src/gallium/drivers/r600/r600_compiler_r700.c

   1 /*
   2  * Copyright 2010 Jerome Glisse <glisse@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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  */
  23 #include <stdlib.h>
  24 #include <string.h>
  25 #include <stdint.h>
  26 #include <stdio.h>
  27 #include <errno.h>
  28 #include "r600_context.h"
  29 #include "r700_sq.h"
  30
  31 static int r700_shader_cf_node_bytecode(struct r600_shader *rshader,
  32                                         struct r600_shader_node *rnode,
  33                                         unsigned *cid)
  34 {
  35         unsigned id = *cid;
  36
  37         if (rnode->nfetch) {
  38                 rshader->bcode[id++] = S_SQ_CF_WORD0_ADDR(rnode->cf_addr >> 1);
  39                 rshader->bcode[id++] = S_SQ_CF_WORD1_CF_INST(V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
  40                                         S_SQ_CF_WORD1_BARRIER(1) |
  41                                         S_SQ_CF_WORD1_COUNT(rnode->nfetch - 1);
  42         } else {
  43                 rshader->bcode[id++] = S_SQ_CF_ALU_WORD0_ADDR(rnode->cf_addr >> 1);
  44                 rshader->bcode[id++] = S_SQ_CF_ALU_WORD1_CF_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU) |
  45                                         S_SQ_CF_ALU_WORD1_BARRIER(1) |
  46                                         S_SQ_CF_ALU_WORD1_COUNT(rnode->nslot - 1);
  47         }
  48         *cid = id;
  49         return 0;
  50 }
  51
  52 static int r700_shader_cf_output_bytecode(struct r600_shader *rshader,
  53                                                 struct c_vector *v,
  54                                                 unsigned *cid,
  55                                                 unsigned end)
  56 {
  57         struct r600_shader_operand out;
  58         unsigned id = *cid;
  59         int r;
  60
  61         r = r600_shader_find_gpr(rshader, v, 0, &out);
  62         if (r)
  63                 return r;
  64         rshader->bcode[id + 0] = S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(out.sel) |
  65                                 S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(3);
  66         rshader->bcode[id + 1] = S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(0) |
  67                 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(1) |
  68                 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(2) |
  69                 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(3) |
  70                 S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(1) |
  71                 S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE) |
  72                 S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(end);
  73         switch (v->name) {
  74         case C_SEMANTIC_POSITION:
  75                 rshader->bcode[id + 0] |= S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(60) |
  76                         S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS);
  77                 break;
  78         case C_SEMANTIC_COLOR:
  79                 if (rshader->cshader.type == C_PROGRAM_TYPE_VS) {
  80                         rshader->output[rshader->noutput].gpr = out.sel;
  81                         rshader->output[rshader->noutput].sid = v->sid;
  82                         rshader->output[rshader->noutput].name = v->name;
  83                         rshader->bcode[id + 0] |= S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(rshader->noutput++) |
  84                                 S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM);
  85                 } else {
  86                         rshader->bcode[id + 0] |= S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(0) |
  87                                 S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL);
  88                 }
  89                 break;
  90         case C_SEMANTIC_GENERIC:
  91                 rshader->output[rshader->noutput].gpr = out.sel;
  92                 rshader->output[rshader->noutput].sid = v->sid;
  93                 rshader->output[rshader->noutput].name = v->name;
  94                 rshader->bcode[id + 0] |= S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(rshader->noutput++) |
  95                         S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM);
  96                 break;
  97         default:
  98                 fprintf(stderr, "%s:%d unsupported\n", __func__, __LINE__);
  99                 return -EINVAL;
 100         }
 101         *cid = id + 2;
 102         return 0;
 103 }
 104
 105 static int r700_shader_alu_bytecode(struct r600_shader *rshader,
 106                                         struct r600_shader_node *rnode,
 107                                         struct r600_shader_inst *alu,
 108                                         unsigned *cid)
 109 {
 110         unsigned id = *cid;
 111
 112         /* don't replace gpr by pv or ps for destination register */
 113         if (alu->is_op3) {
 114                 rshader->bcode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
 115                                         S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) |
 116                                         S_SQ_ALU_WORD0_SRC1_SEL(alu->src[1].sel) |
 117                                         S_SQ_ALU_WORD0_SRC1_CHAN(alu->src[1].chan) |
 118                                         S_SQ_ALU_WORD0_LAST(alu->last);
 119                 rshader->bcode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) |
 120                                         S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
 121                                         S_SQ_ALU_WORD1_OP3_SRC2_SEL(alu->src[2].sel) |
 122                                         S_SQ_ALU_WORD1_OP3_SRC2_CHAN(alu->src[2].chan) |
 123                                         S_SQ_ALU_WORD1_OP3_SRC2_NEG(alu->src[2].neg) |
 124                                         S_SQ_ALU_WORD1_OP3_ALU_INST(alu->opcode) |
 125                                         S_SQ_ALU_WORD1_BANK_SWIZZLE(0);
 126         } else {
 127                 rshader->bcode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
 128                                         S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) |
 129                                         S_SQ_ALU_WORD0_SRC0_NEG(alu->src[0].neg) |
 130                                         S_SQ_ALU_WORD0_SRC1_SEL(alu->src[1].sel) |
 131                                         S_SQ_ALU_WORD0_SRC1_CHAN(alu->src[1].chan) |
 132                                         S_SQ_ALU_WORD0_SRC1_NEG(alu->src[1].neg) |
 133                                         S_SQ_ALU_WORD0_LAST(alu->last);
 134                 rshader->bcode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) |
 135                                         S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
 136                                         S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) |
 137                                         S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) |
 138                                         S_SQ_ALU_WORD1_OP2_WRITE_MASK(1) |
 139                                         S_SQ_ALU_WORD1_OP2_ALU_INST(alu->opcode) |
 140                                         S_SQ_ALU_WORD1_BANK_SWIZZLE(0);
 141         }
 142         *cid = id;
 143         return 0;
 144 }
 145
 146 static int r700_shader_alu_translate(struct r600_shader *rshader,
 147                                      struct r600_shader_node *rnode,
 148                                      unsigned *cid)
 149
 150 {
 151         struct r600_shader_alu *alu;
 152         unsigned id = *cid;
 153         int i;
 154         int r = 0;
 155         LIST_FOR_EACH_ENTRY(alu, &rnode->alu, head) {
 156                 for (i = 0; i < alu->nalu; i++) {
 157                         r = r700_shader_alu_bytecode(rshader, rnode, &alu->alu[i], &id);
 158                         if (r)
 159                                 goto out;
 160                 }
 161                 for (i = 0; i < alu->nliteral; i++) {
 162                         rshader->bcode[id++] = alu->literal[i];
 163                 }
 164         }
 165  out:
 166         *cid = id;
 167         return r;
 168 }
 169
 170 int r700_shader_translate(struct r600_shader *rshader)
 171 {
 172         struct c_shader *shader = &rshader->cshader;
 173         struct r600_shader_node *rnode;
 174         struct r600_shader_vfetch *vfetch;
 175         struct c_vector *v;
 176         unsigned id, end;
 177         int r;
 178
 179         r = r600_shader_register(rshader);
 180         if (r) {
 181                 fprintf(stderr, "%s %d register allocation failed\n", __FILE__, __LINE__);
 182                 return r;
 183         }
 184         r = r600_shader_translate_rec(rshader, &shader->entry);
 185         if (r) {
 186                 fprintf(stderr, "%s %d translation failed\n", __FILE__, __LINE__);
 187                 return r;
 188         }
 189         r = r600_shader_legalize(rshader);
 190         if (r) {
 191                 fprintf(stderr, "%s %d legalize failed\n", __FILE__, __LINE__);
 192                 return r;
 193         }
 194         r600_shader_node_place(rshader);
 195         rshader->bcode = malloc(rshader->ndw * 4);
 196         if (rshader->bcode == NULL)
 197                 return -ENOMEM;
 198         LIST_FOR_EACH_ENTRY(rnode, &rshader->nodes, head) {
 199                 id = rnode->cf_addr;
 200                 LIST_FOR_EACH_ENTRY(vfetch, &rnode->vfetch, head) {
 201                         r = r600_shader_vfetch_bytecode(rshader, rnode, vfetch, &id);
 202                         if (r)
 203                                 return r;
 204                 }
 205                 if (rshader->r6xx_compile)
 206                         r = r6xx_shader_alu_translate(rshader, rnode, &id);
 207                 else
 208                         r = r700_shader_alu_translate(rshader, rnode, &id);
 209                 if (r)
 210                         return r;
 211         }
 212         id = 0;
 213         LIST_FOR_EACH_ENTRY(rnode, &rshader->nodes, head) {
 214                 r = r700_shader_cf_node_bytecode(rshader, rnode, &id);
 215                 if (r)
 216                         return r;
 217         }
 218         LIST_FOR_EACH_ENTRY(v, &rshader->cshader.files[C_FILE_OUTPUT].vectors, head) {
 219                 end = 0;
 220                 if (v->head.next == &rshader->cshader.files[C_FILE_OUTPUT].vectors)
 221                         end = 1;
 222                 r = r700_shader_cf_output_bytecode(rshader, v, &id, end);
 223                 if (r)
 224                         return r;
 225         }
 226         LIST_FOR_EACH_ENTRY(v, &rshader->cshader.files[C_FILE_INPUT].vectors, head) {
 227                 rshader->input[rshader->ninput].gpr = rshader->ninput;
 228                 rshader->input[rshader->ninput].sid = v->sid;
 229                 rshader->input[rshader->ninput].name = v->name;
 230                 rshader->ninput++;
 231         }
 232         return 0;
 233 }