Path: blob/21.2-virgl/src/gallium/drivers/r300/r300_tgsi_to_rc.c
4570 views
/*1* Copyright 2009 Nicolai Hähnle <[email protected]>2*3* Permission is hereby granted, free of charge, to any person obtaining a4* copy of this software and associated documentation files (the "Software"),5* to deal in the Software without restriction, including without limitation6* on the rights to use, copy, modify, merge, publish, distribute, sub7* license, and/or sell copies of the Software, and to permit persons to whom8* the Software is furnished to do so, subject to the following conditions:9*10* The above copyright notice and this permission notice (including the next11* paragraph) shall be included in all copies or substantial portions of the12* Software.13*14* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR15* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,16* FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL17* THE COPYRIGHT HOLDER(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,18* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR19* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE20* USE OR OTHER DEALINGS IN THE SOFTWARE. */2122#include <stdio.h>23#include "r300_tgsi_to_rc.h"2425#include "compiler/radeon_compiler.h"2627#include "tgsi/tgsi_info.h"28#include "tgsi/tgsi_parse.h"29#include "tgsi/tgsi_scan.h"30#include "tgsi/tgsi_util.h"3132#include "util/compiler.h"3334static unsigned translate_opcode(unsigned opcode)35{36switch(opcode) {37case TGSI_OPCODE_ARL: return RC_OPCODE_ARL;38case TGSI_OPCODE_MOV: return RC_OPCODE_MOV;39case TGSI_OPCODE_LIT: return RC_OPCODE_LIT;40case TGSI_OPCODE_RCP: return RC_OPCODE_RCP;41case TGSI_OPCODE_RSQ: return RC_OPCODE_RSQ;42case TGSI_OPCODE_EXP: return RC_OPCODE_EXP;43case TGSI_OPCODE_LOG: return RC_OPCODE_LOG;44case TGSI_OPCODE_MUL: return RC_OPCODE_MUL;45case TGSI_OPCODE_ADD: return RC_OPCODE_ADD;46case TGSI_OPCODE_DP3: return RC_OPCODE_DP3;47case TGSI_OPCODE_DP4: return RC_OPCODE_DP4;48case TGSI_OPCODE_DST: return RC_OPCODE_DST;49case TGSI_OPCODE_MIN: return RC_OPCODE_MIN;50case TGSI_OPCODE_MAX: return RC_OPCODE_MAX;51case TGSI_OPCODE_SLT: return RC_OPCODE_SLT;52case TGSI_OPCODE_SGE: return RC_OPCODE_SGE;53case TGSI_OPCODE_MAD: return RC_OPCODE_MAD;54case TGSI_OPCODE_LRP: return RC_OPCODE_LRP;55case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;56case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;57case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND;58case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;59case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;60case TGSI_OPCODE_POW: return RC_OPCODE_POW;61case TGSI_OPCODE_COS: return RC_OPCODE_COS;62case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;63case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;64case TGSI_OPCODE_KILL: return RC_OPCODE_KILP;65/* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */66/* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */67/* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */68/* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */69case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;70case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;71case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;72case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;73case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;74case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;75case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;76case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;77/* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */78/* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */79/* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */80/* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */81case TGSI_OPCODE_ARR: return RC_OPCODE_ARR;82/* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */83/* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */84case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;85case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;86case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;87/* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */88case TGSI_OPCODE_DP2: return RC_OPCODE_DP2;89case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;90case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;91case TGSI_OPCODE_IF: return RC_OPCODE_IF;92case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;93case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;94case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;95case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;96/* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */97/* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */98case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;99/* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */100/* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */101case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC;102/* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */103/* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */104/* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */105/* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */106/* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */107/* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */108/* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */109/* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */110case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;111/* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */112/* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */113/* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */114/* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */115/* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */116/* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */117case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;118case TGSI_OPCODE_KILL_IF: return RC_OPCODE_KIL;119}120121fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));122return RC_OPCODE_ILLEGAL_OPCODE;123}124125static unsigned translate_saturate(unsigned saturate)126{127return saturate ? RC_SATURATE_ZERO_ONE : RC_SATURATE_NONE;128}129130static unsigned translate_register_file(unsigned file)131{132switch(file) {133case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;134case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;135case TGSI_FILE_INPUT: return RC_FILE_INPUT;136case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;137default:138fprintf(stderr, "Unhandled register file: %i\n", file);139FALLTHROUGH;140case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;141case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;142}143}144145static int translate_register_index(146struct tgsi_to_rc * ttr,147unsigned file,148int index)149{150if (file == TGSI_FILE_IMMEDIATE)151return ttr->immediate_offset + index;152153return index;154}155156static void transform_dstreg(157struct tgsi_to_rc * ttr,158struct rc_dst_register * dst,159struct tgsi_full_dst_register * src)160{161dst->File = translate_register_file(src->Register.File);162dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);163dst->WriteMask = src->Register.WriteMask;164165if (src->Register.Indirect) {166ttr->error = TRUE;167fprintf(stderr, "r300: Relative addressing of destination operands "168"is unsupported.\n");169}170}171172static void transform_srcreg(173struct tgsi_to_rc * ttr,174struct rc_src_register * dst,175struct tgsi_full_src_register * src)176{177unsigned i, j;178179dst->File = translate_register_file(src->Register.File);180dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);181dst->RelAddr = src->Register.Indirect;182dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);183dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;184dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;185dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;186dst->Abs = src->Register.Absolute;187dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;188189if (src->Register.File == TGSI_FILE_IMMEDIATE) {190for (i = 0; i < ttr->imms_to_swizzle_count; i++) {191if (ttr->imms_to_swizzle[i].index == src->Register.Index) {192dst->File = RC_FILE_TEMPORARY;193dst->Index = 0;194dst->Swizzle = 0;195for (j = 0; j < 4; j++) {196dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,197tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);198}199break;200}201}202}203}204205static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,206uint32_t *shadowSamplers)207{208switch(src.Texture) {209case TGSI_TEXTURE_1D:210dst->U.I.TexSrcTarget = RC_TEXTURE_1D;211break;212case TGSI_TEXTURE_2D:213dst->U.I.TexSrcTarget = RC_TEXTURE_2D;214break;215case TGSI_TEXTURE_3D:216dst->U.I.TexSrcTarget = RC_TEXTURE_3D;217break;218case TGSI_TEXTURE_CUBE:219dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;220break;221case TGSI_TEXTURE_RECT:222dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;223break;224case TGSI_TEXTURE_SHADOW1D:225dst->U.I.TexSrcTarget = RC_TEXTURE_1D;226dst->U.I.TexShadow = 1;227*shadowSamplers |= 1 << dst->U.I.TexSrcUnit;228break;229case TGSI_TEXTURE_SHADOW2D:230dst->U.I.TexSrcTarget = RC_TEXTURE_2D;231dst->U.I.TexShadow = 1;232*shadowSamplers |= 1 << dst->U.I.TexSrcUnit;233break;234case TGSI_TEXTURE_SHADOWRECT:235dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;236dst->U.I.TexShadow = 1;237*shadowSamplers |= 1 << dst->U.I.TexSrcUnit;238break;239}240dst->U.I.TexSwizzle = RC_SWIZZLE_XYZW;241}242243static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)244{245struct rc_instruction * dst;246int i;247248dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);249dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);250dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);251252if (src->Instruction.NumDstRegs)253transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);254255for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {256if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)257dst->U.I.TexSrcUnit = src->Src[i].Register.Index;258else259transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);260}261262/* Texturing. */263if (src->Instruction.Texture)264transform_texture(dst, src->Texture,265&ttr->compiler->Program.ShadowSamplers);266}267268static void handle_immediate(struct tgsi_to_rc * ttr,269struct tgsi_full_immediate * imm,270unsigned index)271{272struct rc_constant constant;273unsigned swizzle = 0;274boolean can_swizzle = TRUE;275unsigned i;276277for (i = 0; i < 4; i++) {278if (imm->u[i].Float == 0.0f) {279swizzle |= RC_SWIZZLE_ZERO << (i * 3);280} else if (imm->u[i].Float == 0.5f && ttr->use_half_swizzles) {281swizzle |= RC_SWIZZLE_HALF << (i * 3);282} else if (imm->u[i].Float == 1.0f) {283swizzle |= RC_SWIZZLE_ONE << (i * 3);284} else {285can_swizzle = FALSE;286break;287}288}289290if (can_swizzle) {291ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;292ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;293ttr->imms_to_swizzle_count++;294} else {295constant.Type = RC_CONSTANT_IMMEDIATE;296constant.Size = 4;297for(i = 0; i < 4; ++i)298constant.u.Immediate[i] = imm->u[i].Float;299rc_constants_add(&ttr->compiler->Program.Constants, &constant);300}301}302303void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,304const struct tgsi_token * tokens)305{306struct tgsi_full_instruction *inst;307struct tgsi_parse_context parser;308unsigned imm_index = 0;309int i;310311ttr->error = FALSE;312313/* Allocate constants placeholders.314*315* Note: What if declared constants are not contiguous? */316for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {317struct rc_constant constant;318memset(&constant, 0, sizeof(constant));319constant.Type = RC_CONSTANT_EXTERNAL;320constant.Size = 4;321constant.u.External = i;322rc_constants_add(&ttr->compiler->Program.Constants, &constant);323}324325ttr->immediate_offset = ttr->compiler->Program.Constants.Count;326327ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));328ttr->imms_to_swizzle_count = 0;329330tgsi_parse_init(&parser, tokens);331332while (!tgsi_parse_end_of_tokens(&parser)) {333tgsi_parse_token(&parser);334335switch (parser.FullToken.Token.Type) {336case TGSI_TOKEN_TYPE_DECLARATION:337break;338case TGSI_TOKEN_TYPE_IMMEDIATE:339handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);340imm_index++;341break;342case TGSI_TOKEN_TYPE_INSTRUCTION:343inst = &parser.FullToken.FullInstruction;344if (inst->Instruction.Opcode == TGSI_OPCODE_END) {345break;346}347348transform_instruction(ttr, inst);349break;350}351}352353tgsi_parse_free(&parser);354355free(ttr->imms_to_swizzle);356357rc_calculate_inputs_outputs(ttr->compiler);358}359360361