X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fi915%2Fi915_fpc_translate.c;h=70016ed51e6a0b6bb3b60e4123cdb6e26bb804d7;hb=31da39ddc92e780dc539bf34d2de7f82fc65fa86;hp=df606e27e265f6d221f2aca7799e0e427be659d3;hpb=cc78eb63cdc0022684ca6816f258ea4492431916;p=mesa.git diff --git a/src/gallium/drivers/i915/i915_fpc_translate.c b/src/gallium/drivers/i915/i915_fpc_translate.c index df606e27e26..70016ed51e6 100644 --- a/src/gallium/drivers/i915/i915_fpc_translate.c +++ b/src/gallium/drivers/i915/i915_fpc_translate.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * Copyright 2007 VMware, Inc. * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -18,7 +18,7 @@ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. @@ -31,6 +31,7 @@ #include "i915_reg.h" #include "i915_context.h" #include "i915_fpc.h" +#include "i915_debug_private.h" #include "pipe/p_shader_tokens.h" #include "util/u_math.h" @@ -49,24 +50,27 @@ * Simple pass-through fragment shader to use when we don't have * a real shader (or it fails to compile for some reason). */ -static unsigned passthrough[] = +static unsigned passthrough_decl[] = { _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1), /* declare input color: */ - (D0_DCL | - (REG_TYPE_T << D0_TYPE_SHIFT) | - (T_DIFFUSE << D0_NR_SHIFT) | + (D0_DCL | + (REG_TYPE_T << D0_TYPE_SHIFT) | + (T_DIFFUSE << D0_NR_SHIFT) | D0_CHANNEL_ALL), 0, 0, +}; +static unsigned passthrough_program[] = +{ /* move to output color: */ - (A0_MOV | - (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) | - A0_DEST_CHANNEL_ALL | + (A0_MOV | + (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) | + A0_DEST_CHANNEL_ALL | (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) | (T_DIFFUSE << A0_SRC0_NR_SHIFT)), 0x01230000, /* .xyzw */ @@ -107,7 +111,7 @@ static const float cos_constants[4] = { 1.0, /** * component-wise negation of ureg */ -static INLINE int +static inline int negate(int reg, int x, int y, int z, int w) { /* Another neat thing about the UREG representation */ @@ -125,10 +129,13 @@ negate(int reg, int x, int y, int z, int w) static void i915_use_passthrough_shader(struct i915_fragment_shader *fs) { - fs->program = (uint *) MALLOC(sizeof(passthrough)); + fs->program = (uint *) MALLOC(sizeof(passthrough_program)); + fs->decl = (uint *) MALLOC(sizeof(passthrough_decl)); if (fs->program) { - memcpy(fs->program, passthrough, sizeof(passthrough)); - fs->program_len = Elements(passthrough); + memcpy(fs->program, passthrough_program, sizeof(passthrough_program)); + memcpy(fs->decl, passthrough_decl, sizeof(passthrough_decl)); + fs->program_len = ARRAY_SIZE(passthrough_program); + fs->decl_len = ARRAY_SIZE(passthrough_decl); } fs->num_constants = 0; } @@ -141,7 +148,7 @@ i915_program_error(struct i915_fp_compile *p, const char *msg, ...) char buffer[1024]; debug_printf("i915_program_error: "); - va_start( args, msg ); + va_start( args, msg ); util_vsnprintf( buffer, sizeof(buffer), msg, args ); va_end( args ); debug_printf("%s", buffer); @@ -172,8 +179,8 @@ static uint get_mapping(struct i915_fragment_shader* fs, int unit) */ static uint src_vector(struct i915_fp_compile *p, - const struct tgsi_full_src_register *source, - struct i915_fragment_shader* fs) + const struct i915_full_src_register *source, + struct i915_fragment_shader *fs) { uint index = source->Register.Index; uint src = 0, sem_name, sem_ind; @@ -287,7 +294,7 @@ src_vector(struct i915_fp_compile *p, */ static uint get_result_vector(struct i915_fp_compile *p, - const struct tgsi_full_dst_register *dest) + const struct i915_full_dst_register *dest) { switch (dest->Register.File) { case TGSI_FILE_OUTPUT: @@ -316,13 +323,13 @@ get_result_vector(struct i915_fp_compile *p, * Compute flags for saturation and writemask. */ static uint -get_result_flags(const struct tgsi_full_instruction *inst) +get_result_flags(const struct i915_full_instruction *inst) { const uint writeMask = inst->Dst[0].Register.WriteMask; uint flags = 0x0; - if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE) + if (inst->Instruction.Saturate) flags |= A0_DEST_SATURATE; if (writeMask & TGSI_WRITEMASK_X) @@ -372,13 +379,40 @@ translate_tex_src_target(struct i915_fp_compile *p, uint tex) } } +/** + * Return the number of coords needed to access a given TGSI_TEXTURE_* + */ +uint +i915_num_coords(uint tex) +{ + switch (tex) { + case TGSI_TEXTURE_SHADOW1D: + case TGSI_TEXTURE_1D: + return 1; + + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_SHADOWRECT: + case TGSI_TEXTURE_RECT: + return 2; + + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: + return 3; + + default: + debug_printf("Unknown texture target for num coords"); + return 2; + } +} + /** * Generate texel lookup instruction. */ static void emit_tex(struct i915_fp_compile *p, - const struct tgsi_full_instruction *inst, + const struct i915_full_instruction *inst, uint opcode, struct i915_fragment_shader* fs) { @@ -393,7 +427,8 @@ emit_tex(struct i915_fp_compile *p, get_result_flags( inst ), sampler, coord, - opcode); + opcode, + i915_num_coords(texture) ); } @@ -404,9 +439,9 @@ emit_tex(struct i915_fp_compile *p, */ static void emit_simple_arith(struct i915_fp_compile *p, - const struct tgsi_full_instruction *inst, + const struct i915_full_instruction *inst, uint opcode, uint numArgs, - struct i915_fragment_shader* fs) + struct i915_fragment_shader *fs) { uint arg1, arg2, arg3; @@ -429,11 +464,11 @@ emit_simple_arith(struct i915_fp_compile *p, /** As above, but swap the first two src regs */ static void emit_simple_arith_swap2(struct i915_fp_compile *p, - const struct tgsi_full_instruction *inst, + const struct i915_full_instruction *inst, uint opcode, uint numArgs, - struct i915_fragment_shader* fs) + struct i915_fragment_shader *fs) { - struct tgsi_full_instruction inst2; + struct i915_full_instruction inst2; assert(numArgs == 2); @@ -454,10 +489,10 @@ emit_simple_arith_swap2(struct i915_fp_compile *p, * SIN, COS -- could use another taylor step? * LIT -- results seem a little different to sw mesa * LOG -- different to mesa on negative numbers, but this is conformant. - */ + */ static void i915_translate_instruction(struct i915_fp_compile *p, - const struct tgsi_full_instruction *inst, + const struct i915_full_instruction *inst, struct i915_fragment_shader *fs) { uint writemask; @@ -478,13 +513,29 @@ i915_translate_instruction(struct i915_fp_compile *p, emit_simple_arith(p, inst, A0_ADD, 2, fs); break; + case TGSI_OPCODE_CEIL: + src0 = src_vector(p, &inst->Src[0], fs); + tmp = i915_get_utemp(p); + flags = get_result_flags(inst); + i915_emit_arith(p, + A0_FLR, + tmp, + flags & A0_DEST_CHANNEL_ALL, 0, + negate(src0, 1, 1, 1, 1), 0, 0); + i915_emit_arith(p, + A0_MOV, + get_result_vector(p, &inst->Dst[0]), + flags, 0, + negate(tmp, 1, 1, 1, 1), 0, 0); + break; + case TGSI_OPCODE_CMP: src0 = src_vector(p, &inst->Src[0], fs); src1 = src_vector(p, &inst->Src[1], fs); src2 = src_vector(p, &inst->Src[2], fs); - i915_emit_arith(p, A0_CMP, + i915_emit_arith(p, A0_CMP, get_result_vector(p, &inst->Dst[0]), - get_result_flags(inst), + get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */ break; @@ -500,7 +551,7 @@ i915_translate_instruction(struct i915_fp_compile *p, i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0); /* - * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1 + * t0.xy = MUL x.xx11, x.x111 ; x^2, x, 1, 1 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1 * result = DP4 t0, cos_constants @@ -612,7 +663,7 @@ i915_translate_instruction(struct i915_fp_compile *p, emit_simple_arith(p, inst, A0_FRC, 1, fs); break; - case TGSI_OPCODE_KIL: + case TGSI_OPCODE_KILL_IF: /* kill if src[0].x < 0 || src[0].y < 0 ... */ src0 = src_vector(p, &inst->Src[0], fs); tmp = i915_get_utemp(p); @@ -622,11 +673,21 @@ i915_translate_instruction(struct i915_fp_compile *p, A0_DEST_CHANNEL_ALL, /* dest writemask */ 0, /* sampler */ src0, /* coord*/ - T0_TEXKILL); /* opcode */ + T0_TEXKILL, /* opcode */ + 1); /* num_coord */ break; - case TGSI_OPCODE_KILP: - assert(0); /* not tested yet */ + case TGSI_OPCODE_KILL: + /* unconditional kill */ + tmp = i915_get_utemp(p); + + i915_emit_texld(p, + tmp, /* dest reg: a dummy reg */ + A0_DEST_CHANNEL_ALL, /* dest writemask */ + 0, /* sampler */ + negate(swizzle(0, ONE, ONE, ONE, ONE), 1, 1, 1, 1), /* coord */ + T0_TEXKILL, /* opcode */ + 1); /* num_coord */ break; case TGSI_OPCODE_LG2: @@ -703,21 +764,7 @@ i915_translate_instruction(struct i915_fp_compile *p, break; case TGSI_OPCODE_MIN: - src0 = src_vector(p, &inst->Src[0], fs); - src1 = src_vector(p, &inst->Src[1], fs); - tmp = i915_get_utemp(p); - flags = get_result_flags(inst); - - i915_emit_arith(p, - A0_MAX, - tmp, flags & A0_DEST_CHANNEL_ALL, 0, - negate(src0, 1, 1, 1, 1), - negate(src1, 1, 1, 1, 1), 0); - - i915_emit_arith(p, - A0_MOV, - get_result_vector(p, &inst->Dst[0]), - flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0); + emit_simple_arith(p, inst, A0_MIN, 2, fs); break; case TGSI_OPCODE_MOV: @@ -728,6 +775,9 @@ i915_translate_instruction(struct i915_fp_compile *p, emit_simple_arith(p, inst, A0_MUL, 2, fs); break; + case TGSI_OPCODE_NOP: + break; + case TGSI_OPCODE_POW: src0 = src_vector(p, &inst->Src[0], fs); src1 = src_vector(p, &inst->Src[1], fs); @@ -748,11 +798,11 @@ i915_translate_instruction(struct i915_fp_compile *p, get_result_vector(p, &inst->Dst[0]), flags, 0, swizzle(tmp, X, X, X, X), 0, 0); break; - + case TGSI_OPCODE_RET: /* XXX: no-op? */ break; - + case TGSI_OPCODE_RCP: src0 = src_vector(p, &inst->Src[0], fs); @@ -1044,107 +1094,107 @@ i915_translate_instruction(struct i915_fp_compile *p, } -/** - * Translate TGSI fragment shader into i915 hardware instructions. - * \param p the translation state - * \param tokens the TGSI token array - */ -static void -i915_translate_instructions(struct i915_fp_compile *p, - const struct tgsi_token *tokens, - struct i915_fragment_shader *fs) +static void i915_translate_token(struct i915_fp_compile *p, + const union i915_full_token *token, + struct i915_fragment_shader *fs) { struct i915_fragment_shader *ifs = p->shader; - struct tgsi_parse_context parse; - - tgsi_parse_init( &parse, tokens ); - - while( !tgsi_parse_end_of_tokens( &parse ) ) { - - tgsi_parse_token( &parse ); + switch( token->Token.Type ) { + case TGSI_TOKEN_TYPE_PROPERTY: + /* + * We only support one cbuf, but we still need to ignore the property + * correctly so we don't hit the assert at the end of the switch case. + */ + assert(token->FullProperty.Property.PropertyName == + TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS); + break; - switch( parse.FullToken.Token.Type ) { - case TGSI_TOKEN_TYPE_PROPERTY: - /* - * We only support one cbuf, but we still need to ignore the property - * correctly so we don't hit the assert at the end of the switch case. - */ - assert(parse.FullToken.FullProperty.Property.PropertyName == - TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS); - break; - case TGSI_TOKEN_TYPE_DECLARATION: - if (parse.FullToken.FullDeclaration.Declaration.File - == TGSI_FILE_CONSTANT) { - uint i; - for (i = parse.FullToken.FullDeclaration.Range.First; - i <= parse.FullToken.FullDeclaration.Range.Last; - i++) { - assert(ifs->constant_flags[i] == 0x0); - ifs->constant_flags[i] = I915_CONSTFLAG_USER; - ifs->num_constants = MAX2(ifs->num_constants, i + 1); - } + case TGSI_TOKEN_TYPE_DECLARATION: + if (token->FullDeclaration.Declaration.File + == TGSI_FILE_CONSTANT) { + uint i; + for (i = token->FullDeclaration.Range.First; + i <= MIN2(token->FullDeclaration.Range.Last, I915_MAX_CONSTANT - 1); + i++) { + assert(ifs->constant_flags[i] == 0x0); + ifs->constant_flags[i] = I915_CONSTFLAG_USER; + ifs->num_constants = MAX2(ifs->num_constants, i + 1); } - else if (parse.FullToken.FullDeclaration.Declaration.File - == TGSI_FILE_TEMPORARY) { - uint i; - for (i = parse.FullToken.FullDeclaration.Range.First; - i <= parse.FullToken.FullDeclaration.Range.Last; - i++) { - if (i >= I915_MAX_TEMPORARY) - debug_printf("Too many temps (%d)\n",i); - else - /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */ - p->temp_flag |= (1 << i); /* mark temp as used */ - } + } + else if (token->FullDeclaration.Declaration.File + == TGSI_FILE_TEMPORARY) { + uint i; + for (i = token->FullDeclaration.Range.First; + i <= token->FullDeclaration.Range.Last; + i++) { + if (i >= I915_MAX_TEMPORARY) + debug_printf("Too many temps (%d)\n",i); + else + /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */ + p->temp_flag |= (1 << i); /* mark temp as used */ } - break; + } + break; - case TGSI_TOKEN_TYPE_IMMEDIATE: - { - const struct tgsi_full_immediate *imm - = &parse.FullToken.FullImmediate; - const uint pos = p->num_immediates++; - uint j; - assert( imm->Immediate.NrTokens <= 4 + 1 ); - for (j = 0; j < imm->Immediate.NrTokens - 1; j++) { - p->immediates[pos][j] = imm->u[j].Float; - } + case TGSI_TOKEN_TYPE_IMMEDIATE: + { + const struct tgsi_full_immediate *imm + = &token->FullImmediate; + const uint pos = p->num_immediates++; + uint j; + assert( imm->Immediate.NrTokens <= 4 + 1 ); + for (j = 0; j < imm->Immediate.NrTokens - 1; j++) { + p->immediates[pos][j] = imm->u[j].Float; } - break; + } + break; - case TGSI_TOKEN_TYPE_INSTRUCTION: - if (p->first_instruction) { - /* resolve location of immediates */ - uint i, j; - for (i = 0; i < p->num_immediates; i++) { - /* find constant slot for this immediate */ - for (j = 0; j < I915_MAX_CONSTANT; j++) { - if (ifs->constant_flags[j] == 0x0) { - memcpy(ifs->constants[j], - p->immediates[i], - 4 * sizeof(float)); - /*printf("immediate %d maps to const %d\n", i, j);*/ - ifs->constant_flags[j] = 0xf; /* all four comps used */ - p->immediates_map[i] = j; - ifs->num_constants = MAX2(ifs->num_constants, j + 1); - break; - } + case TGSI_TOKEN_TYPE_INSTRUCTION: + if (p->first_instruction) { + /* resolve location of immediates */ + uint i, j; + for (i = 0; i < p->num_immediates; i++) { + /* find constant slot for this immediate */ + for (j = 0; j < I915_MAX_CONSTANT; j++) { + if (ifs->constant_flags[j] == 0x0) { + memcpy(ifs->constants[j], + p->immediates[i], + 4 * sizeof(float)); + /*printf("immediate %d maps to const %d\n", i, j);*/ + ifs->constant_flags[j] = 0xf; /* all four comps used */ + p->immediates_map[i] = j; + ifs->num_constants = MAX2(ifs->num_constants, j + 1); + break; } } - - p->first_instruction = FALSE; } - i915_translate_instruction(p, &parse.FullToken.FullInstruction, fs); - break; - - default: - assert( 0 ); + p->first_instruction = FALSE; } - } /* while */ + i915_translate_instruction(p, &token->FullInstruction, fs); + break; + + default: + assert( 0 ); + } - tgsi_parse_free (&parse); +} + +/** + * Translate TGSI fragment shader into i915 hardware instructions. + * \param p the translation state + * \param tokens the TGSI token array + */ +static void +i915_translate_instructions(struct i915_fp_compile *p, + const struct i915_token_list *tokens, + struct i915_fragment_shader *fs) +{ + int i; + for(i = 0; iNumTokens; i++) { + i915_translate_token(p, &tokens->Tokens[i], fs); + } } @@ -1234,18 +1284,27 @@ i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p) /* Copy compilation results to fragment program struct: */ + assert(!ifs->decl); assert(!ifs->program); + + ifs->decl + = (uint *) MALLOC(decl_size * sizeof(uint)); ifs->program - = (uint *) MALLOC((program_size + decl_size) * sizeof(uint)); - if (ifs->program) { - ifs->program_len = program_size + decl_size; + = (uint *) MALLOC(program_size * sizeof(uint)); - memcpy(ifs->program, - p->declarations, + if (ifs->decl) { + ifs->decl_len = decl_size; + + memcpy(ifs->decl, + p->declarations, decl_size * sizeof(uint)); + } + + if (ifs->program) { + ifs->program_len = program_size; - memcpy(ifs->program + decl_size, - p->program, + memcpy(ifs->program, + p->program, program_size * sizeof(uint)); } } @@ -1288,6 +1347,7 @@ i915_translate_fragment_program( struct i915_context *i915, { struct i915_fp_compile *p; const struct tgsi_token *tokens = fs->state.tokens; + struct i915_token_list* i_tokens; #if 0 tgsi_dump(tokens, 0); @@ -1303,8 +1363,14 @@ i915_translate_fragment_program( struct i915_context *i915, p = i915_init_compile(i915, fs); - i915_translate_instructions(p, tokens, fs); + i_tokens = i915_optimize(tokens); + i915_translate_instructions(p, i_tokens, fs); i915_fixup_depth_write(p); i915_fini_compile(i915, p); + i915_optimize_free(i_tokens); + +#if 0 + i915_disassemble_program(NULL, fs->program, fs->program_len); +#endif }