3 * Copyright © 2011-2015 Intel Corporation
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28 #include "brw_shader.h"
29 #include "brw_context.h"
35 class src_reg
: public backend_reg
38 DECLARE_RALLOC_CXX_OPERATORS(src_reg
)
42 src_reg(enum brw_reg_file file
, int nr
, const glsl_type
*type
);
44 src_reg(struct ::brw_reg reg
);
46 bool equals(const src_reg
&r
) const;
48 src_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
49 src_reg(class vec4_visitor
*v
, const struct glsl_type
*type
, int size
);
51 explicit src_reg(const dst_reg
®
);
57 retype(src_reg reg
, enum brw_reg_type type
)
66 add_byte_offset(backend_reg
*reg
, unsigned bytes
)
75 assert(reg
->offset
% 16 == 0);
78 const unsigned suboffset
= reg
->offset
+ bytes
;
79 reg
->nr
+= suboffset
/ REG_SIZE
;
80 reg
->offset
= suboffset
% REG_SIZE
;
81 assert(reg
->offset
% 16 == 0);
86 const unsigned suboffset
= reg
->subnr
+ bytes
;
87 reg
->nr
+= suboffset
/ REG_SIZE
;
88 reg
->subnr
= suboffset
% REG_SIZE
;
89 assert(reg
->subnr
% 16 == 0);
97 } /* namepace detail */
100 byte_offset(src_reg reg
, unsigned bytes
)
102 detail::add_byte_offset(®
, bytes
);
106 static inline src_reg
107 offset(src_reg reg
, unsigned width
, unsigned delta
)
109 const unsigned stride
= (reg
.file
== UNIFORM
? 0 : 4);
110 const unsigned num_components
= MAX2(width
/ 4 * stride
, 4);
111 return byte_offset(reg
, num_components
* type_sz(reg
.type
) * delta
);
115 * Reswizzle a given source register.
118 static inline src_reg
119 swizzle(src_reg reg
, unsigned swizzle
)
122 reg
.ud
= brw_swizzle_immediate(reg
.type
, reg
.ud
, swizzle
);
124 reg
.swizzle
= brw_compose_swizzle(swizzle
, reg
.swizzle
);
129 static inline src_reg
132 assert(reg
.file
!= IMM
);
133 reg
.negate
= !reg
.negate
;
138 is_uniform(const src_reg
®
)
140 return (reg
.file
== IMM
|| reg
.file
== UNIFORM
|| reg
.is_null()) &&
141 (!reg
.reladdr
|| is_uniform(*reg
.reladdr
));
144 class dst_reg
: public backend_reg
147 DECLARE_RALLOC_CXX_OPERATORS(dst_reg
)
152 dst_reg(enum brw_reg_file file
, int nr
);
153 dst_reg(enum brw_reg_file file
, int nr
, const glsl_type
*type
,
155 dst_reg(enum brw_reg_file file
, int nr
, brw_reg_type type
,
157 dst_reg(struct ::brw_reg reg
);
158 dst_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
160 explicit dst_reg(const src_reg
®
);
162 bool equals(const dst_reg
&r
) const;
167 static inline dst_reg
168 retype(dst_reg reg
, enum brw_reg_type type
)
174 static inline dst_reg
175 byte_offset(dst_reg reg
, unsigned bytes
)
177 detail::add_byte_offset(®
, bytes
);
181 static inline dst_reg
182 offset(dst_reg reg
, unsigned width
, unsigned delta
)
184 const unsigned stride
= (reg
.file
== UNIFORM
? 0 : 4);
185 const unsigned num_components
= MAX2(width
/ 4 * stride
, 4);
186 return byte_offset(reg
, num_components
* type_sz(reg
.type
) * delta
);
189 static inline dst_reg
190 writemask(dst_reg reg
, unsigned mask
)
192 assert(reg
.file
!= IMM
);
193 assert((reg
.writemask
& mask
) != 0);
194 reg
.writemask
&= mask
;
199 * Return an integer identifying the discrete address space a register is
200 * contained in. A register is by definition fully contained in the single
201 * reg_space it belongs to, so two registers with different reg_space ids are
202 * guaranteed not to overlap. Most register files are a single reg_space of
203 * its own, only the VGRF file is composed of multiple discrete address
204 * spaces, one for each VGRF allocation.
206 static inline uint32_t
207 reg_space(const backend_reg
&r
)
209 return r
.file
<< 16 | (r
.file
== VGRF
? r
.nr
: 0);
213 * Return the base offset in bytes of a register relative to the start of its
216 static inline unsigned
217 reg_offset(const backend_reg
&r
)
219 return (r
.file
== VGRF
|| r
.file
== IMM
? 0 : r
.nr
) *
220 (r
.file
== UNIFORM
? 16 : REG_SIZE
) + r
.offset
+
221 (r
.file
== ARF
|| r
.file
== FIXED_GRF
? r
.subnr
: 0);
225 * Return whether the register region starting at \p r and spanning \p dr
226 * bytes could potentially overlap the register region starting at \p s and
227 * spanning \p ds bytes.
230 regions_overlap(const backend_reg
&r
, unsigned dr
,
231 const backend_reg
&s
, unsigned ds
)
233 if (r
.file
== MRF
&& (r
.nr
& BRW_MRF_COMPR4
)) {
234 /* COMPR4 regions are translated by the hardware during decompression
235 * into two separate half-regions 4 MRFs apart from each other.
238 t0
.nr
&= ~BRW_MRF_COMPR4
;
240 t1
.offset
+= 4 * REG_SIZE
;
241 return regions_overlap(t0
, dr
/ 2, s
, ds
) ||
242 regions_overlap(t1
, dr
/ 2, s
, ds
);
244 } else if (s
.file
== MRF
&& (s
.nr
& BRW_MRF_COMPR4
)) {
245 return regions_overlap(s
, ds
, r
, dr
);
248 return reg_space(r
) == reg_space(s
) &&
249 !(reg_offset(r
) + dr
<= reg_offset(s
) ||
250 reg_offset(s
) + ds
<= reg_offset(r
));
254 class vec4_instruction
: public backend_instruction
{
256 DECLARE_RALLOC_CXX_OPERATORS(vec4_instruction
)
258 vec4_instruction(enum opcode opcode
,
259 const dst_reg
&dst
= dst_reg(),
260 const src_reg
&src0
= src_reg(),
261 const src_reg
&src1
= src_reg(),
262 const src_reg
&src2
= src_reg());
267 enum brw_urb_write_flags urb_write_flags
;
269 unsigned sol_binding
; /**< gen6: SOL binding table index */
270 bool sol_final_write
; /**< gen6: send commit message */
271 unsigned sol_vertex
; /**< gen6: used for setting dst index in SVB header */
273 bool is_send_from_grf();
274 unsigned size_read(unsigned arg
) const;
275 bool can_reswizzle(const struct gen_device_info
*devinfo
, int dst_writemask
,
276 int swizzle
, int swizzle_mask
);
277 void reswizzle(int dst_writemask
, int swizzle
);
278 bool can_do_source_mods(const struct gen_device_info
*devinfo
);
279 bool can_do_writemask(const struct gen_device_info
*devinfo
);
280 bool can_change_types() const;
281 bool has_source_and_destination_hazard() const;
285 return predicate
|| opcode
== VS_OPCODE_UNPACK_FLAGS_SIMD4X2
;
288 bool reads_flag(unsigned c
)
290 if (opcode
== VS_OPCODE_UNPACK_FLAGS_SIMD4X2
)
294 case BRW_PREDICATE_NONE
:
296 case BRW_PREDICATE_ALIGN16_REPLICATE_X
:
298 case BRW_PREDICATE_ALIGN16_REPLICATE_Y
:
300 case BRW_PREDICATE_ALIGN16_REPLICATE_Z
:
302 case BRW_PREDICATE_ALIGN16_REPLICATE_W
:
311 return (conditional_mod
&& (opcode
!= BRW_OPCODE_SEL
&&
312 opcode
!= BRW_OPCODE_IF
&&
313 opcode
!= BRW_OPCODE_WHILE
));
318 * Make the execution of \p inst dependent on the evaluation of a possibly
319 * inverted predicate.
321 inline vec4_instruction
*
322 set_predicate_inv(enum brw_predicate pred
, bool inverse
,
323 vec4_instruction
*inst
)
325 inst
->predicate
= pred
;
326 inst
->predicate_inverse
= inverse
;
331 * Make the execution of \p inst dependent on the evaluation of a predicate.
333 inline vec4_instruction
*
334 set_predicate(enum brw_predicate pred
, vec4_instruction
*inst
)
336 return set_predicate_inv(pred
, false, inst
);
340 * Write the result of evaluating the condition given by \p mod to a flag
343 inline vec4_instruction
*
344 set_condmod(enum brw_conditional_mod mod
, vec4_instruction
*inst
)
346 inst
->conditional_mod
= mod
;
351 * Clamp the result of \p inst to the saturation range of its destination
354 inline vec4_instruction
*
355 set_saturate(bool saturate
, vec4_instruction
*inst
)
357 inst
->saturate
= saturate
;
362 * Return the number of dataflow registers written by the instruction (either
363 * fully or partially) counted from 'floor(reg_offset(inst->dst) /
364 * register_size)'. The somewhat arbitrary register size unit is 16B for the
365 * UNIFORM and IMM files and 32B for all other files.
368 regs_written(const vec4_instruction
*inst
)
370 assert(inst
->dst
.file
!= UNIFORM
&& inst
->dst
.file
!= IMM
);
371 return DIV_ROUND_UP(reg_offset(inst
->dst
) % REG_SIZE
+ inst
->size_written
,
376 * Return the number of dataflow registers read by the instruction (either
377 * fully or partially) counted from 'floor(reg_offset(inst->src[i]) /
378 * register_size)'. The somewhat arbitrary register size unit is 16B for the
379 * UNIFORM and IMM files and 32B for all other files.
382 regs_read(const vec4_instruction
*inst
, unsigned i
)
384 const unsigned reg_size
=
385 inst
->src
[i
].file
== UNIFORM
|| inst
->src
[i
].file
== IMM
? 16 : REG_SIZE
;
386 return DIV_ROUND_UP(reg_offset(inst
->src
[i
]) % reg_size
+ inst
->size_read(i
),
390 } /* namespace brw */