2 * Copyright © 2014 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 * Connor Abbott (cwabbott0@gmail.com)
29 * This header file defines all the available intrinsics in one place. It
30 * expands to a list of macros of the form:
32 * INTRINSIC(name, num_srcs, src_components, has_dest, dest_components,
33 * num_variables, num_indices, idx0, idx1, idx2, flags)
35 * Which should correspond one-to-one with the nir_intrinsic_info structure. It
36 * is included in both ir.h to create the nir_intrinsic enum (with members of
37 * the form nir_intrinsic_(name)) and and in opcodes.c to create
38 * nir_intrinsic_infos, which is a const array of nir_intrinsic_info structures
42 #define ARR(...) { __VA_ARGS__ }
44 INTRINSIC(nop
, 0, ARR(0), false, 0, 0, 0, xx
, xx
, xx
,
45 NIR_INTRINSIC_CAN_ELIMINATE
)
47 INTRINSIC(load_var
, 0, ARR(0), true, 0, 1, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
48 INTRINSIC(store_var
, 1, ARR(0), false, 0, 1, 1, WRMASK
, xx
, xx
, 0)
49 INTRINSIC(copy_var
, 0, ARR(0), false, 0, 2, 0, xx
, xx
, xx
, 0)
52 * Interpolation of input. The interp_var_at* intrinsics are similar to the
53 * load_var intrinsic acting on a shader input except that they interpolate
54 * the input differently. The at_sample and at_offset intrinsics take an
55 * additional source that is an integer sample id or a vec2 position offset
59 INTRINSIC(interp_var_at_centroid
, 0, ARR(0), true, 0, 1, 0, xx
, xx
, xx
,
60 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
61 INTRINSIC(interp_var_at_sample
, 1, ARR(1), true, 0, 1, 0, xx
, xx
, xx
,
62 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
63 INTRINSIC(interp_var_at_offset
, 1, ARR(2), true, 0, 1, 0, xx
, xx
, xx
,
64 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
67 * Ask the driver for the size of a given buffer. It takes the buffer index
70 INTRINSIC(get_buffer_size
, 1, ARR(1), true, 1, 0, 0, xx
, xx
, xx
,
71 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
74 * a barrier is an intrinsic with no inputs/outputs but which can't be moved
75 * around/optimized in general
77 #define BARRIER(name) INTRINSIC(name, 0, ARR(0), false, 0, 0, 0, xx, xx, xx, 0)
83 * Memory barrier with semantics analogous to the memoryBarrier() GLSL
86 BARRIER(memory_barrier
)
89 * Shader clock intrinsic with semantics analogous to the clock2x32ARB()
91 * The latter can be used as code motion barrier, which is currently not
94 INTRINSIC(shader_clock
, 0, ARR(0), true, 2, 0, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
97 * Shader ballot intrinsics with semantics analogous to the
100 * readInvocationARB()
101 * readFirstInvocationARB()
103 * GLSL functions from ARB_shader_ballot.
105 INTRINSIC(ballot
, 1, ARR(1), true, 1, 0, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
106 INTRINSIC(read_invocation
, 2, ARR(0, 1), true, 0, 0, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
107 INTRINSIC(read_first_invocation
, 1, ARR(0), true, 0, 0, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
110 * Memory barrier with semantics analogous to the compute shader
111 * groupMemoryBarrier(), memoryBarrierAtomicCounter(), memoryBarrierBuffer(),
112 * memoryBarrierImage() and memoryBarrierShared() GLSL intrinsics.
114 BARRIER(group_memory_barrier
)
115 BARRIER(memory_barrier_atomic_counter
)
116 BARRIER(memory_barrier_buffer
)
117 BARRIER(memory_barrier_image
)
118 BARRIER(memory_barrier_shared
)
120 /** A conditional discard, with a single boolean source. */
121 INTRINSIC(discard_if
, 1, ARR(1), false, 0, 0, 0, xx
, xx
, xx
, 0)
123 /** ARB_shader_group_vote intrinsics */
124 INTRINSIC(vote_any
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
125 INTRINSIC(vote_all
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
126 INTRINSIC(vote_eq
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
129 * Basic Geometry Shader intrinsics.
131 * emit_vertex implements GLSL's EmitStreamVertex() built-in. It takes a single
132 * index, which is the stream ID to write to.
134 * end_primitive implements GLSL's EndPrimitive() built-in.
136 INTRINSIC(emit_vertex
, 0, ARR(0), false, 0, 0, 1, STREAM_ID
, xx
, xx
, 0)
137 INTRINSIC(end_primitive
, 0, ARR(0), false, 0, 0, 1, STREAM_ID
, xx
, xx
, 0)
140 * Geometry Shader intrinsics with a vertex count.
142 * Alternatively, drivers may implement these intrinsics, and use
143 * nir_lower_gs_intrinsics() to convert from the basic intrinsics.
145 * These maintain a count of the number of vertices emitted, as an additional
146 * unsigned integer source.
148 INTRINSIC(emit_vertex_with_counter
, 1, ARR(1), false, 0, 0, 1, STREAM_ID
, xx
, xx
, 0)
149 INTRINSIC(end_primitive_with_counter
, 1, ARR(1), false, 0, 0, 1, STREAM_ID
, xx
, xx
, 0)
150 INTRINSIC(set_vertex_count
, 1, ARR(1), false, 0, 0, 0, xx
, xx
, xx
, 0)
155 * The *_var variants take an atomic_uint nir_variable, while the other,
156 * lowered, variants take a constant buffer index and register offset.
159 #define ATOMIC(name, flags) \
160 INTRINSIC(name##_var, 0, ARR(0), true, 1, 1, 0, xx, xx, xx, flags) \
161 INTRINSIC(name, 1, ARR(1), true, 1, 0, 1, BASE, xx, xx, flags)
162 #define ATOMIC2(name) \
163 INTRINSIC(name##_var, 1, ARR(1), true, 1, 1, 0, xx, xx, xx, 0) \
164 INTRINSIC(name, 2, ARR(1, 1), true, 1, 0, 1, BASE, xx, xx, 0)
165 #define ATOMIC3(name) \
166 INTRINSIC(name##_var, 2, ARR(1, 1), true, 1, 1, 0, xx, xx, xx, 0) \
167 INTRINSIC(name, 3, ARR(1, 1, 1), true, 1, 0, 1, BASE, xx, xx, 0)
169 ATOMIC(atomic_counter_inc
, 0)
170 ATOMIC(atomic_counter_dec
, 0)
171 ATOMIC(atomic_counter_read
, NIR_INTRINSIC_CAN_ELIMINATE
)
172 ATOMIC2(atomic_counter_add
)
173 ATOMIC2(atomic_counter_min
)
174 ATOMIC2(atomic_counter_max
)
175 ATOMIC2(atomic_counter_and
)
176 ATOMIC2(atomic_counter_or
)
177 ATOMIC2(atomic_counter_xor
)
178 ATOMIC2(atomic_counter_exchange
)
179 ATOMIC3(atomic_counter_comp_swap
)
182 * Image load, store and atomic intrinsics.
184 * All image intrinsics take an image target passed as a nir_variable. Image
185 * variables contain a number of memory and layout qualifiers that influence
186 * the semantics of the intrinsic.
188 * All image intrinsics take a four-coordinate vector and a sample index as
189 * first two sources, determining the location within the image that will be
190 * accessed by the intrinsic. Components not applicable to the image target
191 * in use are undefined. Image store takes an additional four-component
192 * argument with the value to be written, and image atomic operations take
193 * either one or two additional scalar arguments with the same meaning as in
194 * the ARB_shader_image_load_store specification.
196 INTRINSIC(image_load
, 2, ARR(4, 1), true, 4, 1, 0, xx
, xx
, xx
,
197 NIR_INTRINSIC_CAN_ELIMINATE
)
198 INTRINSIC(image_store
, 3, ARR(4, 1, 4), false, 0, 1, 0, xx
, xx
, xx
, 0)
199 INTRINSIC(image_atomic_add
, 3, ARR(4, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
200 INTRINSIC(image_atomic_min
, 3, ARR(4, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
201 INTRINSIC(image_atomic_max
, 3, ARR(4, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
202 INTRINSIC(image_atomic_and
, 3, ARR(4, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
203 INTRINSIC(image_atomic_or
, 3, ARR(4, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
204 INTRINSIC(image_atomic_xor
, 3, ARR(4, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
205 INTRINSIC(image_atomic_exchange
, 3, ARR(4, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
206 INTRINSIC(image_atomic_comp_swap
, 4, ARR(4, 1, 1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
207 INTRINSIC(image_size
, 0, ARR(0), true, 0, 1, 0, xx
, xx
, xx
,
208 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
209 INTRINSIC(image_samples
, 0, ARR(0), true, 1, 1, 0, xx
, xx
, xx
,
210 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
213 * Vulkan descriptor set intrinsic
215 * The Vulkan API uses a different binding model from GL. In the Vulkan
216 * API, all external resources are represented by a tuple:
218 * (descriptor set, binding, array index)
220 * where the array index is the only thing allowed to be indirect. The
221 * vulkan_surface_index intrinsic takes the descriptor set and binding as
222 * its first two indices and the array index as its source. The third
223 * index is a nir_variable_mode in case that's useful to the backend.
225 * The intended usage is that the shader will call vulkan_surface_index to
226 * get an index and then pass that as the buffer index ubo/ssbo calls.
228 INTRINSIC(vulkan_resource_index
, 1, ARR(1), true, 1, 0, 2,
229 DESC_SET
, BINDING
, xx
,
230 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
233 * variable atomic intrinsics
235 * All of these variable atomic memory operations read a value from memory,
236 * compute a new value using one of the operations below, write the new value
237 * to memory, and return the original value read.
239 * All operations take 1 source except CompSwap that takes 2. These sources
242 * 0: The data parameter to the atomic function (i.e. the value to add
243 * in shared_atomic_add, etc).
244 * 1: For CompSwap only: the second data parameter.
246 * All operations take 1 variable deref.
248 INTRINSIC(var_atomic_add
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
249 INTRINSIC(var_atomic_imin
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
250 INTRINSIC(var_atomic_umin
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
251 INTRINSIC(var_atomic_imax
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
252 INTRINSIC(var_atomic_umax
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
253 INTRINSIC(var_atomic_and
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
254 INTRINSIC(var_atomic_or
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
255 INTRINSIC(var_atomic_xor
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
256 INTRINSIC(var_atomic_exchange
, 1, ARR(1), true, 1, 1, 0, xx
, xx
, xx
, 0)
257 INTRINSIC(var_atomic_comp_swap
, 2, ARR(1, 1), true, 1, 1, 0, xx
, xx
, xx
, 0)
260 * SSBO atomic intrinsics
262 * All of the SSBO atomic memory operations read a value from memory,
263 * compute a new value using one of the operations below, write the new
264 * value to memory, and return the original value read.
266 * All operations take 3 sources except CompSwap that takes 4. These
269 * 0: The SSBO buffer index.
270 * 1: The offset into the SSBO buffer of the variable that the atomic
271 * operation will operate on.
272 * 2: The data parameter to the atomic function (i.e. the value to add
273 * in ssbo_atomic_add, etc).
274 * 3: For CompSwap only: the second data parameter.
276 INTRINSIC(ssbo_atomic_add
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
277 INTRINSIC(ssbo_atomic_imin
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
278 INTRINSIC(ssbo_atomic_umin
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
279 INTRINSIC(ssbo_atomic_imax
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
280 INTRINSIC(ssbo_atomic_umax
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
281 INTRINSIC(ssbo_atomic_and
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
282 INTRINSIC(ssbo_atomic_or
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
283 INTRINSIC(ssbo_atomic_xor
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
284 INTRINSIC(ssbo_atomic_exchange
, 3, ARR(1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
285 INTRINSIC(ssbo_atomic_comp_swap
, 4, ARR(1, 1, 1, 1), true, 1, 0, 0, xx
, xx
, xx
, 0)
288 * CS shared variable atomic intrinsics
290 * All of the shared variable atomic memory operations read a value from
291 * memory, compute a new value using one of the operations below, write the
292 * new value to memory, and return the original value read.
294 * All operations take 2 sources except CompSwap that takes 3. These
297 * 0: The offset into the shared variable storage region that the atomic
298 * operation will operate on.
299 * 1: The data parameter to the atomic function (i.e. the value to add
300 * in shared_atomic_add, etc).
301 * 2: For CompSwap only: the second data parameter.
303 INTRINSIC(shared_atomic_add
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
304 INTRINSIC(shared_atomic_imin
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
305 INTRINSIC(shared_atomic_umin
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
306 INTRINSIC(shared_atomic_imax
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
307 INTRINSIC(shared_atomic_umax
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
308 INTRINSIC(shared_atomic_and
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
309 INTRINSIC(shared_atomic_or
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
310 INTRINSIC(shared_atomic_xor
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
311 INTRINSIC(shared_atomic_exchange
, 2, ARR(1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
312 INTRINSIC(shared_atomic_comp_swap
, 3, ARR(1, 1, 1), true, 1, 0, 1, BASE
, xx
, xx
, 0)
314 /* Used by nir_builder.h to generate loader helpers for the system values. */
315 #ifndef DEFINE_SYSTEM_VALUE
316 #define DEFINE_SYSTEM_VALUE(name)
319 #define SYSTEM_VALUE(name, components, num_indices, idx0, idx1, idx2) \
320 DEFINE_SYSTEM_VALUE(name) \
321 INTRINSIC(load_##name, 0, ARR(0), true, components, 0, num_indices, \
323 NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
325 SYSTEM_VALUE(front_face
, 1, 0, xx
, xx
, xx
)
326 SYSTEM_VALUE(vertex_id
, 1, 0, xx
, xx
, xx
)
327 SYSTEM_VALUE(vertex_id_zero_base
, 1, 0, xx
, xx
, xx
)
328 SYSTEM_VALUE(base_vertex
, 1, 0, xx
, xx
, xx
)
329 SYSTEM_VALUE(instance_id
, 1, 0, xx
, xx
, xx
)
330 SYSTEM_VALUE(base_instance
, 1, 0, xx
, xx
, xx
)
331 SYSTEM_VALUE(draw_id
, 1, 0, xx
, xx
, xx
)
332 SYSTEM_VALUE(sample_id
, 1, 0, xx
, xx
, xx
)
333 SYSTEM_VALUE(sample_pos
, 2, 0, xx
, xx
, xx
)
334 SYSTEM_VALUE(sample_mask_in
, 1, 0, xx
, xx
, xx
)
335 SYSTEM_VALUE(primitive_id
, 1, 0, xx
, xx
, xx
)
336 SYSTEM_VALUE(invocation_id
, 1, 0, xx
, xx
, xx
)
337 SYSTEM_VALUE(tess_coord
, 3, 0, xx
, xx
, xx
)
338 SYSTEM_VALUE(tess_level_outer
, 4, 0, xx
, xx
, xx
)
339 SYSTEM_VALUE(tess_level_inner
, 2, 0, xx
, xx
, xx
)
340 SYSTEM_VALUE(patch_vertices_in
, 1, 0, xx
, xx
, xx
)
341 SYSTEM_VALUE(local_invocation_id
, 3, 0, xx
, xx
, xx
)
342 SYSTEM_VALUE(local_invocation_index
, 1, 0, xx
, xx
, xx
)
343 SYSTEM_VALUE(work_group_id
, 3, 0, xx
, xx
, xx
)
344 SYSTEM_VALUE(user_clip_plane
, 4, 1, UCP_ID
, xx
, xx
)
345 SYSTEM_VALUE(num_work_groups
, 3, 0, xx
, xx
, xx
)
346 SYSTEM_VALUE(helper_invocation
, 1, 0, xx
, xx
, xx
)
347 SYSTEM_VALUE(channel_num
, 1, 0, xx
, xx
, xx
)
348 SYSTEM_VALUE(alpha_ref_float
, 1, 0, xx
, xx
, xx
)
349 SYSTEM_VALUE(layer_id
, 1, 0, xx
, xx
, xx
)
350 SYSTEM_VALUE(view_index
, 1, 0, xx
, xx
, xx
)
352 /* Blend constant color values. Float values are clamped. */
353 SYSTEM_VALUE(blend_const_color_r_float
, 1, 0, xx
, xx
, xx
)
354 SYSTEM_VALUE(blend_const_color_g_float
, 1, 0, xx
, xx
, xx
)
355 SYSTEM_VALUE(blend_const_color_b_float
, 1, 0, xx
, xx
, xx
)
356 SYSTEM_VALUE(blend_const_color_a_float
, 1, 0, xx
, xx
, xx
)
357 SYSTEM_VALUE(blend_const_color_rgba8888_unorm
, 1, 0, xx
, xx
, xx
)
358 SYSTEM_VALUE(blend_const_color_aaaa8888_unorm
, 1, 0, xx
, xx
, xx
)
361 * Barycentric coordinate intrinsics.
363 * These set up the barycentric coordinates for a particular interpolation.
364 * The first three are for the simple cases: pixel, centroid, or per-sample
365 * (at gl_SampleID). The next two handle interpolating at a specified
366 * sample location, or interpolating with a vec2 offset,
368 * The interp_mode index should be either the INTERP_MODE_SMOOTH or
369 * INTERP_MODE_NOPERSPECTIVE enum values.
371 * The vec2 value produced by these intrinsics is intended for use as the
372 * barycoord source of a load_interpolated_input intrinsic.
375 #define BARYCENTRIC(name, sources, source_components) \
376 INTRINSIC(load_barycentric_##name, sources, ARR(source_components), \
377 true, 2, 0, 1, INTERP_MODE, xx, xx, \
378 NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
380 /* no sources. const_index[] = { interp_mode } */
381 BARYCENTRIC(pixel
, 0, 0)
382 BARYCENTRIC(centroid
, 0, 0)
383 BARYCENTRIC(sample
, 0, 0)
384 /* src[] = { sample_id }. const_index[] = { interp_mode } */
385 BARYCENTRIC(at_sample
, 1, 1)
386 /* src[] = { offset.xy }. const_index[] = { interp_mode } */
387 BARYCENTRIC(at_offset
, 1, 2)
390 * Load operations pull data from some piece of GPU memory. All load
391 * operations operate in terms of offsets into some piece of theoretical
392 * memory. Loads from externally visible memory (UBO and SSBO) simply take a
393 * byte offset as a source. Loads from opaque memory (uniforms, inputs, etc.)
394 * take a base+offset pair where the base (const_index[0]) gives the location
395 * of the start of the variable being loaded and and the offset source is a
396 * offset into that variable.
398 * Uniform load operations have a second "range" index that specifies the
399 * range (starting at base) of the data from which we are loading. If
400 * const_index[1] == 0, then the range is unknown.
402 * Some load operations such as UBO/SSBO load and per_vertex loads take an
403 * additional source to specify which UBO/SSBO/vertex to load from.
405 * The exact address type depends on the lowering pass that generates the
406 * load/store intrinsics. Typically, this is vec4 units for things such as
407 * varying slots and float units for fragment shader inputs. UBO and SSBO
408 * offsets are always in bytes.
411 #define LOAD(name, srcs, num_indices, idx0, idx1, idx2, flags) \
412 INTRINSIC(load_##name, srcs, ARR(1, 1, 1, 1), true, 0, 0, num_indices, idx0, idx1, idx2, flags)
414 /* src[] = { offset }. const_index[] = { base, range } */
415 LOAD(uniform
, 1, 2, BASE
, RANGE
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
416 /* src[] = { buffer_index, offset }. No const_index */
417 LOAD(ubo
, 2, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
418 /* src[] = { offset }. const_index[] = { base, component } */
419 LOAD(input
, 1, 2, BASE
, COMPONENT
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
420 /* src[] = { vertex, offset }. const_index[] = { base, component } */
421 LOAD(per_vertex_input
, 2, 2, BASE
, COMPONENT
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
422 /* src[] = { barycoord, offset }. const_index[] = { base, component } */
423 INTRINSIC(load_interpolated_input
, 2, ARR(2, 1), true, 0, 0,
424 2, BASE
, COMPONENT
, xx
,
425 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
427 /* src[] = { buffer_index, offset }. No const_index */
428 LOAD(ssbo
, 2, 0, xx
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
429 /* src[] = { offset }. const_index[] = { base, component } */
430 LOAD(output
, 1, 1, BASE
, COMPONENT
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
431 /* src[] = { vertex, offset }. const_index[] = { base, component } */
432 LOAD(per_vertex_output
, 2, 1, BASE
, COMPONENT
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
433 /* src[] = { offset }. const_index[] = { base } */
434 LOAD(shared
, 1, 1, BASE
, xx
, xx
, NIR_INTRINSIC_CAN_ELIMINATE
)
435 /* src[] = { offset }. const_index[] = { base, range } */
436 LOAD(push_constant
, 1, 2, BASE
, RANGE
, xx
,
437 NIR_INTRINSIC_CAN_ELIMINATE
| NIR_INTRINSIC_CAN_REORDER
)
440 * Stores work the same way as loads, except now the first source is the value
441 * to store and the second (and possibly third) source specify where to store
442 * the value. SSBO and shared memory stores also have a write mask as
446 #define STORE(name, srcs, num_indices, idx0, idx1, idx2, flags) \
447 INTRINSIC(store_##name, srcs, ARR(0, 1, 1, 1), false, 0, 0, num_indices, idx0, idx1, idx2, flags)
449 /* src[] = { value, offset }. const_index[] = { base, write_mask, component } */
450 STORE(output
, 2, 3, BASE
, WRMASK
, COMPONENT
, 0)
451 /* src[] = { value, vertex, offset }.
452 * const_index[] = { base, write_mask, component }
454 STORE(per_vertex_output
, 3, 3, BASE
, WRMASK
, COMPONENT
, 0)
455 /* src[] = { value, block_index, offset }. const_index[] = { write_mask } */
456 STORE(ssbo
, 3, 1, WRMASK
, xx
, xx
, 0)
457 /* src[] = { value, offset }. const_index[] = { base, write_mask } */
458 STORE(shared
, 2, 2, BASE
, WRMASK
, xx
, 0)
460 LAST_INTRINSIC(store_shared
)
462 #undef DEFINE_SYSTEM_VALUE
464 #undef LAST_INTRINSIC