2 * Copyright © 2018 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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Rob Clark (robdclark@gmail.com)
29 #include "nir/nir_builtin_builder.h"
31 #include "vtn_private.h"
32 #include "OpenCL.std.h"
34 typedef nir_ssa_def
*(*nir_handler
)(struct vtn_builder
*b
,
35 enum OpenCLstd_Entrypoints opcode
,
36 unsigned num_srcs
, nir_ssa_def
**srcs
,
37 const struct glsl_type
*dest_type
);
40 handle_instr(struct vtn_builder
*b
, enum OpenCLstd_Entrypoints opcode
,
41 const uint32_t *w
, unsigned count
, nir_handler handler
)
43 const struct glsl_type
*dest_type
=
44 vtn_value(b
, w
[1], vtn_value_type_type
)->type
->type
;
46 unsigned num_srcs
= count
- 5;
47 nir_ssa_def
*srcs
[3] = { NULL
};
48 vtn_assert(num_srcs
<= ARRAY_SIZE(srcs
));
49 for (unsigned i
= 0; i
< num_srcs
; i
++) {
50 srcs
[i
] = vtn_ssa_value(b
, w
[i
+ 5])->def
;
53 nir_ssa_def
*result
= handler(b
, opcode
, num_srcs
, srcs
, dest_type
);
55 struct vtn_value
*val
= vtn_push_value(b
, w
[2], vtn_value_type_ssa
);
56 val
->ssa
= vtn_create_ssa_value(b
, dest_type
);
57 val
->ssa
->def
= result
;
59 vtn_assert(dest_type
== glsl_void_type());
64 nir_alu_op_for_opencl_opcode(struct vtn_builder
*b
,
65 enum OpenCLstd_Entrypoints opcode
)
68 case OpenCLstd_Fabs
: return nir_op_fabs
;
69 case OpenCLstd_SAbs
: return nir_op_iabs
;
70 case OpenCLstd_SAdd_sat
: return nir_op_iadd_sat
;
71 case OpenCLstd_UAdd_sat
: return nir_op_uadd_sat
;
72 case OpenCLstd_Ceil
: return nir_op_fceil
;
73 case OpenCLstd_Cos
: return nir_op_fcos
;
74 case OpenCLstd_Exp2
: return nir_op_fexp2
;
75 case OpenCLstd_Log2
: return nir_op_flog2
;
76 case OpenCLstd_Floor
: return nir_op_ffloor
;
77 case OpenCLstd_SHadd
: return nir_op_ihadd
;
78 case OpenCLstd_UHadd
: return nir_op_uhadd
;
79 case OpenCLstd_Fma
: return nir_op_ffma
;
80 case OpenCLstd_Fmax
: return nir_op_fmax
;
81 case OpenCLstd_SMax
: return nir_op_imax
;
82 case OpenCLstd_UMax
: return nir_op_umax
;
83 case OpenCLstd_Fmin
: return nir_op_fmin
;
84 case OpenCLstd_SMin
: return nir_op_imin
;
85 case OpenCLstd_UMin
: return nir_op_umin
;
86 case OpenCLstd_Fmod
: return nir_op_fmod
;
87 case OpenCLstd_Mix
: return nir_op_flrp
;
88 case OpenCLstd_SMul_hi
: return nir_op_imul_high
;
89 case OpenCLstd_UMul_hi
: return nir_op_umul_high
;
90 case OpenCLstd_Popcount
: return nir_op_bit_count
;
91 case OpenCLstd_Pow
: return nir_op_fpow
;
92 case OpenCLstd_Remainder
: return nir_op_frem
;
93 case OpenCLstd_SRhadd
: return nir_op_irhadd
;
94 case OpenCLstd_URhadd
: return nir_op_urhadd
;
95 case OpenCLstd_Rsqrt
: return nir_op_frsq
;
96 case OpenCLstd_Sign
: return nir_op_fsign
;
97 case OpenCLstd_Sin
: return nir_op_fsin
;
98 case OpenCLstd_Sqrt
: return nir_op_fsqrt
;
99 case OpenCLstd_SSub_sat
: return nir_op_isub_sat
;
100 case OpenCLstd_USub_sat
: return nir_op_usub_sat
;
101 case OpenCLstd_Trunc
: return nir_op_ftrunc
;
103 case OpenCLstd_UAbs
: return nir_op_mov
;
105 vtn_fail("No NIR equivalent");
110 handle_alu(struct vtn_builder
*b
, enum OpenCLstd_Entrypoints opcode
,
111 unsigned num_srcs
, nir_ssa_def
**srcs
,
112 const struct glsl_type
*dest_type
)
114 return nir_build_alu(&b
->nb
, nir_alu_op_for_opencl_opcode(b
, opcode
),
115 srcs
[0], srcs
[1], srcs
[2], NULL
);
119 handle_special(struct vtn_builder
*b
, enum OpenCLstd_Entrypoints opcode
,
120 unsigned num_srcs
, nir_ssa_def
**srcs
,
121 const struct glsl_type
*dest_type
)
123 nir_builder
*nb
= &b
->nb
;
126 case OpenCLstd_SAbs_diff
:
127 return nir_iabs_diff(nb
, srcs
[0], srcs
[1]);
128 case OpenCLstd_UAbs_diff
:
129 return nir_uabs_diff(nb
, srcs
[0], srcs
[1]);
130 case OpenCLstd_Bitselect
:
131 return nir_bitselect(nb
, srcs
[0], srcs
[1], srcs
[2]);
132 case OpenCLstd_FClamp
:
133 return nir_fclamp(nb
, srcs
[0], srcs
[1], srcs
[2]);
134 case OpenCLstd_SClamp
:
135 return nir_iclamp(nb
, srcs
[0], srcs
[1], srcs
[2]);
136 case OpenCLstd_UClamp
:
137 return nir_uclamp(nb
, srcs
[0], srcs
[1], srcs
[2]);
138 case OpenCLstd_Copysign
:
139 return nir_copysign(nb
, srcs
[0], srcs
[1]);
140 case OpenCLstd_Cross
:
141 if (glsl_get_components(dest_type
) == 4)
142 return nir_cross4(nb
, srcs
[0], srcs
[1]);
143 return nir_cross3(nb
, srcs
[0], srcs
[1]);
144 case OpenCLstd_Degrees
:
145 return nir_degrees(nb
, srcs
[0]);
147 return nir_fdim(nb
, srcs
[0], srcs
[1]);
148 case OpenCLstd_Distance
:
149 return nir_distance(nb
, srcs
[0], srcs
[1]);
150 case OpenCLstd_Fast_distance
:
151 return nir_fast_distance(nb
, srcs
[0], srcs
[1]);
152 case OpenCLstd_Fast_length
:
153 return nir_fast_length(nb
, srcs
[0]);
154 case OpenCLstd_Fast_normalize
:
155 return nir_fast_normalize(nb
, srcs
[0]);
156 case OpenCLstd_Length
:
157 return nir_length(nb
, srcs
[0]);
159 return nir_fmad(nb
, srcs
[0], srcs
[1], srcs
[2]);
160 case OpenCLstd_Maxmag
:
161 return nir_maxmag(nb
, srcs
[0], srcs
[1]);
162 case OpenCLstd_Minmag
:
163 return nir_minmag(nb
, srcs
[0], srcs
[1]);
165 return nir_nan(nb
, srcs
[0]);
166 case OpenCLstd_Nextafter
:
167 return nir_nextafter(nb
, srcs
[0], srcs
[1]);
168 case OpenCLstd_Normalize
:
169 return nir_normalize(nb
, srcs
[0]);
170 case OpenCLstd_Radians
:
171 return nir_radians(nb
, srcs
[0]);
172 case OpenCLstd_Rotate
:
173 return nir_rotate(nb
, srcs
[0], srcs
[1]);
174 case OpenCLstd_Smoothstep
:
175 return nir_smoothstep(nb
, srcs
[0], srcs
[1], srcs
[2]);
176 case OpenCLstd_Select
:
177 return nir_select(nb
, srcs
[0], srcs
[1], srcs
[2]);
179 return nir_sge(nb
, srcs
[1], srcs
[0]);
180 case OpenCLstd_S_Upsample
:
181 case OpenCLstd_U_Upsample
:
182 return nir_upsample(nb
, srcs
[0], srcs
[1]);
184 vtn_fail("No NIR equivalent");
190 _handle_v_load_store(struct vtn_builder
*b
, enum OpenCLstd_Entrypoints opcode
,
191 const uint32_t *w
, unsigned count
, bool load
)
193 struct vtn_type
*type
;
195 type
= vtn_value(b
, w
[1], vtn_value_type_type
)->type
;
197 type
= vtn_untyped_value(b
, w
[5])->type
;
198 unsigned a
= load
? 0 : 1;
200 const struct glsl_type
*dest_type
= type
->type
;
201 unsigned components
= glsl_get_vector_elements(dest_type
);
202 unsigned stride
= components
* glsl_get_bit_size(dest_type
) / 8;
204 nir_ssa_def
*offset
= vtn_ssa_value(b
, w
[5 + a
])->def
;
205 struct vtn_value
*p
= vtn_value(b
, w
[6 + a
], vtn_value_type_pointer
);
207 nir_deref_instr
*deref
= vtn_pointer_to_deref(b
, p
->pointer
);
209 /* 1. cast to vec type with adjusted stride */
210 deref
= nir_build_deref_cast(&b
->nb
, &deref
->dest
.ssa
, deref
->mode
,
212 /* 2. deref ptr_as_array */
213 deref
= nir_build_deref_ptr_as_array(&b
->nb
, deref
, offset
);
216 struct vtn_ssa_value
*val
= vtn_local_load(b
, deref
, p
->type
->access
);
217 vtn_push_ssa(b
, w
[2], type
, val
);
219 struct vtn_ssa_value
*val
= vtn_ssa_value(b
, w
[5]);
220 vtn_local_store(b
, val
, deref
, p
->type
->access
);
225 vtn_handle_opencl_vload(struct vtn_builder
*b
, enum OpenCLstd_Entrypoints opcode
,
226 const uint32_t *w
, unsigned count
)
228 _handle_v_load_store(b
, opcode
, w
, count
, true);
232 vtn_handle_opencl_vstore(struct vtn_builder
*b
, enum OpenCLstd_Entrypoints opcode
,
233 const uint32_t *w
, unsigned count
)
235 _handle_v_load_store(b
, opcode
, w
, count
, false);
239 handle_printf(struct vtn_builder
*b
, enum OpenCLstd_Entrypoints opcode
,
240 unsigned num_srcs
, nir_ssa_def
**srcs
,
241 const struct glsl_type
*dest_type
)
243 /* hahah, yeah, right.. */
244 return nir_imm_int(&b
->nb
, -1);
248 vtn_handle_opencl_instruction(struct vtn_builder
*b
, SpvOp ext_opcode
,
249 const uint32_t *w
, unsigned count
)
251 switch ((enum OpenCLstd_Entrypoints
)ext_opcode
) {
255 case OpenCLstd_SAdd_sat
:
256 case OpenCLstd_UAdd_sat
:
261 case OpenCLstd_Floor
:
264 case OpenCLstd_SHadd
:
265 case OpenCLstd_UHadd
:
273 case OpenCLstd_SMul_hi
:
274 case OpenCLstd_UMul_hi
:
275 case OpenCLstd_Popcount
:
277 case OpenCLstd_Remainder
:
278 case OpenCLstd_SRhadd
:
279 case OpenCLstd_URhadd
:
280 case OpenCLstd_Rsqrt
:
284 case OpenCLstd_SSub_sat
:
285 case OpenCLstd_USub_sat
:
286 case OpenCLstd_Trunc
:
287 handle_instr(b
, ext_opcode
, w
, count
, handle_alu
);
289 case OpenCLstd_SAbs_diff
:
290 case OpenCLstd_UAbs_diff
:
291 case OpenCLstd_Bitselect
:
292 case OpenCLstd_FClamp
:
293 case OpenCLstd_SClamp
:
294 case OpenCLstd_UClamp
:
295 case OpenCLstd_Copysign
:
296 case OpenCLstd_Cross
:
297 case OpenCLstd_Degrees
:
299 case OpenCLstd_Distance
:
300 case OpenCLstd_Fast_distance
:
301 case OpenCLstd_Fast_length
:
302 case OpenCLstd_Fast_normalize
:
303 case OpenCLstd_Length
:
305 case OpenCLstd_Maxmag
:
306 case OpenCLstd_Minmag
:
308 case OpenCLstd_Nextafter
:
309 case OpenCLstd_Normalize
:
310 case OpenCLstd_Radians
:
311 case OpenCLstd_Rotate
:
312 case OpenCLstd_Select
:
314 case OpenCLstd_Smoothstep
:
315 case OpenCLstd_S_Upsample
:
316 case OpenCLstd_U_Upsample
:
317 handle_instr(b
, ext_opcode
, w
, count
, handle_special
);
319 case OpenCLstd_Vloadn
:
320 vtn_handle_opencl_vload(b
, ext_opcode
, w
, count
);
322 case OpenCLstd_Vstoren
:
323 vtn_handle_opencl_vstore(b
, ext_opcode
, w
, count
);
325 case OpenCLstd_Printf
:
326 handle_instr(b
, ext_opcode
, w
, count
, handle_printf
);
328 case OpenCLstd_Prefetch
:
329 /* TODO maybe add a nir instruction for this? */
332 vtn_fail("unhandled opencl opc: %u\n", ext_opcode
);