src/compiler/nir/nir_intrinsics.h

   1 /*
   2  * Copyright © 2014 Intel Corporation
   3  *
   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:
  10  *
  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
  13  * Software.
  14  *
  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
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Connor Abbott (cwabbott0@gmail.com)
  25  *
  26  */
  27
  28 /**
  29  * This header file defines all the available intrinsics in one place. It
  30  * expands to a list of macros of the form:
  31  *
  32  * INTRINSIC(name, num_srcs, src_components, has_dest, dest_components,
  33  *              num_variables, num_indices, idx0, idx1, idx2, flags)
  34  *
  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
  39  * for each intrinsic.
  40  */
  41
  42 #define ARR(...) { __VA_ARGS__ }
  43
  44 INTRINSIC(nop, 0, ARR(0), false, 0, 0, 0, xx, xx, xx,
  45           NIR_INTRINSIC_CAN_ELIMINATE)
  46
  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)
  50
  51 /*
  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
  56  * respectively.
  57  */
  58
  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)
  65
  66 /*
  67  * Ask the driver for the size of a given buffer. It takes the buffer index
  68  * as source.
  69  */
  70 INTRINSIC(get_buffer_size, 1, ARR(1), true, 1, 0, 0, xx, xx, xx,
  71           NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
  72
  73 /*
  74  * a barrier is an intrinsic with no inputs/outputs but which can't be moved
  75  * around/optimized in general
  76  */
  77 #define BARRIER(name) INTRINSIC(name, 0, ARR(0), false, 0, 0, 0, xx, xx, xx, 0)
  78
  79 BARRIER(barrier)
  80 BARRIER(discard)
  81
  82 /*
  83  * Memory barrier with semantics analogous to the memoryBarrier() GLSL
  84  * intrinsic.
  85  */
  86 BARRIER(memory_barrier)
  87
  88 /*
  89  * Shader clock intrinsic with semantics analogous to the clock2x32ARB()
  90  * GLSL intrinsic.
  91  * The latter can be used as code motion barrier, which is currently not
  92  * feasible with NIR.
  93  */
  94 INTRINSIC(shader_clock, 0, ARR(0), true, 2, 0, 0, xx, xx, xx, NIR_INTRINSIC_CAN_ELIMINATE)
  95
  96 /*
  97  * Shader ballot intrinsics with semantics analogous to the
  98  *
  99  *    ballotARB()
 100  *    readInvocationARB()
 101  *    readFirstInvocationARB()
 102  *
 103  * GLSL functions from ARB_shader_ballot.
 104  */
 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)
 108
 109 /*
 110  * Memory barrier with semantics analogous to the compute shader
 111  * groupMemoryBarrier(), memoryBarrierAtomicCounter(), memoryBarrierBuffer(),
 112  * memoryBarrierImage() and memoryBarrierShared() GLSL intrinsics.
 113  */
 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)
 119
 120 /** A conditional discard, with a single boolean source. */
 121 INTRINSIC(discard_if, 1, ARR(1), false, 0, 0, 0, xx, xx, xx, 0)
 122
 123 /** ARB_shader_group_vote intrinsics */
 124 INTRINSIC(vote_any, 1, ARR(1), true, 1, 0, 0, xx, xx, xx, NIR_INTRINSIC_CAN_ELIMINATE)
 125 INTRINSIC(vote_all, 1, ARR(1), true, 1, 0, 0, xx, xx, xx, NIR_INTRINSIC_CAN_ELIMINATE)
 126 INTRINSIC(vote_eq,  1, ARR(1), true, 1, 0, 0, xx, xx, xx, NIR_INTRINSIC_CAN_ELIMINATE)
 127
 128 /**
 129  * Basic Geometry Shader intrinsics.
 130  *
 131  * emit_vertex implements GLSL's EmitStreamVertex() built-in.  It takes a single
 132  * index, which is the stream ID to write to.
 133  *
 134  * end_primitive implements GLSL's EndPrimitive() built-in.
 135  */
 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)
 138
 139 /**
 140  * Geometry Shader intrinsics with a vertex count.
 141  *
 142  * Alternatively, drivers may implement these intrinsics, and use
 143  * nir_lower_gs_intrinsics() to convert from the basic intrinsics.
 144  *
 145  * These maintain a count of the number of vertices emitted, as an additional
 146  * unsigned integer source.
 147  */
 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)
 151
 152 /*
 153  * Atomic counters
 154  *
 155  * The *_var variants take an atomic_uint nir_variable, while the other,
 156  * lowered, variants take a constant buffer index and register offset.
 157  */
 158
 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)
 168
 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)
 180
 181 /*
 182  * Image load, store and atomic intrinsics.
 183  *
 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.
 187  *
 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.
 195  */
 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)
 211
 212 /*
 213  * Vulkan descriptor set intrinsic
 214  *
 215  * The Vulkan API uses a different binding model from GL.  In the Vulkan
 216  * API, all external resources are represented by a tuple:
 217  *
 218  * (descriptor set, binding, array index)
 219  *
 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.
 224  *
 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.
 227  */
 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)
 231
 232 /*
 233  * variable atomic intrinsics
 234  *
 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.
 238  *
 239  * All operations take 1 source except CompSwap that takes 2. These sources
 240  * represent:
 241  *
 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.
 245  *
 246  * All operations take 1 variable deref.
 247  */
 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)
 258
 259 /*
 260  * SSBO atomic intrinsics
 261  *
 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.
 265  *
 266  * All operations take 3 sources except CompSwap that takes 4. These
 267  * sources represent:
 268  *
 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.
 275  */
 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)
 286
 287 /*
 288  * CS shared variable atomic intrinsics
 289  *
 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.
 293  *
 294  * All operations take 2 sources except CompSwap that takes 3. These
 295  * sources represent:
 296  *
 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.
 302  */
 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)
 313
 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)
 317 #endif
 318
 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, \
 322    idx0, idx1, idx2, \
 323    NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 324
 325 SYSTEM_VALUE(frag_coord, 4, 0, xx, xx, xx)
 326 SYSTEM_VALUE(front_face, 1, 0, xx, xx, xx)
 327 SYSTEM_VALUE(vertex_id, 1, 0, xx, xx, xx)
 328 SYSTEM_VALUE(vertex_id_zero_base, 1, 0, xx, xx, xx)
 329 SYSTEM_VALUE(base_vertex, 1, 0, xx, xx, xx)
 330 SYSTEM_VALUE(instance_id, 1, 0, xx, xx, xx)
 331 SYSTEM_VALUE(base_instance, 1, 0, xx, xx, xx)
 332 SYSTEM_VALUE(draw_id, 1, 0, xx, xx, xx)
 333 SYSTEM_VALUE(sample_id, 1, 0, xx, xx, xx)
 334 SYSTEM_VALUE(sample_pos, 2, 0, xx, xx, xx)
 335 SYSTEM_VALUE(sample_mask_in, 1, 0, xx, xx, xx)
 336 SYSTEM_VALUE(primitive_id, 1, 0, xx, xx, xx)
 337 SYSTEM_VALUE(invocation_id, 1, 0, xx, xx, xx)
 338 SYSTEM_VALUE(tess_coord, 3, 0, xx, xx, xx)
 339 SYSTEM_VALUE(tess_level_outer, 4, 0, xx, xx, xx)
 340 SYSTEM_VALUE(tess_level_inner, 2, 0, xx, xx, xx)
 341 SYSTEM_VALUE(patch_vertices_in, 1, 0, xx, xx, xx)
 342 SYSTEM_VALUE(local_invocation_id, 3, 0, xx, xx, xx)
 343 SYSTEM_VALUE(local_invocation_index, 1, 0, xx, xx, xx)
 344 SYSTEM_VALUE(work_group_id, 3, 0, xx, xx, xx)
 345 SYSTEM_VALUE(user_clip_plane, 4, 1, UCP_ID, xx, xx)
 346 SYSTEM_VALUE(num_work_groups, 3, 0, xx, xx, xx)
 347 SYSTEM_VALUE(helper_invocation, 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)
 351 SYSTEM_VALUE(subgroup_size, 1, 0, xx, xx, xx)
 352 SYSTEM_VALUE(subgroup_invocation, 1, 0, xx, xx, xx)
 353 SYSTEM_VALUE(subgroup_eq_mask, 1, 0, xx, xx, xx)
 354 SYSTEM_VALUE(subgroup_ge_mask, 1, 0, xx, xx, xx)
 355 SYSTEM_VALUE(subgroup_gt_mask, 1, 0, xx, xx, xx)
 356 SYSTEM_VALUE(subgroup_le_mask, 1, 0, xx, xx, xx)
 357 SYSTEM_VALUE(subgroup_lt_mask, 1, 0, xx, xx, xx)
 358
 359 /* Blend constant color values.  Float values are clamped. */
 360 SYSTEM_VALUE(blend_const_color_r_float, 1, 0, xx, xx, xx)
 361 SYSTEM_VALUE(blend_const_color_g_float, 1, 0, xx, xx, xx)
 362 SYSTEM_VALUE(blend_const_color_b_float, 1, 0, xx, xx, xx)
 363 SYSTEM_VALUE(blend_const_color_a_float, 1, 0, xx, xx, xx)
 364 SYSTEM_VALUE(blend_const_color_rgba8888_unorm, 1, 0, xx, xx, xx)
 365 SYSTEM_VALUE(blend_const_color_aaaa8888_unorm, 1, 0, xx, xx, xx)
 366
 367 /**
 368  * Barycentric coordinate intrinsics.
 369  *
 370  * These set up the barycentric coordinates for a particular interpolation.
 371  * The first three are for the simple cases: pixel, centroid, or per-sample
 372  * (at gl_SampleID).  The next two handle interpolating at a specified
 373  * sample location, or interpolating with a vec2 offset,
 374  *
 375  * The interp_mode index should be either the INTERP_MODE_SMOOTH or
 376  * INTERP_MODE_NOPERSPECTIVE enum values.
 377  *
 378  * The vec2 value produced by these intrinsics is intended for use as the
 379  * barycoord source of a load_interpolated_input intrinsic.
 380  */
 381
 382 #define BARYCENTRIC(name, sources, source_components) \
 383    INTRINSIC(load_barycentric_##name, sources, ARR(source_components), \
 384              true, 2, 0, 1, INTERP_MODE, xx, xx, \
 385              NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 386
 387 /* no sources.  const_index[] = { interp_mode } */
 388 BARYCENTRIC(pixel, 0, 0)
 389 BARYCENTRIC(centroid, 0, 0)
 390 BARYCENTRIC(sample, 0, 0)
 391 /* src[] = { sample_id }.  const_index[] = { interp_mode } */
 392 BARYCENTRIC(at_sample, 1, 1)
 393 /* src[] = { offset.xy }.  const_index[] = { interp_mode } */
 394 BARYCENTRIC(at_offset, 1, 2)
 395
 396 /*
 397  * Load operations pull data from some piece of GPU memory.  All load
 398  * operations operate in terms of offsets into some piece of theoretical
 399  * memory.  Loads from externally visible memory (UBO and SSBO) simply take a
 400  * byte offset as a source.  Loads from opaque memory (uniforms, inputs, etc.)
 401  * take a base+offset pair where the base (const_index[0]) gives the location
 402  * of the start of the variable being loaded and and the offset source is a
 403  * offset into that variable.
 404  *
 405  * Uniform load operations have a second "range" index that specifies the
 406  * range (starting at base) of the data from which we are loading.  If
 407  * const_index[1] == 0, then the range is unknown.
 408  *
 409  * Some load operations such as UBO/SSBO load and per_vertex loads take an
 410  * additional source to specify which UBO/SSBO/vertex to load from.
 411  *
 412  * The exact address type depends on the lowering pass that generates the
 413  * load/store intrinsics.  Typically, this is vec4 units for things such as
 414  * varying slots and float units for fragment shader inputs.  UBO and SSBO
 415  * offsets are always in bytes.
 416  */
 417
 418 #define LOAD(name, srcs, num_indices, idx0, idx1, idx2, flags) \
 419    INTRINSIC(load_##name, srcs, ARR(1, 1, 1, 1), true, 0, 0, num_indices, idx0, idx1, idx2, flags)
 420
 421 /* src[] = { offset }. const_index[] = { base, range } */
 422 LOAD(uniform, 1, 2, BASE, RANGE, xx, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 423 /* src[] = { buffer_index, offset }. No const_index */
 424 LOAD(ubo, 2, 0, xx, xx, xx, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 425 /* src[] = { offset }. const_index[] = { base, component } */
 426 LOAD(input, 1, 2, BASE, COMPONENT, xx, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 427 /* src[] = { vertex, offset }. const_index[] = { base, component } */
 428 LOAD(per_vertex_input, 2, 2, BASE, COMPONENT, xx, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 429 /* src[] = { barycoord, offset }. const_index[] = { base, component } */
 430 INTRINSIC(load_interpolated_input, 2, ARR(2, 1), true, 0, 0,
 431           2, BASE, COMPONENT, xx,
 432           NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 433
 434 /* src[] = { buffer_index, offset }. No const_index */
 435 LOAD(ssbo, 2, 0, xx, xx, xx, NIR_INTRINSIC_CAN_ELIMINATE)
 436 /* src[] = { offset }. const_index[] = { base, component } */
 437 LOAD(output, 1, 1, BASE, COMPONENT, xx, NIR_INTRINSIC_CAN_ELIMINATE)
 438 /* src[] = { vertex, offset }. const_index[] = { base, component } */
 439 LOAD(per_vertex_output, 2, 1, BASE, COMPONENT, xx, NIR_INTRINSIC_CAN_ELIMINATE)
 440 /* src[] = { offset }. const_index[] = { base } */
 441 LOAD(shared, 1, 1, BASE, xx, xx, NIR_INTRINSIC_CAN_ELIMINATE)
 442 /* src[] = { offset }. const_index[] = { base, range } */
 443 LOAD(push_constant, 1, 2, BASE, RANGE, xx,
 444      NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 445
 446 /*
 447  * Stores work the same way as loads, except now the first source is the value
 448  * to store and the second (and possibly third) source specify where to store
 449  * the value.  SSBO and shared memory stores also have a write mask as
 450  * const_index[0].
 451  */
 452
 453 #define STORE(name, srcs, num_indices, idx0, idx1, idx2, flags) \
 454    INTRINSIC(store_##name, srcs, ARR(0, 1, 1, 1), false, 0, 0, num_indices, idx0, idx1, idx2, flags)
 455
 456 /* src[] = { value, offset }. const_index[] = { base, write_mask, component } */
 457 STORE(output, 2, 3, BASE, WRMASK, COMPONENT, 0)
 458 /* src[] = { value, vertex, offset }.
 459  * const_index[] = { base, write_mask, component }
 460  */
 461 STORE(per_vertex_output, 3, 3, BASE, WRMASK, COMPONENT, 0)
 462 /* src[] = { value, block_index, offset }. const_index[] = { write_mask } */
 463 STORE(ssbo, 3, 1, WRMASK, xx, xx, 0)
 464 /* src[] = { value, offset }. const_index[] = { base, write_mask } */
 465 STORE(shared, 2, 2, BASE, WRMASK, xx, 0)
 466
 467 LAST_INTRINSIC(store_shared)
 468
 469 #undef DEFINE_SYSTEM_VALUE
 470 #undef INTRINSIC
 471 #undef LAST_INTRINSIC