src/freedreno/ir3/ir3_context.h

   1 /*
   2  * Copyright (C) 2015-2018 Rob Clark <robclark@freedesktop.org>
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #ifndef IR3_CONTEXT_H_
  28 #define IR3_CONTEXT_H_
  29
  30 #include "ir3_compiler.h"
  31 #include "ir3_nir.h"
  32 #include "ir3.h"
  33
  34 /* for conditionally setting boolean flag(s): */
  35 #define COND(bool, val) ((bool) ? (val) : 0)
  36
  37 #define DBG(fmt, ...) \
  38                 do { debug_printf("%s:%d: "fmt "\n", \
  39                                 __FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0)
  40
  41 /**
  42  * The context for compilation of a single shader.
  43  */
  44 struct ir3_context {
  45         struct ir3_compiler *compiler;
  46         const struct ir3_context_funcs *funcs;
  47
  48         struct nir_shader *s;
  49
  50         struct nir_instr *cur_instr;  /* current instruction, just for debug */
  51
  52         struct ir3 *ir;
  53         struct ir3_shader_variant *so;
  54
  55         /* Tables of scalar inputs/outputs.  Because of the way varying packing
  56          * works, we could have inputs w/ fractional location, which is a bit
  57          * awkward to deal with unless we keep track of the split scalar in/
  58          * out components.
  59          *
  60          * These *only* have inputs/outputs that are touched by load_*input and
  61          * store_output.
  62          */
  63         unsigned ninputs, noutputs;
  64         struct ir3_instruction **inputs;
  65         struct ir3_instruction **outputs;
  66
  67         struct ir3_block *block;      /* the current block */
  68         struct ir3_block *in_block;   /* block created for shader inputs */
  69
  70         nir_function_impl *impl;
  71
  72         /* For fragment shaders, varyings are not actual shader inputs,
  73          * instead the hw passes a ij coord which is used with
  74          * bary.f.
  75          *
  76          * But NIR doesn't know that, it still declares varyings as
  77          * inputs.  So we do all the input tracking normally and fix
  78          * things up after compile_instructions()
  79          */
  80         struct ir3_instruction *ij_pixel, *ij_sample, *ij_centroid, *ij_size;
  81
  82         /* for fragment shaders, for gl_FrontFacing and gl_FragCoord: */
  83         struct ir3_instruction *frag_face, *frag_coord;
  84
  85         /* For vertex shaders, keep track of the system values sources */
  86         struct ir3_instruction *vertex_id, *basevertex, *instance_id, *base_instance;
  87
  88         /* For fragment shaders: */
  89         struct ir3_instruction *samp_id, *samp_mask_in;
  90
  91         /* For geometry shaders: */
  92         struct ir3_instruction *primitive_id;
  93         struct ir3_instruction *gs_header;
  94
  95         /* For tessellation shaders: */
  96         struct ir3_instruction *patch_vertices_in;
  97         struct ir3_instruction *tcs_header;
  98         struct ir3_instruction *tess_coord;
  99
 100         /* Compute shader inputs: */
 101         struct ir3_instruction *local_invocation_id, *work_group_id;
 102
 103         /* mapping from nir_register to defining instruction: */
 104         struct hash_table *def_ht;
 105
 106         unsigned num_arrays;
 107
 108         /* Tracking for max level of flowcontrol (branchstack) needed
 109          * by a5xx+:
 110          */
 111         unsigned stack, max_stack;
 112
 113         /* a common pattern for indirect addressing is to request the
 114          * same address register multiple times.  To avoid generating
 115          * duplicate instruction sequences (which our backend does not
 116          * try to clean up, since that should be done as the NIR stage)
 117          * we cache the address value generated for a given src value:
 118          *
 119          * Note that we have to cache these per alignment, since same
 120          * src used for an array of vec1 cannot be also used for an
 121          * array of vec4.
 122          */
 123         struct hash_table *addr0_ht[4];
 124
 125         /* The same for a1.x. We only support immediate values for a1.x, as this
 126          * is the only use so far.
 127          */
 128         struct hash_table_u64 *addr1_ht;
 129
 130         struct hash_table *sel_cond_conversions;
 131
 132         /* last dst array, for indirect we need to insert a var-store.
 133          */
 134         struct ir3_instruction **last_dst;
 135         unsigned last_dst_n;
 136
 137         /* maps nir_block to ir3_block, mostly for the purposes of
 138          * figuring out the blocks successors
 139          */
 140         struct hash_table *block_ht;
 141
 142         /* on a4xx, bitmask of samplers which need astc+srgb workaround: */
 143         unsigned astc_srgb;
 144
 145         unsigned samples;             /* bitmask of x,y sample shifts */
 146
 147         unsigned max_texture_index;
 148
 149         unsigned prefetch_limit;
 150
 151         /* set if we encounter something we can't handle yet, so we
 152          * can bail cleanly and fallback to TGSI compiler f/e
 153          */
 154         bool error;
 155 };
 156
 157 struct ir3_context_funcs {
 158         void (*emit_intrinsic_load_ssbo)(struct ir3_context *ctx, nir_intrinsic_instr *intr,
 159                         struct ir3_instruction **dst);
 160         void (*emit_intrinsic_store_ssbo)(struct ir3_context *ctx, nir_intrinsic_instr *intr);
 161         struct ir3_instruction * (*emit_intrinsic_atomic_ssbo)(struct ir3_context *ctx, nir_intrinsic_instr *intr);
 162         void (*emit_intrinsic_load_image)(struct ir3_context *ctx, nir_intrinsic_instr *intr,
 163                         struct ir3_instruction **dst);
 164         void (*emit_intrinsic_store_image)(struct ir3_context *ctx, nir_intrinsic_instr *intr);
 165         struct ir3_instruction * (*emit_intrinsic_atomic_image)(struct ir3_context *ctx, nir_intrinsic_instr *intr);
 166         void (*emit_intrinsic_image_size)(struct ir3_context *ctx, nir_intrinsic_instr *intr,
 167                         struct ir3_instruction **dst);
 168 };
 169
 170 extern const struct ir3_context_funcs ir3_a4xx_funcs;
 171 extern const struct ir3_context_funcs ir3_a6xx_funcs;
 172
 173 struct ir3_context * ir3_context_init(struct ir3_compiler *compiler,
 174                 struct ir3_shader_variant *so);
 175 void ir3_context_free(struct ir3_context *ctx);
 176
 177 struct ir3_instruction ** ir3_get_dst_ssa(struct ir3_context *ctx, nir_ssa_def *dst, unsigned n);
 178 struct ir3_instruction ** ir3_get_dst(struct ir3_context *ctx, nir_dest *dst, unsigned n);
 179 struct ir3_instruction * const * ir3_get_src(struct ir3_context *ctx, nir_src *src);
 180 void ir3_put_dst(struct ir3_context *ctx, nir_dest *dst);
 181 struct ir3_instruction * ir3_create_collect(struct ir3_context *ctx,
 182                 struct ir3_instruction *const *arr, unsigned arrsz);
 183 void ir3_split_dest(struct ir3_block *block, struct ir3_instruction **dst,
 184                 struct ir3_instruction *src, unsigned base, unsigned n);
 185 void ir3_handle_bindless_cat6(struct ir3_instruction *instr, nir_src rsrc);
 186 void emit_intrinsic_image_size_tex(struct ir3_context *ctx, nir_intrinsic_instr *intr,
 187                 struct ir3_instruction **dst);
 188
 189 NORETURN void ir3_context_error(struct ir3_context *ctx, const char *format, ...);
 190
 191 #define compile_assert(ctx, cond) do { \
 192                 if (!(cond)) ir3_context_error((ctx), "failed assert: "#cond"\n"); \
 193         } while (0)
 194
 195 struct ir3_instruction * ir3_get_addr0(struct ir3_context *ctx,
 196                 struct ir3_instruction *src, int align);
 197 struct ir3_instruction * ir3_get_addr1(struct ir3_context *ctx,
 198                 unsigned const_val);
 199 struct ir3_instruction * ir3_get_predicate(struct ir3_context *ctx,
 200                 struct ir3_instruction *src);
 201
 202 void ir3_declare_array(struct ir3_context *ctx, nir_register *reg);
 203 struct ir3_array * ir3_get_array(struct ir3_context *ctx, nir_register *reg);
 204 struct ir3_instruction *ir3_create_array_load(struct ir3_context *ctx,
 205                 struct ir3_array *arr, int n, struct ir3_instruction *address,
 206                 unsigned bitsize);
 207 void ir3_create_array_store(struct ir3_context *ctx, struct ir3_array *arr, int n,
 208                 struct ir3_instruction *src, struct ir3_instruction *address);
 209
 210 static inline type_t utype_for_size(unsigned bit_size)
 211 {
 212         switch (bit_size) {
 213         case 32: return TYPE_U32;
 214         case 16: return TYPE_U16;
 215         case  8: return TYPE_U8;
 216         default: unreachable("bad bitsize"); return ~0;
 217         }
 218 }
 219
 220 static inline type_t utype_src(nir_src src)
 221 { return utype_for_size(nir_src_bit_size(src)); }
 222
 223 static inline type_t utype_dst(nir_dest dst)
 224 { return utype_for_size(nir_dest_bit_size(dst)); }
 225
 226 #endif /* IR3_CONTEXT_H_ */