, nir_ssa_def
*c
)
200 nir_ssa_def
*linear
= nir_fdiv(b
, c
, nir_imm_float(b
, 12.92f
));
201 nir_ssa_def
*curved
=
202 nir_fpow(b
, nir_fdiv(b
, nir_fadd(b
, c
, nir_imm_float(b
, 0.055f
)),
203 nir_imm_float(b
, 1.055f
)),
204 nir_imm_float(b
, 2.4f
));
206 return nir_fsat(b
, nir_bcsel(b
, nir_fge(b
, nir_imm_float(b
, 0.04045f
), c
),
210 static inline nir_ssa_def
*
211 nir_format_unpack_11f11f10f(nir_builder
*b
, nir_ssa_def
*packed
)
213 nir_ssa_def
*chans
[3];
214 chans
[0] = nir_mask_shift(b
, packed
, 0x000007ff, 4);
215 chans
[1] = nir_mask_shift(b
, packed
, 0x003ff100, -7);
216 chans
[2] = nir_mask_shift(b
, packed
, 0xffc00000, -17);
218 for (unsigned i
= 0; i
< 3; i
++)
219 chans
[i
] = nir_unpack_half_2x16_split_x(b
, chans
[i
]);
221 return nir_vec(b
, chans
, 3);
224 static inline nir_ssa_def
*
225 nir_format_pack_r11g11b10f(nir_builder
*b
, nir_ssa_def
*color
)
227 /* 10 and 11-bit floats are unsigned. Clamp to non-negative */
228 nir_ssa_def
*clamped
= nir_fmax(b
, color
, nir_imm_float(b
, 0));
230 nir_ssa_def
*undef
= nir_ssa_undef(b
, 1, color
->bit_size
);
231 nir_ssa_def
*p1
= nir_pack_half_2x16_split(b
, nir_channel(b
, clamped
, 0),
232 nir_channel(b
, clamped
, 1));
233 nir_ssa_def
*p2
= nir_pack_half_2x16_split(b
, nir_channel(b
, clamped
, 2),
236 /* A 10 or 11-bit float has the same exponent as a 16-bit float but with
237 * fewer mantissa bits and no sign bit. All we have to do is throw away
238 * the sign bit and the bottom mantissa bits and shift it into place.
240 nir_ssa_def
*packed
= nir_imm_int(b
, 0);
241 packed
= nir_mask_shift_or(b
, packed
, p1
, 0x00007ff0, -4);
242 packed
= nir_mask_shift_or(b
, packed
, p1
, 0x7ff00000, -9);
243 packed
= nir_mask_shift_or(b
, packed
, p2
, 0x00007fe0, 17);
248 static inline nir_ssa_def
*
249 nir_format_pack_r9g9b9e5(nir_builder
*b
, nir_ssa_def
*color
)
251 /* See also float3_to_rgb9e5 */
253 /* First, we need to clamp it to range. */
254 nir_ssa_def
*clamped
= nir_fmin(b
, color
, nir_imm_float(b
, MAX_RGB9E5
));
256 /* Get rid of negatives and NaN */
257 clamped
= nir_bcsel(b
, nir_ult(b
, nir_imm_int(b
, 0x7f800000), color
),
258 nir_imm_float(b
, 0), clamped
);
260 /* maxrgb.u = MAX3(rc.u, gc.u, bc.u); */
261 nir_ssa_def
*maxu
= nir_umax(b
, nir_channel(b
, clamped
, 0),
262 nir_umax(b
, nir_channel(b
, clamped
, 1),
263 nir_channel(b
, clamped
, 2)));
265 /* maxrgb.u += maxrgb.u & (1 << (23-9)); */
266 maxu
= nir_iadd(b
, maxu
, nir_iand(b
, maxu
, nir_imm_int(b
, 1 << 14)));
268 /* exp_shared = MAX2((maxrgb.u >> 23), -RGB9E5_EXP_BIAS - 1 + 127) +
269 * 1 + RGB9E5_EXP_BIAS - 127;
271 nir_ssa_def
*exp_shared
=
272 nir_iadd(b
, nir_umax(b
, nir_ushr(b
, maxu
, nir_imm_int(b
, 23)),
273 nir_imm_int(b
, -RGB9E5_EXP_BIAS
- 1 + 127)),
274 nir_imm_int(b
, 1 + RGB9E5_EXP_BIAS
- 127));
276 /* revdenom_biasedexp = 127 - (exp_shared - RGB9E5_EXP_BIAS -
277 * RGB9E5_MANTISSA_BITS) + 1;
279 nir_ssa_def
*revdenom_biasedexp
=
280 nir_isub(b
, nir_imm_int(b
, 127 + RGB9E5_EXP_BIAS
+
281 RGB9E5_MANTISSA_BITS
+ 1),
284 /* revdenom.u = revdenom_biasedexp << 23; */
285 nir_ssa_def
*revdenom
=
286 nir_ishl(b
, revdenom_biasedexp
, nir_imm_int(b
, 23));
288 /* rm = (int) (rc.f * revdenom.f);
289 * gm = (int) (gc.f * revdenom.f);
290 * bm = (int) (bc.f * revdenom.f);
292 nir_ssa_def
*mantissa
=
293 nir_f2i32(b
, nir_fmul(b
, clamped
, revdenom
));
295 /* rm = (rm & 1) + (rm >> 1);
296 * gm = (gm & 1) + (gm >> 1);
297 * bm = (bm & 1) + (bm >> 1);
299 mantissa
= nir_iadd(b
, nir_iand(b
, mantissa
, nir_imm_int(b
, 1)),
300 nir_ushr(b
, mantissa
, nir_imm_int(b
, 1)));
302 nir_ssa_def
*packed
= nir_channel(b
, mantissa
, 0);
303 packed
= nir_mask_shift_or(b
, packed
, nir_channel(b
, mantissa
, 1), ~0, 9);
304 packed
= nir_mask_shift_or(b
, packed
, nir_channel(b
, mantissa
, 2), ~0, 18);
305 packed
= nir_mask_shift_or(b
, packed
, exp_shared
, ~0, 27);