src/glsl/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, 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
  45 INTRINSIC(load_var, 0, ARR(), true, 0, 1, 0, NIR_INTRINSIC_CAN_ELIMINATE)
  46 INTRINSIC(store_var, 1, ARR(0), false, 0, 1, 0, 0)
  47 INTRINSIC(copy_var, 0, ARR(), false, 0, 2, 0, 0)
  48
  49 /*
  50  * Interpolation of input.  The interp_var_at* intrinsics are similar to the
  51  * load_var intrinsic acting an a shader input except that they interpolate
  52  * the input differently.  The at_sample and at_offset intrinsics take an
  53  * aditional source that is a integer sample id or a vec2 position offset
  54  * respectively.
  55  */
  56
  57 INTRINSIC(interp_var_at_centroid, 0, ARR(0), true, 0, 1, 0,
  58           NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
  59 INTRINSIC(interp_var_at_sample, 1, ARR(1), true, 0, 1, 0,
  60           NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
  61 INTRINSIC(interp_var_at_offset, 1, ARR(2), true, 0, 1, 0,
  62           NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
  63
  64 /*
  65  * a barrier is an intrinsic with no inputs/outputs but which can't be moved
  66  * around/optimized in general
  67  */
  68 #define BARRIER(name) INTRINSIC(name, 0, ARR(), false, 0, 0, 0, 0)
  69
  70 BARRIER(discard)
  71 /** A conditional discard, with a single boolean source. */
  72 INTRINSIC(discard_if, 1, ARR(1), false, 0, 0, 0, 0)
  73
  74 INTRINSIC(emit_vertex,   0, ARR(), false, 0, 0, 1, 0)
  75 INTRINSIC(end_primitive, 0, ARR(), false, 0, 0, 1, 0)
  76
  77 /*
  78  * Atomic counters
  79  *
  80  * The *_var variants take an atomic_uint nir_variable, while the other,
  81  * lowered, variants take a constant buffer index and register offset.
  82  */
  83
  84 #define ATOMIC(name, flags) \
  85    INTRINSIC(atomic_counter_##name##_var, 0, ARR(), true, 1, 1, 0, flags) \
  86    INTRINSIC(atomic_counter_##name, 1, ARR(1), true, 1, 0, 1, flags)
  87
  88 ATOMIC(inc, 0)
  89 ATOMIC(dec, 0)
  90 ATOMIC(read, NIR_INTRINSIC_CAN_ELIMINATE)
  91
  92 /*
  93  * Image load, store and atomic intrinsics.
  94  *
  95  * All image intrinsics take an image target passed as a nir_variable.  Image
  96  * variables contain a number of memory and layout qualifiers that influence
  97  * the semantics of the intrinsic.
  98  *
  99  * All image intrinsics take a four-coordinate vector and a sample index as
 100  * first two sources, determining the location within the image that will be
 101  * accessed by the intrinsic.  Components not applicable to the image target
 102  * in use are undefined.  Image store takes an additional four-component
 103  * argument with the value to be written, and image atomic operations take
 104  * either one or two additional scalar arguments with the same meaning as in
 105  * the ARB_shader_image_load_store specification.
 106  */
 107 INTRINSIC(image_load, 2, ARR(4, 1), true, 4, 1, 0,
 108           NIR_INTRINSIC_CAN_ELIMINATE)
 109 INTRINSIC(image_store, 3, ARR(4, 1, 4), false, 0, 1, 0, 0)
 110 INTRINSIC(image_atomic_add, 3, ARR(4, 1, 1), true, 1, 1, 0, 0)
 111 INTRINSIC(image_atomic_min, 3, ARR(4, 1, 1), true, 1, 1, 0, 0)
 112 INTRINSIC(image_atomic_max, 3, ARR(4, 1, 1), true, 1, 1, 0, 0)
 113 INTRINSIC(image_atomic_and, 3, ARR(4, 1, 1), true, 1, 1, 0, 0)
 114 INTRINSIC(image_atomic_or, 3, ARR(4, 1, 1), true, 1, 1, 0, 0)
 115 INTRINSIC(image_atomic_xor, 3, ARR(4, 1, 1), true, 1, 1, 0, 0)
 116 INTRINSIC(image_atomic_exchange, 3, ARR(4, 1, 1), true, 1, 1, 0, 0)
 117 INTRINSIC(image_atomic_comp_swap, 4, ARR(4, 1, 1, 1), true, 1, 1, 0, 0)
 118
 119 #define SYSTEM_VALUE(name, components) \
 120    INTRINSIC(load_##name, 0, ARR(), true, components, 0, 0, \
 121    NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 122
 123 SYSTEM_VALUE(front_face, 1)
 124 SYSTEM_VALUE(vertex_id, 1)
 125 SYSTEM_VALUE(vertex_id_zero_base, 1)
 126 SYSTEM_VALUE(base_vertex, 1)
 127 SYSTEM_VALUE(instance_id, 1)
 128 SYSTEM_VALUE(sample_id, 1)
 129 SYSTEM_VALUE(sample_pos, 2)
 130 SYSTEM_VALUE(sample_mask_in, 1)
 131 SYSTEM_VALUE(invocation_id, 1)
 132
 133 /*
 134  * The first index is the address to load from, and the second index is the
 135  * number of array elements to load.  Indirect loads have an additional
 136  * register input, which is added to the constant address to compute the
 137  * final address to load from.  For UBO's (and SSBO's), the first source is
 138  * the (possibly constant) UBO buffer index and the indirect (if it exists)
 139  * is the second source.
 140  *
 141  * For vector backends, the address is in terms of one vec4, and so each array
 142  * element is +4 scalar components from the previous array element. For scalar
 143  * backends, the address is in terms of a single 4-byte float/int and arrays
 144  * elements begin immediately after the previous array element.
 145  */
 146
 147 #define LOAD(name, extra_srcs, flags) \
 148    INTRINSIC(load_##name, extra_srcs, ARR(1), true, 0, 0, 2, flags) \
 149    INTRINSIC(load_##name##_indirect, extra_srcs + 1, ARR(1, 1), \
 150              true, 0, 0, 2, flags)
 151
 152 LOAD(uniform, 0, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 153 LOAD(ubo, 1, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 154 LOAD(input, 0, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REORDER)
 155 /* LOAD(ssbo, 1, 0) */
 156
 157 /*
 158  * Stores work the same way as loads, except now the first register input is
 159  * the value or array to store and the optional second input is the indirect
 160  * offset.
 161  */
 162
 163 #define STORE(name, num_indices, flags) \
 164    INTRINSIC(store_##name, 1, ARR(0), false, 0, 0, num_indices, flags) \
 165    INTRINSIC(store_##name##_indirect, 2, ARR(0, 1), false, 0, 0, \
 166              num_indices, flags) \
 167
 168 STORE(output, 2, 0)
 169 /* STORE(ssbo, 3, 0) */
 170
 171 LAST_INTRINSIC(store_output_indirect)