13464ebd5Sriastradh/* 23464ebd5Sriastradh * Copyright 2010 Jerome Glisse <glisse@freedesktop.org> 33464ebd5Sriastradh * 43464ebd5Sriastradh * Permission is hereby granted, free of charge, to any person obtaining a 53464ebd5Sriastradh * copy of this software and associated documentation files (the "Software"), 63464ebd5Sriastradh * to deal in the Software without restriction, including without limitation 73464ebd5Sriastradh * on the rights to use, copy, modify, merge, publish, distribute, sub 83464ebd5Sriastradh * license, and/or sell copies of the Software, and to permit persons to whom 93464ebd5Sriastradh * the Software is furnished to do so, subject to the following conditions: 103464ebd5Sriastradh * 113464ebd5Sriastradh * The above copyright notice and this permission notice (including the next 123464ebd5Sriastradh * paragraph) shall be included in all copies or substantial portions of the 133464ebd5Sriastradh * Software. 143464ebd5Sriastradh * 153464ebd5Sriastradh * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 163464ebd5Sriastradh * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 173464ebd5Sriastradh * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 183464ebd5Sriastradh * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 193464ebd5Sriastradh * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 203464ebd5Sriastradh * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 213464ebd5Sriastradh * USE OR OTHER DEALINGS IN THE SOFTWARE. 223464ebd5Sriastradh */ 233464ebd5Sriastradh#include "r600_asm.h" 243464ebd5Sriastradh#include "r700_sq.h" 253464ebd5Sriastradh 26af69d88dSmrgvoid r700_bytecode_cf_vtx_build(uint32_t *bytecode, const struct r600_bytecode_cf *cf) 273464ebd5Sriastradh{ 283464ebd5Sriastradh unsigned count = (cf->ndw / 4) - 1; 293464ebd5Sriastradh *bytecode++ = S_SQ_CF_WORD0_ADDR(cf->addr >> 1); 30af69d88dSmrg *bytecode++ = S_SQ_CF_WORD1_CF_INST(r600_isa_cf_opcode(ISA_CC_R700, cf->op)) | 313464ebd5Sriastradh S_SQ_CF_WORD1_BARRIER(1) | 323464ebd5Sriastradh S_SQ_CF_WORD1_COUNT(count) | 3301e04c3fSmrg S_SQ_CF_WORD1_COUNT_3(count >> 3)| 3401e04c3fSmrg S_SQ_CF_WORD1_END_OF_PROGRAM(cf->end_of_program); 353464ebd5Sriastradh} 363464ebd5Sriastradh 37af69d88dSmrgint r700_bytecode_alu_build(struct r600_bytecode *bc, struct r600_bytecode_alu *alu, unsigned id) 383464ebd5Sriastradh{ 393464ebd5Sriastradh bc->bytecode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) | 403464ebd5Sriastradh S_SQ_ALU_WORD0_SRC0_REL(alu->src[0].rel) | 413464ebd5Sriastradh S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) | 423464ebd5Sriastradh S_SQ_ALU_WORD0_SRC0_NEG(alu->src[0].neg) | 433464ebd5Sriastradh S_SQ_ALU_WORD0_SRC1_SEL(alu->src[1].sel) | 443464ebd5Sriastradh S_SQ_ALU_WORD0_SRC1_REL(alu->src[1].rel) | 453464ebd5Sriastradh S_SQ_ALU_WORD0_SRC1_CHAN(alu->src[1].chan) | 463464ebd5Sriastradh S_SQ_ALU_WORD0_SRC1_NEG(alu->src[1].neg) | 47af69d88dSmrg S_SQ_ALU_WORD0_PRED_SEL(alu->pred_sel) | 483464ebd5Sriastradh S_SQ_ALU_WORD0_LAST(alu->last); 493464ebd5Sriastradh 503464ebd5Sriastradh /* don't replace gpr by pv or ps for destination register */ 513464ebd5Sriastradh if (alu->is_op3) { 5201e04c3fSmrg assert(!alu->src[0].abs && !alu->src[1].abs && !alu->src[2].abs); 533464ebd5Sriastradh bc->bytecode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) | 543464ebd5Sriastradh S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) | 553464ebd5Sriastradh S_SQ_ALU_WORD1_DST_REL(alu->dst.rel) | 563464ebd5Sriastradh S_SQ_ALU_WORD1_CLAMP(alu->dst.clamp) | 573464ebd5Sriastradh S_SQ_ALU_WORD1_OP3_SRC2_SEL(alu->src[2].sel) | 583464ebd5Sriastradh S_SQ_ALU_WORD1_OP3_SRC2_REL(alu->src[2].rel) | 593464ebd5Sriastradh S_SQ_ALU_WORD1_OP3_SRC2_CHAN(alu->src[2].chan) | 603464ebd5Sriastradh S_SQ_ALU_WORD1_OP3_SRC2_NEG(alu->src[2].neg) | 61af69d88dSmrg S_SQ_ALU_WORD1_OP3_ALU_INST(r600_isa_alu_opcode(bc->isa->hw_class, alu->op)) | 623464ebd5Sriastradh S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle); 633464ebd5Sriastradh } else { 643464ebd5Sriastradh bc->bytecode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) | 653464ebd5Sriastradh S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) | 663464ebd5Sriastradh S_SQ_ALU_WORD1_DST_REL(alu->dst.rel) | 673464ebd5Sriastradh S_SQ_ALU_WORD1_CLAMP(alu->dst.clamp) | 683464ebd5Sriastradh S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) | 693464ebd5Sriastradh S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) | 703464ebd5Sriastradh S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) | 713464ebd5Sriastradh S_SQ_ALU_WORD1_OP2_OMOD(alu->omod) | 72af69d88dSmrg S_SQ_ALU_WORD1_OP2_ALU_INST(r600_isa_alu_opcode(bc->isa->hw_class, alu->op)) | 733464ebd5Sriastradh S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) | 74af69d88dSmrg S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->execute_mask) | 75af69d88dSmrg S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu->update_pred); 763464ebd5Sriastradh } 773464ebd5Sriastradh return 0; 783464ebd5Sriastradh} 79af69d88dSmrg 80af69d88dSmrgvoid r700_bytecode_alu_read(struct r600_bytecode *bc, 81af69d88dSmrg struct r600_bytecode_alu *alu, uint32_t word0, uint32_t word1) 82af69d88dSmrg{ 83af69d88dSmrg /* WORD0 */ 84af69d88dSmrg alu->src[0].sel = G_SQ_ALU_WORD0_SRC0_SEL(word0); 85af69d88dSmrg alu->src[0].rel = G_SQ_ALU_WORD0_SRC0_REL(word0); 86af69d88dSmrg alu->src[0].chan = G_SQ_ALU_WORD0_SRC0_CHAN(word0); 87af69d88dSmrg alu->src[0].neg = G_SQ_ALU_WORD0_SRC0_NEG(word0); 88af69d88dSmrg alu->src[1].sel = G_SQ_ALU_WORD0_SRC1_SEL(word0); 89af69d88dSmrg alu->src[1].rel = G_SQ_ALU_WORD0_SRC1_REL(word0); 90af69d88dSmrg alu->src[1].chan = G_SQ_ALU_WORD0_SRC1_CHAN(word0); 91af69d88dSmrg alu->src[1].neg = G_SQ_ALU_WORD0_SRC1_NEG(word0); 92af69d88dSmrg alu->index_mode = G_SQ_ALU_WORD0_INDEX_MODE(word0); 93af69d88dSmrg alu->pred_sel = G_SQ_ALU_WORD0_PRED_SEL(word0); 94af69d88dSmrg alu->last = G_SQ_ALU_WORD0_LAST(word0); 95af69d88dSmrg 96af69d88dSmrg /* WORD1 */ 97af69d88dSmrg alu->bank_swizzle = G_SQ_ALU_WORD1_BANK_SWIZZLE(word1); 98af69d88dSmrg if (alu->bank_swizzle) 99af69d88dSmrg alu->bank_swizzle_force = alu->bank_swizzle; 100af69d88dSmrg alu->dst.sel = G_SQ_ALU_WORD1_DST_GPR(word1); 101af69d88dSmrg alu->dst.rel = G_SQ_ALU_WORD1_DST_REL(word1); 102af69d88dSmrg alu->dst.chan = G_SQ_ALU_WORD1_DST_CHAN(word1); 103af69d88dSmrg alu->dst.clamp = G_SQ_ALU_WORD1_CLAMP(word1); 104af69d88dSmrg if (G_SQ_ALU_WORD1_ENCODING(word1)) /*ALU_DWORD1_OP3*/ 105af69d88dSmrg { 106af69d88dSmrg alu->is_op3 = 1; 107af69d88dSmrg alu->src[2].sel = G_SQ_ALU_WORD1_OP3_SRC2_SEL(word1); 108af69d88dSmrg alu->src[2].rel = G_SQ_ALU_WORD1_OP3_SRC2_REL(word1); 109af69d88dSmrg alu->src[2].chan = G_SQ_ALU_WORD1_OP3_SRC2_CHAN(word1); 110af69d88dSmrg alu->src[2].neg = G_SQ_ALU_WORD1_OP3_SRC2_NEG(word1); 111af69d88dSmrg alu->op = r600_isa_alu_by_opcode(bc->isa, 112af69d88dSmrg G_SQ_ALU_WORD1_OP3_ALU_INST(word1), 1); 113af69d88dSmrg } 114af69d88dSmrg else /*ALU_DWORD1_OP2*/ 115af69d88dSmrg { 116af69d88dSmrg alu->src[0].abs = G_SQ_ALU_WORD1_OP2_SRC0_ABS(word1); 117af69d88dSmrg alu->src[1].abs = G_SQ_ALU_WORD1_OP2_SRC1_ABS(word1); 118af69d88dSmrg alu->op = r600_isa_alu_by_opcode(bc->isa, 119af69d88dSmrg G_SQ_ALU_WORD1_OP2_ALU_INST(word1), 0); 120af69d88dSmrg alu->omod = G_SQ_ALU_WORD1_OP2_OMOD(word1); 121af69d88dSmrg alu->dst.write = G_SQ_ALU_WORD1_OP2_WRITE_MASK(word1); 122af69d88dSmrg alu->update_pred = G_SQ_ALU_WORD1_OP2_UPDATE_PRED(word1); 123af69d88dSmrg alu->execute_mask = 124af69d88dSmrg G_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(word1); 125af69d88dSmrg } 126af69d88dSmrg} 12701e04c3fSmrg 12801e04c3fSmrgint r700_bytecode_fetch_mem_build(struct r600_bytecode *bc, struct r600_bytecode_vtx *mem, unsigned id) 12901e04c3fSmrg{ 13001e04c3fSmrg unsigned opcode = r600_isa_fetch_opcode(bc->isa->hw_class, mem->op) >> 8; 13101e04c3fSmrg 13201e04c3fSmrg bc->bytecode[id++] = S_SQ_MEM_RD_WORD0_MEM_INST(2) | 13301e04c3fSmrg S_SQ_MEM_RD_WORD0_ELEM_SIZE(mem->elem_size) | 13401e04c3fSmrg S_SQ_MEM_RD_WORD0_FETCH_WHOLE_QUAD(0) | 13501e04c3fSmrg S_SQ_MEM_RD_WORD0_MEM_OP(opcode) | 13601e04c3fSmrg S_SQ_MEM_RD_WORD0_UNCACHED(mem->uncached) | 13701e04c3fSmrg S_SQ_MEM_RD_WORD0_INDEXED(mem->indexed) | 13801e04c3fSmrg S_SQ_MEM_RD_WORD0_SRC_SEL_Y(mem->src_sel_y) | 13901e04c3fSmrg S_SQ_MEM_RD_WORD0_SRC_GPR(mem->src_gpr) | 14001e04c3fSmrg S_SQ_MEM_RD_WORD0_SRC_REL(mem->src_rel) | 14101e04c3fSmrg S_SQ_MEM_RD_WORD0_SRC_SEL_X(mem->src_sel_x) | 14201e04c3fSmrg S_SQ_MEM_RD_WORD0_BURST_COUNT(mem->burst_count) | 14301e04c3fSmrg S_SQ_MEM_RD_WORD0_LDS_REQ(0) | 14401e04c3fSmrg S_SQ_MEM_RD_WORD0_COALESCED_READ(0); 14501e04c3fSmrg 14601e04c3fSmrg bc->bytecode[id++] = S_SQ_MEM_RD_WORD1_DST_GPR(mem->dst_gpr) | 14701e04c3fSmrg S_SQ_MEM_RD_WORD1_DST_REL(mem->dst_rel) | 14801e04c3fSmrg S_SQ_MEM_RD_WORD1_DST_SEL_X(mem->dst_sel_x) | 14901e04c3fSmrg S_SQ_MEM_RD_WORD1_DST_SEL_Y(mem->dst_sel_y) | 15001e04c3fSmrg S_SQ_MEM_RD_WORD1_DST_SEL_W(mem->dst_sel_w) | 15101e04c3fSmrg S_SQ_MEM_RD_WORD1_DST_SEL_Z(mem->dst_sel_z) | 15201e04c3fSmrg S_SQ_MEM_RD_WORD1_DATA_FORMAT(mem->data_format) | 15301e04c3fSmrg S_SQ_MEM_RD_WORD1_NUM_FORMAT_ALL(mem->num_format_all) | 15401e04c3fSmrg S_SQ_MEM_RD_WORD1_FORMAT_COMP_ALL(mem->format_comp_all) | 15501e04c3fSmrg S_SQ_MEM_RD_WORD1_SRF_MODE_ALL(mem->srf_mode_all); 15601e04c3fSmrg 15701e04c3fSmrg bc->bytecode[id++] = S_SQ_MEM_RD_WORD2_ARRAY_BASE(mem->array_base) | 15801e04c3fSmrg S_SQ_MEM_RD_WORD2_ENDIAN_SWAP(0) | 15901e04c3fSmrg S_SQ_MEM_RD_WORD2_ARRAY_SIZE(mem->array_size); 16001e04c3fSmrg 16101e04c3fSmrg 16201e04c3fSmrg bc->bytecode[id++] = 0; /* MEM ops are 4 word aligned */ 16301e04c3fSmrg 16401e04c3fSmrg return 0; 16501e04c3fSmrg} 166