X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fmain%2Fff_fragment_shader.cpp;h=9bb706cc54906b6e1bb12d047ab3949d30881852;hb=5980fc35c9e2e25a5a53cb1b38245ad763885cca;hp=bf65a4f64852f557698b58a8a3bc47b6f0a04885;hpb=e5c6a92a12b5cd7db205d72039f58d302b0be9d5;p=mesa.git diff --git a/src/mesa/main/ff_fragment_shader.cpp b/src/mesa/main/ff_fragment_shader.cpp index bf65a4f6485..9bb706cc549 100644 --- a/src/mesa/main/ff_fragment_shader.cpp +++ b/src/mesa/main/ff_fragment_shader.cpp @@ -1,8 +1,9 @@ /************************************************************************** * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * Copyright 2007 VMware, Inc. * All Rights Reserved. * Copyright 2009 VMware, Inc. All Rights Reserved. + * Copyright © 2010-2011 Intel Corporation * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the @@ -19,7 +20,7 @@ * 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. @@ -30,6 +31,9 @@ extern "C" { #include "glheader.h" #include "imports.h" #include "mtypes.h" +#include "main/context.h" +#include "main/macros.h" +#include "main/samplerobj.h" #include "program/program.h" #include "program/prog_parameter.h" #include "program/prog_cache.h" @@ -38,7 +42,18 @@ extern "C" { #include "program/prog_statevars.h" #include "program/programopt.h" #include "texenvprogram.h" +#include "texobj.h" } +#include "main/uniforms.h" +#include "../glsl/glsl_types.h" +#include "../glsl/ir.h" +#include "../glsl/ir_builder.h" +#include "../glsl/glsl_symbol_table.h" +#include "../glsl/glsl_parser_extras.h" +#include "../glsl/ir_optimization.h" +#include "../program/ir_to_mesa.h" + +using namespace ir_builder; /* * Note on texture units: @@ -59,7 +74,7 @@ struct texenvprog_cache_item { GLuint hash; void *key; - struct gl_fragment_program *data; + struct gl_shader_program *data; struct texenvprog_cache_item *next; }; @@ -76,13 +91,6 @@ texenv_doing_secondary_color(struct gl_context *ctx) return GL_FALSE; } -/** - * Up to nine instructions per tex unit, plus fog, specular color. - */ -#define MAX_INSTRUCTIONS ((MAX_TEXTURE_COORD_UNITS * 9) + 12) - -#define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM) - struct mode_opt { #ifdef __GNUC__ __extension__ GLubyte Source:4; /**< SRC_x */ @@ -105,7 +113,7 @@ struct state_key { /* NOTE: This array of structs must be last! (see "keySize" below) */ struct { GLuint enabled:1; - GLuint source_index:3; /**< TEXTURE_x_INDEX */ + GLuint source_index:4; /**< TEXTURE_x_INDEX */ GLuint shadow:1; GLuint ScaleShiftRGB:2; GLuint ScaleShiftA:2; @@ -116,8 +124,6 @@ struct state_key { GLuint NumArgsA:3; /**< up to MAX_COMBINER_TERMS */ GLuint ModeA:5; /**< MODE_x */ - GLuint texture_cyl_wrap:1; /**< For gallium test/debug only */ - struct mode_opt OptRGB[MAX_COMBINER_TERMS]; struct mode_opt OptA[MAX_COMBINER_TERMS]; } unit[MAX_TEXTURE_UNITS]; @@ -214,7 +220,6 @@ static GLuint translate_source( GLenum src ) #define MODE_MODULATE_SUBTRACT_ATI 12 /* r = a0 * a2 - a1 */ #define MODE_ADD_PRODUCTS 13 /* r = a0 * a1 + a2 * a3 */ #define MODE_ADD_PRODUCTS_SIGNED 14 /* r = a0 * a1 + a2 * a3 - 0.5 */ -#define MODE_BUMP_ENVMAP_ATI 15 /* special */ #define MODE_UNKNOWN 16 /** @@ -244,7 +249,6 @@ static GLuint translate_mode( GLenum envMode, GLenum mode ) case GL_MODULATE_ADD_ATI: return MODE_MODULATE_ADD_ATI; case GL_MODULATE_SIGNED_ADD_ATI: return MODE_MODULATE_SIGNED_ADD_ATI; case GL_MODULATE_SUBTRACT_ATI: return MODE_MODULATE_SUBTRACT_ATI; - case GL_BUMP_ENVMAP_ATI: return MODE_BUMP_ENVMAP_ATI; default: assert(0); return MODE_UNKNOWN; @@ -277,7 +281,6 @@ need_saturate( GLuint mode ) case MODE_MODULATE_SUBTRACT_ATI: case MODE_ADD_PRODUCTS: case MODE_ADD_PRODUCTS_SIGNED: - case MODE_BUMP_ENVMAP_ATI: return GL_TRUE; default: assert(0); @@ -285,20 +288,7 @@ need_saturate( GLuint mode ) } } - - -/** - * Translate TEXTURE_x_BIT to TEXTURE_x_INDEX. - */ -static GLuint translate_tex_src_bit( GLbitfield bit ) -{ - ASSERT(bit); - return _mesa_ffs(bit) - 1; -} - - #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0) -#define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0) /** * Identify all possible varying inputs. The fragment program will @@ -313,9 +303,9 @@ static GLbitfield get_fp_input_mask( struct gl_context *ctx ) { /* _NEW_PROGRAM */ const GLboolean vertexShader = - (ctx->Shader.CurrentVertexProgram && - ctx->Shader.CurrentVertexProgram->LinkStatus && - ctx->Shader.CurrentVertexProgram->VertexProgram); + (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX] && + ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]->LinkStatus && + ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]->_LinkedShaders[MESA_SHADER_VERTEX]); const GLboolean vertexProgram = ctx->VertexProgram._Enabled; GLbitfield fp_inputs = 0x0; @@ -328,51 +318,50 @@ static GLbitfield get_fp_input_mask( struct gl_context *ctx ) } else if (ctx->RenderMode == GL_FEEDBACK) { /* _NEW_RENDERMODE */ - fp_inputs = (FRAG_BIT_COL0 | FRAG_BIT_TEX0); + fp_inputs = (VARYING_BIT_COL0 | VARYING_BIT_TEX0); } - else if (!(vertexProgram || vertexShader) || - !ctx->VertexProgram._Current) { + else if (!(vertexProgram || vertexShader)) { /* Fixed function vertex logic */ - /* _NEW_ARRAY */ - GLbitfield varying_inputs = ctx->varying_vp_inputs; + /* _NEW_VARYING_VP_INPUTS */ + GLbitfield64 varying_inputs = ctx->varying_vp_inputs; /* These get generated in the setup routine regardless of the * vertex program: */ /* _NEW_POINT */ if (ctx->Point.PointSprite) - varying_inputs |= FRAG_BITS_TEX_ANY; + varying_inputs |= VARYING_BITS_TEX_ANY; /* First look at what values may be computed by the generated * vertex program: */ /* _NEW_LIGHT */ if (ctx->Light.Enabled) { - fp_inputs |= FRAG_BIT_COL0; + fp_inputs |= VARYING_BIT_COL0; if (texenv_doing_secondary_color(ctx)) - fp_inputs |= FRAG_BIT_COL1; + fp_inputs |= VARYING_BIT_COL1; } /* _NEW_TEXTURE */ fp_inputs |= (ctx->Texture._TexGenEnabled | - ctx->Texture._TexMatEnabled) << FRAG_ATTRIB_TEX0; + ctx->Texture._TexMatEnabled) << VARYING_SLOT_TEX0; /* Then look at what might be varying as a result of enabled * arrays, etc: */ if (varying_inputs & VERT_BIT_COLOR0) - fp_inputs |= FRAG_BIT_COL0; + fp_inputs |= VARYING_BIT_COL0; if (varying_inputs & VERT_BIT_COLOR1) - fp_inputs |= FRAG_BIT_COL1; + fp_inputs |= VARYING_BIT_COL1; fp_inputs |= (((varying_inputs & VERT_BIT_TEX_ANY) >> VERT_ATTRIB_TEX0) - << FRAG_ATTRIB_TEX0); + << VARYING_SLOT_TEX0); } else { /* calculate from vp->outputs */ - struct gl_vertex_program *vprog; + struct gl_program *vprog; GLbitfield64 vp_outputs; /* Choose GLSL vertex shader over ARB vertex program. Need this @@ -380,26 +369,26 @@ static GLbitfield get_fp_input_mask( struct gl_context *ctx ) * validation (see additional comments in state.c). */ if (vertexShader) - vprog = ctx->Shader.CurrentVertexProgram->VertexProgram; + vprog = ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]->_LinkedShaders[MESA_SHADER_VERTEX]->Program; else - vprog = ctx->VertexProgram.Current; + vprog = &ctx->VertexProgram.Current->Base; - vp_outputs = vprog->Base.OutputsWritten; + vp_outputs = vprog->OutputsWritten; /* These get generated in the setup routine regardless of the * vertex program: */ /* _NEW_POINT */ if (ctx->Point.PointSprite) - vp_outputs |= FRAG_BITS_TEX_ANY; + vp_outputs |= VARYING_BITS_TEX_ANY; - if (vp_outputs & (1 << VERT_RESULT_COL0)) - fp_inputs |= FRAG_BIT_COL0; - if (vp_outputs & (1 << VERT_RESULT_COL1)) - fp_inputs |= FRAG_BIT_COL1; + if (vp_outputs & (1 << VARYING_SLOT_COL0)) + fp_inputs |= VARYING_BIT_COL0; + if (vp_outputs & (1 << VARYING_SLOT_COL1)) + fp_inputs |= VARYING_BIT_COL1; - fp_inputs |= (((vp_outputs & VERT_RESULT_TEX_ANY) >> VERT_RESULT_TEX0) - << FRAG_ATTRIB_TEX0); + fp_inputs |= (((vp_outputs & VARYING_BITS_TEX_ANY) >> VARYING_SLOT_TEX0) + << VARYING_SLOT_TEX0); } return fp_inputs; @@ -413,7 +402,7 @@ static GLbitfield get_fp_input_mask( struct gl_context *ctx ) static GLuint make_state_key( struct gl_context *ctx, struct state_key *key ) { GLuint i, j; - GLbitfield inputs_referenced = FRAG_BIT_COL0; + GLbitfield inputs_referenced = VARYING_BIT_COL0; const GLbitfield inputs_available = get_fp_input_mask( ctx ); GLuint keySize; @@ -424,24 +413,27 @@ static GLuint make_state_key( struct gl_context *ctx, struct state_key *key ) const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i]; const struct gl_texture_object *texObj = texUnit->_Current; const struct gl_tex_env_combine_state *comb = texUnit->_CurrentCombine; + const struct gl_sampler_object *samp; GLenum format; - if (!texUnit->_ReallyEnabled || !texUnit->Enabled) + if (!texUnit->_Current || !texUnit->Enabled) continue; + samp = _mesa_get_samplerobj(ctx, i); format = texObj->Image[0][texObj->BaseLevel]->_BaseFormat; key->unit[i].enabled = 1; key->enabled_units |= (1<nr_enabled_units = i + 1; - inputs_referenced |= FRAG_BIT_TEX(i); + inputs_referenced |= VARYING_BIT_TEX(i); - key->unit[i].source_index = - translate_tex_src_bit(texUnit->_ReallyEnabled); + key->unit[i].source_index = _mesa_tex_target_to_index(ctx, + texObj->Target); - key->unit[i].shadow = ((texObj->CompareMode == GL_COMPARE_R_TO_TEXTURE) && - ((format == GL_DEPTH_COMPONENT) || - (format == GL_DEPTH_STENCIL_EXT))); + key->unit[i].shadow = + ((samp->CompareMode == GL_COMPARE_R_TO_TEXTURE) && + ((format == GL_DEPTH_COMPONENT) || + (format == GL_DEPTH_STENCIL_EXT))); key->unit[i].NumArgsRGB = comb->_NumArgsRGB; key->unit[i].NumArgsA = comb->_NumArgsA; @@ -460,38 +452,30 @@ static GLuint make_state_key( struct gl_context *ctx, struct state_key *key ) key->unit[i].OptRGB[j].Source = translate_source(comb->SourceRGB[j]); key->unit[i].OptA[j].Source = translate_source(comb->SourceA[j]); } - - if (key->unit[i].ModeRGB == MODE_BUMP_ENVMAP_ATI) { - /* requires some special translation */ - key->unit[i].NumArgsRGB = 2; - key->unit[i].ScaleShiftRGB = 0; - key->unit[i].OptRGB[0].Operand = OPR_SRC_COLOR; - key->unit[i].OptRGB[0].Source = SRC_TEXTURE; - key->unit[i].OptRGB[1].Operand = OPR_SRC_COLOR; - key->unit[i].OptRGB[1].Source = texUnit->BumpTarget - GL_TEXTURE0 + SRC_TEXTURE0; - } - - /* this is a back-door for enabling cylindrical texture wrap mode */ - if (texObj->Priority == 0.125) - key->unit[i].texture_cyl_wrap = 1; } /* _NEW_LIGHT | _NEW_FOG */ if (texenv_doing_secondary_color(ctx)) { key->separate_specular = 1; - inputs_referenced |= FRAG_BIT_COL1; + inputs_referenced |= VARYING_BIT_COL1; } /* _NEW_FOG */ if (ctx->Fog.Enabled) { key->fog_enabled = 1; key->fog_mode = translate_fog_mode(ctx->Fog.Mode); - inputs_referenced |= FRAG_BIT_FOGC; /* maybe */ + inputs_referenced |= VARYING_BIT_FOGC; /* maybe */ } /* _NEW_BUFFERS */ key->num_draw_buffers = ctx->DrawBuffer->_NumColorDrawBuffers; + /* _NEW_COLOR */ + if (ctx->Color.AlphaEnabled && key->num_draw_buffers == 0) { + /* if alpha test is enabled we need to emit at least one color */ + key->num_draw_buffers = 1; + } + key->inputs_available = (inputs_available & inputs_referenced); /* compute size of state key, ignoring unused texture units */ @@ -502,426 +486,68 @@ static GLuint make_state_key( struct gl_context *ctx, struct state_key *key ) } -/** - * Use uregs to represent registers internally, translate to Mesa's - * expected formats on emit. - * - * NOTE: These are passed by value extensively in this file rather - * than as usual by pointer reference. If this disturbs you, try - * remembering they are just 32bits in size. - * - * GCC is smart enough to deal with these dword-sized structures in - * much the same way as if I had defined them as dwords and was using - * macros to access and set the fields. This is much nicer and easier - * to evolve. - */ -struct ureg { - GLuint file:4; - GLuint idx:8; - GLuint negatebase:1; - GLuint swz:12; - GLuint pad:7; -}; - -static const struct ureg undef = { - PROGRAM_UNDEFINED, - 255, - 0, - 0, - 0 -}; - - /** State used to build the fragment program: */ -struct texenv_fragment_program { - struct gl_fragment_program *program; +class texenv_fragment_program : public ir_factory { +public: + struct gl_shader_program *shader_program; + struct gl_shader *shader; + exec_list *top_instructions; struct state_key *state; - GLbitfield alu_temps; /**< Track texture indirections, see spec. */ - GLbitfield temps_output; /**< Track texture indirections, see spec. */ - GLbitfield temp_in_use; /**< Tracks temporary regs which are in use. */ - GLboolean error; - - struct ureg src_texture[MAX_TEXTURE_COORD_UNITS]; + ir_variable *src_texture[MAX_TEXTURE_COORD_UNITS]; /* Reg containing each texture unit's sampled texture color, * else undef. */ - struct ureg texcoord_tex[MAX_TEXTURE_COORD_UNITS]; + /* Texcoord override from bumpmapping. */ + ir_variable *texcoord_tex[MAX_TEXTURE_COORD_UNITS]; + /* Reg containing texcoord for a texture unit, * needed for bump mapping, else undef. */ - struct ureg src_previous; /**< Reg containing color from previous + ir_rvalue *src_previous; /**< Reg containing color from previous * stage. May need to be decl'd. */ - - GLuint last_tex_stage; /**< Number of last enabled texture unit */ - - struct ureg half; - struct ureg one; - struct ureg zero; }; - - -static struct ureg make_ureg(GLuint file, GLuint idx) -{ - struct ureg reg; - reg.file = file; - reg.idx = idx; - reg.negatebase = 0; - reg.swz = SWIZZLE_NOOP; - reg.pad = 0; - return reg; -} - -static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w ) -{ - reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x), - GET_SWZ(reg.swz, y), - GET_SWZ(reg.swz, z), - GET_SWZ(reg.swz, w)); - - return reg; -} - -static struct ureg swizzle1( struct ureg reg, int x ) -{ - return swizzle(reg, x, x, x, x); -} - -static struct ureg negate( struct ureg reg ) -{ - reg.negatebase ^= 1; - return reg; -} - -static GLboolean is_undef( struct ureg reg ) +static ir_rvalue * +get_current_attrib(texenv_fragment_program *p, GLuint attrib) { - return reg.file == PROGRAM_UNDEFINED; -} - - -static struct ureg get_temp( struct texenv_fragment_program *p ) -{ - GLint bit; - - /* First try and reuse temps which have been used already: - */ - bit = _mesa_ffs( ~p->temp_in_use & p->alu_temps ); - - /* Then any unused temporary: - */ - if (!bit) - bit = _mesa_ffs( ~p->temp_in_use ); - - if (!bit) { - _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__); - exit(1); - } - - if ((GLuint) bit > p->program->Base.NumTemporaries) - p->program->Base.NumTemporaries = bit; - - p->temp_in_use |= 1<<(bit-1); - return make_ureg(PROGRAM_TEMPORARY, (bit-1)); + ir_variable *current; + ir_rvalue *val; + + current = p->shader->symbols->get_variable("gl_CurrentAttribFragMESA"); + assert(current); + current->data.max_array_access = MAX2(current->data.max_array_access, attrib); + val = new(p->mem_ctx) ir_dereference_variable(current); + ir_rvalue *index = new(p->mem_ctx) ir_constant(attrib); + return new(p->mem_ctx) ir_dereference_array(val, index); } -static struct ureg get_tex_temp( struct texenv_fragment_program *p ) +static ir_rvalue * +get_gl_Color(texenv_fragment_program *p) { - int bit; - - /* First try to find available temp not previously used (to avoid - * starting a new texture indirection). According to the spec, the - * ~p->temps_output isn't necessary, but will keep it there for - * now: - */ - bit = _mesa_ffs( ~p->temp_in_use & ~p->alu_temps & ~p->temps_output ); - - /* Then any unused temporary: - */ - if (!bit) - bit = _mesa_ffs( ~p->temp_in_use ); - - if (!bit) { - _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__); - exit(1); + if (p->state->inputs_available & VARYING_BIT_COL0) { + ir_variable *var = p->shader->symbols->get_variable("gl_Color"); + assert(var); + return new(p->mem_ctx) ir_dereference_variable(var); + } else { + return get_current_attrib(p, VERT_ATTRIB_COLOR0); } - - if ((GLuint) bit > p->program->Base.NumTemporaries) - p->program->Base.NumTemporaries = bit; - - p->temp_in_use |= 1<<(bit-1); - return make_ureg(PROGRAM_TEMPORARY, (bit-1)); } - -/** Mark a temp reg as being no longer allocatable. */ -static void reserve_temp( struct texenv_fragment_program *p, struct ureg r ) +static ir_rvalue * +get_source(texenv_fragment_program *p, + GLuint src, GLuint unit) { - if (r.file == PROGRAM_TEMPORARY) - p->temps_output |= (1 << r.idx); -} - + ir_variable *var; + ir_dereference *deref; -static void release_temps(struct gl_context *ctx, struct texenv_fragment_program *p ) -{ - GLuint max_temp = ctx->Const.FragmentProgram.MaxTemps; - - /* KW: To support tex_env_crossbar, don't release the registers in - * temps_output. - */ - if (max_temp >= sizeof(int) * 8) - p->temp_in_use = p->temps_output; - else - p->temp_in_use = ~((1<temps_output; -} - - -static struct ureg register_param5( struct texenv_fragment_program *p, - GLint s0, - GLint s1, - GLint s2, - GLint s3, - GLint s4) -{ - int tokens[STATE_LENGTH]; - GLuint idx; - tokens[0] = s0; - tokens[1] = s1; - tokens[2] = s2; - tokens[3] = s3; - tokens[4] = s4; - idx = _mesa_add_state_reference(p->program->Base.Parameters, - (gl_state_index *)tokens); - return make_ureg(PROGRAM_STATE_VAR, idx); -} - - -#define register_param1(p,s0) register_param5(p,s0,0,0,0,0) -#define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0) -#define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0) -#define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0) - -static GLuint frag_to_vert_attrib( GLuint attrib ) -{ - switch (attrib) { - case FRAG_ATTRIB_COL0: return VERT_ATTRIB_COLOR0; - case FRAG_ATTRIB_COL1: return VERT_ATTRIB_COLOR1; - default: - assert(attrib >= FRAG_ATTRIB_TEX0); - assert(attrib <= FRAG_ATTRIB_TEX7); - return attrib - FRAG_ATTRIB_TEX0 + VERT_ATTRIB_TEX0; - } -} - - -static struct ureg register_input( struct texenv_fragment_program *p, GLuint input ) -{ - if (p->state->inputs_available & (1<program->Base.InputsRead |= (1 << input); - return make_ureg(PROGRAM_INPUT, input); - } - else { - GLuint idx = frag_to_vert_attrib( input ); - return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED, idx ); - } -} - - -static void emit_arg( struct prog_src_register *reg, - struct ureg ureg ) -{ - reg->File = ureg.file; - reg->Index = ureg.idx; - reg->Swizzle = ureg.swz; - reg->Negate = ureg.negatebase ? NEGATE_XYZW : NEGATE_NONE; - reg->Abs = GL_FALSE; -} - -static void emit_dst( struct prog_dst_register *dst, - struct ureg ureg, GLuint mask ) -{ - dst->File = ureg.file; - dst->Index = ureg.idx; - dst->WriteMask = mask; - dst->CondMask = COND_TR; /* always pass cond test */ - dst->CondSwizzle = SWIZZLE_NOOP; -} - -static struct prog_instruction * -emit_op(struct texenv_fragment_program *p, - enum prog_opcode op, - struct ureg dest, - GLuint mask, - GLboolean saturate, - struct ureg src0, - struct ureg src1, - struct ureg src2 ) -{ - const GLuint nr = p->program->Base.NumInstructions++; - struct prog_instruction *inst = &p->program->Base.Instructions[nr]; - - assert(nr < MAX_INSTRUCTIONS); - - _mesa_init_instructions(inst, 1); - inst->Opcode = op; - - emit_arg( &inst->SrcReg[0], src0 ); - emit_arg( &inst->SrcReg[1], src1 ); - emit_arg( &inst->SrcReg[2], src2 ); - - inst->SaturateMode = saturate ? SATURATE_ZERO_ONE : SATURATE_OFF; - - emit_dst( &inst->DstReg, dest, mask ); - -#if 0 - /* Accounting for indirection tracking: - */ - if (dest.file == PROGRAM_TEMPORARY) - p->temps_output |= 1 << dest.idx; -#endif - - return inst; -} - - -static struct ureg emit_arith( struct texenv_fragment_program *p, - enum prog_opcode op, - struct ureg dest, - GLuint mask, - GLboolean saturate, - struct ureg src0, - struct ureg src1, - struct ureg src2 ) -{ - emit_op(p, op, dest, mask, saturate, src0, src1, src2); - - /* Accounting for indirection tracking: - */ - if (src0.file == PROGRAM_TEMPORARY) - p->alu_temps |= 1 << src0.idx; - - if (!is_undef(src1) && src1.file == PROGRAM_TEMPORARY) - p->alu_temps |= 1 << src1.idx; - - if (!is_undef(src2) && src2.file == PROGRAM_TEMPORARY) - p->alu_temps |= 1 << src2.idx; - - if (dest.file == PROGRAM_TEMPORARY) - p->alu_temps |= 1 << dest.idx; - - p->program->Base.NumAluInstructions++; - return dest; -} - -static struct ureg emit_texld( struct texenv_fragment_program *p, - enum prog_opcode op, - struct ureg dest, - GLuint destmask, - GLuint tex_unit, - GLuint tex_idx, - GLuint tex_shadow, - struct ureg coord ) -{ - struct prog_instruction *inst = emit_op( p, op, - dest, destmask, - GL_FALSE, /* don't saturate? */ - coord, /* arg 0? */ - undef, - undef); - - inst->TexSrcTarget = tex_idx; - inst->TexSrcUnit = tex_unit; - inst->TexShadow = tex_shadow; - - p->program->Base.NumTexInstructions++; - - /* Accounting for indirection tracking: - */ - reserve_temp(p, dest); - -#if 0 - /* Is this a texture indirection? - */ - if ((coord.file == PROGRAM_TEMPORARY && - (p->temps_output & (1<alu_temps & (1<program->Base.NumTexIndirections++; - p->temps_output = 1<alu_temps = 0; - assert(0); /* KW: texture env crossbar */ - } -#endif - - return dest; -} - - -static struct ureg register_const4f( struct texenv_fragment_program *p, - GLfloat s0, - GLfloat s1, - GLfloat s2, - GLfloat s3) -{ - GLfloat values[4]; - GLuint idx, swizzle; - struct ureg r; - values[0] = s0; - values[1] = s1; - values[2] = s2; - values[3] = s3; - idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4, - &swizzle ); - r = make_ureg(PROGRAM_CONSTANT, idx); - r.swz = swizzle; - return r; -} - -#define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0) -#define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1) -#define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1) -#define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1) - - -static struct ureg get_one( struct texenv_fragment_program *p ) -{ - if (is_undef(p->one)) - p->one = register_scalar_const(p, 1.0); - return p->one; -} - -static struct ureg get_half( struct texenv_fragment_program *p ) -{ - if (is_undef(p->half)) - p->half = register_scalar_const(p, 0.5); - return p->half; -} - -static struct ureg get_zero( struct texenv_fragment_program *p ) -{ - if (is_undef(p->zero)) - p->zero = register_scalar_const(p, 0.0); - return p->zero; -} - - -static void program_error( struct texenv_fragment_program *p, const char *msg ) -{ - _mesa_problem(NULL, "%s", msg); - p->error = 1; -} - -static struct ureg get_source( struct texenv_fragment_program *p, - GLuint src, GLuint unit ) -{ switch (src) { case SRC_TEXTURE: - assert(!is_undef(p->src_texture[unit])); - return p->src_texture[unit]; + return new(p->mem_ctx) ir_dereference_variable(p->src_texture[unit]); case SRC_TEXTURE0: case SRC_TEXTURE1: @@ -931,66 +557,65 @@ static struct ureg get_source( struct texenv_fragment_program *p, case SRC_TEXTURE5: case SRC_TEXTURE6: case SRC_TEXTURE7: - assert(!is_undef(p->src_texture[src - SRC_TEXTURE0])); - return p->src_texture[src - SRC_TEXTURE0]; + return new(p->mem_ctx) + ir_dereference_variable(p->src_texture[src - SRC_TEXTURE0]); case SRC_CONSTANT: - return register_param2(p, STATE_TEXENV_COLOR, unit); + var = p->shader->symbols->get_variable("gl_TextureEnvColor"); + assert(var); + deref = new(p->mem_ctx) ir_dereference_variable(var); + var->data.max_array_access = MAX2(var->data.max_array_access, unit); + return new(p->mem_ctx) ir_dereference_array(deref, + new(p->mem_ctx) ir_constant(unit)); case SRC_PRIMARY_COLOR: - return register_input(p, FRAG_ATTRIB_COL0); + var = p->shader->symbols->get_variable("gl_Color"); + assert(var); + return new(p->mem_ctx) ir_dereference_variable(var); case SRC_ZERO: - return get_zero(p); + return new(p->mem_ctx) ir_constant(0.0f); case SRC_PREVIOUS: - if (is_undef(p->src_previous)) - return register_input(p, FRAG_ATTRIB_COL0); - else - return p->src_previous; + if (!p->src_previous) { + return get_gl_Color(p); + } else { + return p->src_previous->clone(p->mem_ctx, NULL); + } default: assert(0); - return undef; + return NULL; } } -static struct ureg emit_combine_source( struct texenv_fragment_program *p, - GLuint mask, - GLuint unit, - GLuint source, - GLuint operand ) +static ir_rvalue * +emit_combine_source(texenv_fragment_program *p, + GLuint unit, + GLuint source, + GLuint operand) { - struct ureg arg, src, one; + ir_rvalue *src; src = get_source(p, source, unit); switch (operand) { case OPR_ONE_MINUS_SRC_COLOR: - /* Get unused tmp, - * Emit tmp = 1.0 - arg.xyzw - */ - arg = get_temp( p ); - one = get_one( p ); - return emit_arith( p, OPCODE_SUB, arg, mask, 0, one, src, undef); + return sub(new(p->mem_ctx) ir_constant(1.0f), src); + + case OPR_SRC_ALPHA: + return src->type->is_scalar() ? src : swizzle_w(src); + + case OPR_ONE_MINUS_SRC_ALPHA: { + ir_rvalue *const scalar = src->type->is_scalar() ? src : swizzle_w(src); + + return sub(new(p->mem_ctx) ir_constant(1.0f), scalar); + } - case OPR_SRC_ALPHA: - if (mask == WRITEMASK_W) - return src; - else - return swizzle1( src, SWIZZLE_W ); - case OPR_ONE_MINUS_SRC_ALPHA: - /* Get unused tmp, - * Emit tmp = 1.0 - arg.wwww - */ - arg = get_temp(p); - one = get_one(p); - return emit_arith(p, OPCODE_SUB, arg, mask, 0, - one, swizzle1(src, SWIZZLE_W), undef); case OPR_ZERO: - return get_zero(p); + return new(p->mem_ctx) ir_constant(0.0f); case OPR_ONE: - return get_one(p); + return new(p->mem_ctx) ir_constant(1.0f); case OPR_SRC_COLOR: return src; default: @@ -1039,141 +664,101 @@ static GLboolean args_match( const struct state_key *key, GLuint unit ) return GL_TRUE; } -static struct ureg emit_combine( struct texenv_fragment_program *p, - struct ureg dest, - GLuint mask, - GLboolean saturate, - GLuint unit, - GLuint nr, - GLuint mode, - const struct mode_opt *opt) +static ir_rvalue * +smear(texenv_fragment_program *p, ir_rvalue *val) { - struct ureg src[MAX_COMBINER_TERMS]; - struct ureg tmp, half; + if (!val->type->is_scalar()) + return val; + + return swizzle_xxxx(val); +} + +static ir_rvalue * +emit_combine(texenv_fragment_program *p, + GLuint unit, + GLuint nr, + GLuint mode, + const struct mode_opt *opt) +{ + ir_rvalue *src[MAX_COMBINER_TERMS]; + ir_rvalue *tmp0, *tmp1; GLuint i; assert(nr <= MAX_COMBINER_TERMS); for (i = 0; i < nr; i++) - src[i] = emit_combine_source( p, mask, unit, opt[i].Source, opt[i].Operand ); + src[i] = emit_combine_source( p, unit, opt[i].Source, opt[i].Operand ); switch (mode) { case MODE_REPLACE: - if (mask == WRITEMASK_XYZW && !saturate) - return src[0]; - else - return emit_arith( p, OPCODE_MOV, dest, mask, saturate, src[0], undef, undef ); + return src[0]; + case MODE_MODULATE: - return emit_arith( p, OPCODE_MUL, dest, mask, saturate, - src[0], src[1], undef ); + return mul(src[0], src[1]); + case MODE_ADD: - return emit_arith( p, OPCODE_ADD, dest, mask, saturate, - src[0], src[1], undef ); + return add(src[0], src[1]); + case MODE_ADD_SIGNED: - /* tmp = arg0 + arg1 - * result = tmp - .5 - */ - half = get_half(p); - tmp = get_temp( p ); - emit_arith( p, OPCODE_ADD, tmp, mask, 0, src[0], src[1], undef ); - emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp, half, undef ); - return dest; + return add(add(src[0], src[1]), new(p->mem_ctx) ir_constant(-0.5f)); + case MODE_INTERPOLATE: - /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered: - */ - return emit_arith( p, OPCODE_LRP, dest, mask, saturate, src[2], src[0], src[1] ); + /* Arg0 * (Arg2) + Arg1 * (1-Arg2) */ + tmp0 = mul(src[0], src[2]); + tmp1 = mul(src[1], sub(new(p->mem_ctx) ir_constant(1.0f), + src[2]->clone(p->mem_ctx, NULL))); + return add(tmp0, tmp1); case MODE_SUBTRACT: - return emit_arith( p, OPCODE_SUB, dest, mask, saturate, src[0], src[1], undef ); + return sub(src[0], src[1]); case MODE_DOT3_RGBA: case MODE_DOT3_RGBA_EXT: case MODE_DOT3_RGB_EXT: case MODE_DOT3_RGB: { - struct ureg tmp0 = get_temp( p ); - struct ureg tmp1 = get_temp( p ); - struct ureg neg1 = register_scalar_const(p, -1); - struct ureg two = register_scalar_const(p, 2); + tmp0 = mul(src[0], new(p->mem_ctx) ir_constant(2.0f)); + tmp0 = add(tmp0, new(p->mem_ctx) ir_constant(-1.0f)); - /* tmp0 = 2*src0 - 1 - * tmp1 = 2*src1 - 1 - * - * dst = tmp0 dot3 tmp1 - */ - emit_arith( p, OPCODE_MAD, tmp0, WRITEMASK_XYZW, 0, - two, src[0], neg1); + tmp1 = mul(src[1], new(p->mem_ctx) ir_constant(2.0f)); + tmp1 = add(tmp1, new(p->mem_ctx) ir_constant(-1.0f)); - if (memcmp(&src[0], &src[1], sizeof(struct ureg)) == 0) - tmp1 = tmp0; - else - emit_arith( p, OPCODE_MAD, tmp1, WRITEMASK_XYZW, 0, - two, src[1], neg1); - emit_arith( p, OPCODE_DP3, dest, mask, saturate, tmp0, tmp1, undef); - return dest; + return dot(swizzle_xyz(smear(p, tmp0)), swizzle_xyz(smear(p, tmp1))); } case MODE_MODULATE_ADD_ATI: - /* Arg0 * Arg2 + Arg1 */ - return emit_arith( p, OPCODE_MAD, dest, mask, saturate, - src[0], src[2], src[1] ); - case MODE_MODULATE_SIGNED_ADD_ATI: { - /* Arg0 * Arg2 + Arg1 - 0.5 */ - struct ureg tmp0 = get_temp(p); - half = get_half(p); - emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[0], src[2], src[1] ); - emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef ); - return dest; - } + return add(mul(src[0], src[2]), src[1]); + + case MODE_MODULATE_SIGNED_ADD_ATI: + return add(add(mul(src[0], src[2]), src[1]), + new(p->mem_ctx) ir_constant(-0.5f)); + case MODE_MODULATE_SUBTRACT_ATI: - /* Arg0 * Arg2 - Arg1 */ - emit_arith( p, OPCODE_MAD, dest, mask, 0, src[0], src[2], negate(src[1]) ); - return dest; + return sub(mul(src[0], src[2]), src[1]); + case MODE_ADD_PRODUCTS: - /* Arg0 * Arg1 + Arg2 * Arg3 */ - { - struct ureg tmp0 = get_temp(p); - emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef ); - emit_arith( p, OPCODE_MAD, dest, mask, saturate, src[2], src[3], tmp0 ); - } - return dest; + return add(mul(src[0], src[1]), mul(src[2], src[3])); + case MODE_ADD_PRODUCTS_SIGNED: - /* Arg0 * Arg1 + Arg2 * Arg3 - 0.5 */ - { - struct ureg tmp0 = get_temp(p); - half = get_half(p); - emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef ); - emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[2], src[3], tmp0 ); - emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef ); - } - return dest; - case MODE_BUMP_ENVMAP_ATI: - /* special - not handled here */ - assert(0); - return src[0]; + return add(add(mul(src[0], src[1]), mul(src[2], src[3])), + new(p->mem_ctx) ir_constant(-0.5f)); default: assert(0); return src[0]; } } - /** * Generate instructions for one texture unit's env/combiner mode. */ -static struct ureg -emit_texenv(struct texenv_fragment_program *p, GLuint unit) +static ir_rvalue * +emit_texenv(texenv_fragment_program *p, GLuint unit) { const struct state_key *key = p->state; GLboolean rgb_saturate, alpha_saturate; GLuint rgb_shift, alpha_shift; - struct ureg out, dest; if (!key->unit[unit].enabled) { return get_source(p, SRC_PREVIOUS, 0); } - if (key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) { - /* this isn't really a env stage delivering a color and handled elsewhere */ - return get_source(p, SRC_PREVIOUS, 0); - } switch (key->unit[unit].ModeRGB) { case MODE_DOT3_RGB_EXT: @@ -1207,129 +792,222 @@ emit_texenv(struct texenv_fragment_program *p, GLuint unit) else alpha_saturate = GL_FALSE; - /* If this is the very last calculation (and various other conditions - * are met), emit directly to the color output register. Otherwise, - * emit to a temporary register. - */ - if (key->separate_specular || - unit != p->last_tex_stage || - alpha_shift || - key->num_draw_buffers != 1 || - rgb_shift) - dest = get_temp( p ); - else - dest = make_ureg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR); + ir_variable *temp_var = p->make_temp(glsl_type::vec4_type, "texenv_combine"); + ir_dereference *deref; + ir_rvalue *val; /* Emit the RGB and A combine ops */ if (key->unit[unit].ModeRGB == key->unit[unit].ModeA && args_match(key, unit)) { - out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate, - unit, - key->unit[unit].NumArgsRGB, - key->unit[unit].ModeRGB, - key->unit[unit].OptRGB); + val = emit_combine(p, unit, + key->unit[unit].NumArgsRGB, + key->unit[unit].ModeRGB, + key->unit[unit].OptRGB); + val = smear(p, val); + if (rgb_saturate) + val = saturate(val); + + p->emit(assign(temp_var, val)); } else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT || key->unit[unit].ModeRGB == MODE_DOT3_RGBA) { - out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate, - unit, - key->unit[unit].NumArgsRGB, - key->unit[unit].ModeRGB, - key->unit[unit].OptRGB); + ir_rvalue *val = emit_combine(p, unit, + key->unit[unit].NumArgsRGB, + key->unit[unit].ModeRGB, + key->unit[unit].OptRGB); + val = smear(p, val); + if (rgb_saturate) + val = saturate(val); + p->emit(assign(temp_var, val)); } else { /* Need to do something to stop from re-emitting identical * argument calculations here: */ - out = emit_combine( p, dest, WRITEMASK_XYZ, rgb_saturate, - unit, - key->unit[unit].NumArgsRGB, - key->unit[unit].ModeRGB, - key->unit[unit].OptRGB); - out = emit_combine( p, dest, WRITEMASK_W, alpha_saturate, - unit, - key->unit[unit].NumArgsA, - key->unit[unit].ModeA, - key->unit[unit].OptA); + val = emit_combine(p, unit, + key->unit[unit].NumArgsRGB, + key->unit[unit].ModeRGB, + key->unit[unit].OptRGB); + val = swizzle_xyz(smear(p, val)); + if (rgb_saturate) + val = saturate(val); + p->emit(assign(temp_var, val, WRITEMASK_XYZ)); + + val = emit_combine(p, unit, + key->unit[unit].NumArgsA, + key->unit[unit].ModeA, + key->unit[unit].OptA); + val = swizzle_w(smear(p, val)); + if (alpha_saturate) + val = saturate(val); + p->emit(assign(temp_var, val, WRITEMASK_W)); } + deref = new(p->mem_ctx) ir_dereference_variable(temp_var); + /* Deal with the final shift: */ if (alpha_shift || rgb_shift) { - struct ureg shift; - GLboolean saturate = GL_TRUE; /* always saturate at this point */ + ir_constant *shift; if (rgb_shift == alpha_shift) { - shift = register_scalar_const(p, (GLfloat)(1<mem_ctx) ir_constant((float)(1 << rgb_shift)); } else { - shift = register_const4f(p, - (GLfloat)(1<mem_ctx) ir_constant(glsl_type::vec4_type, + &const_data); } - return emit_arith( p, OPCODE_MUL, dest, WRITEMASK_XYZW, - saturate, out, shift, undef ); + + return saturate(mul(deref, shift)); } else - return out; + return deref; } /** * Generate instruction for getting a texture source term. */ -static void load_texture( struct texenv_fragment_program *p, GLuint unit ) +static void load_texture( texenv_fragment_program *p, GLuint unit ) { - if (is_undef(p->src_texture[unit])) { - const GLuint texTarget = p->state->unit[unit].source_index; - struct ureg texcoord; - struct ureg tmp = get_tex_temp( p ); + ir_dereference *deref; - if (is_undef(p->texcoord_tex[unit])) { - texcoord = register_input(p, FRAG_ATTRIB_TEX0+unit); - } - else { - /* might want to reuse this reg for tex output actually */ - texcoord = p->texcoord_tex[unit]; - } + if (p->src_texture[unit]) + return; - /* TODO: Use D0_MASK_XY where possible. - */ - if (p->state->unit[unit].enabled) { - GLboolean shadow = GL_FALSE; - - if (p->state->unit[unit].shadow) { - p->program->Base.ShadowSamplers |= 1 << unit; - shadow = GL_TRUE; - } - - p->src_texture[unit] = emit_texld( p, OPCODE_TXP, - tmp, WRITEMASK_XYZW, - unit, texTarget, shadow, - texcoord ); - - p->program->Base.SamplersUsed |= (1 << unit); - /* This identity mapping should already be in place - * (see _mesa_init_program_struct()) but let's be safe. - */ - p->program->Base.SamplerUnits[unit] = unit; - } + const GLuint texTarget = p->state->unit[unit].source_index; + ir_rvalue *texcoord; + + if (!(p->state->inputs_available & (VARYING_BIT_TEX0 << unit))) { + texcoord = get_current_attrib(p, VERT_ATTRIB_TEX0 + unit); + } else if (p->texcoord_tex[unit]) { + texcoord = new(p->mem_ctx) ir_dereference_variable(p->texcoord_tex[unit]); + } else { + ir_variable *tc_array = p->shader->symbols->get_variable("gl_TexCoord"); + assert(tc_array); + texcoord = new(p->mem_ctx) ir_dereference_variable(tc_array); + ir_rvalue *index = new(p->mem_ctx) ir_constant(unit); + texcoord = new(p->mem_ctx) ir_dereference_array(texcoord, index); + tc_array->data.max_array_access = MAX2(tc_array->data.max_array_access, unit); + } + + if (!p->state->unit[unit].enabled) { + p->src_texture[unit] = p->make_temp(glsl_type::vec4_type, + "dummy_tex"); + p->emit(p->src_texture[unit]); + + p->emit(assign(p->src_texture[unit], new(p->mem_ctx) ir_constant(0.0f))); + return ; + } + + const glsl_type *sampler_type = NULL; + int coords = 0; + + switch (texTarget) { + case TEXTURE_1D_INDEX: + if (p->state->unit[unit].shadow) + sampler_type = glsl_type::sampler1DShadow_type; + else + sampler_type = glsl_type::sampler1D_type; + coords = 1; + break; + case TEXTURE_1D_ARRAY_INDEX: + if (p->state->unit[unit].shadow) + sampler_type = glsl_type::sampler1DArrayShadow_type; else - p->src_texture[unit] = get_zero(p); + sampler_type = glsl_type::sampler1DArray_type; + coords = 2; + break; + case TEXTURE_2D_INDEX: + if (p->state->unit[unit].shadow) + sampler_type = glsl_type::sampler2DShadow_type; + else + sampler_type = glsl_type::sampler2D_type; + coords = 2; + break; + case TEXTURE_2D_ARRAY_INDEX: + if (p->state->unit[unit].shadow) + sampler_type = glsl_type::sampler2DArrayShadow_type; + else + sampler_type = glsl_type::sampler2DArray_type; + coords = 3; + break; + case TEXTURE_RECT_INDEX: + if (p->state->unit[unit].shadow) + sampler_type = glsl_type::sampler2DRectShadow_type; + else + sampler_type = glsl_type::sampler2DRect_type; + coords = 2; + break; + case TEXTURE_3D_INDEX: + assert(!p->state->unit[unit].shadow); + sampler_type = glsl_type::sampler3D_type; + coords = 3; + break; + case TEXTURE_CUBE_INDEX: + if (p->state->unit[unit].shadow) + sampler_type = glsl_type::samplerCubeShadow_type; + else + sampler_type = glsl_type::samplerCube_type; + coords = 3; + break; + case TEXTURE_EXTERNAL_INDEX: + assert(!p->state->unit[unit].shadow); + sampler_type = glsl_type::samplerExternalOES_type; + coords = 2; + break; + } - if (p->state->unit[unit].texture_cyl_wrap) { - /* set flag which is checked by Mesa->Gallium program translation */ - p->program->Base.InputFlags[0] |= PROG_PARAM_BIT_CYL_WRAP; - } + p->src_texture[unit] = p->make_temp(glsl_type::vec4_type, + "tex"); + + ir_texture *tex = new(p->mem_ctx) ir_texture(ir_tex); + + + char *sampler_name = ralloc_asprintf(p->mem_ctx, "sampler_%d", unit); + ir_variable *sampler = new(p->mem_ctx) ir_variable(sampler_type, + sampler_name, + ir_var_uniform); + p->top_instructions->push_head(sampler); + + /* Set the texture unit for this sampler. The linker will pick this value + * up and do-the-right-thing. + * + * NOTE: The cast to int is important. Without it, the constant will have + * type uint, and things later on may get confused. + */ + sampler->constant_value = new(p->mem_ctx) ir_constant(int(unit)); + + deref = new(p->mem_ctx) ir_dereference_variable(sampler); + tex->set_sampler(deref, glsl_type::vec4_type); + tex->coordinate = new(p->mem_ctx) ir_swizzle(texcoord, 0, 1, 2, 3, coords); + + if (p->state->unit[unit].shadow) { + texcoord = texcoord->clone(p->mem_ctx, NULL); + tex->shadow_comparitor = new(p->mem_ctx) ir_swizzle(texcoord, + coords, 0, 0, 0, + 1); + coords++; } + + texcoord = texcoord->clone(p->mem_ctx, NULL); + tex->projector = swizzle_w(texcoord); + + p->emit(assign(p->src_texture[unit], tex)); } -static GLboolean load_texenv_source( struct texenv_fragment_program *p, - GLuint src, GLuint unit ) +static void +load_texenv_source(texenv_fragment_program *p, + GLuint src, GLuint unit) { switch (src) { case SRC_TEXTURE: @@ -1351,8 +1029,6 @@ static GLboolean load_texenv_source( struct texenv_fragment_program *p, /* not a texture src - do nothing */ break; } - - return GL_TRUE; } @@ -1360,7 +1036,7 @@ static GLboolean load_texenv_source( struct texenv_fragment_program *p, * Generate instructions for loading all texture source terms. */ static GLboolean -load_texunit_sources( struct texenv_fragment_program *p, GLuint unit ) +load_texunit_sources( texenv_fragment_program *p, GLuint unit ) { const struct state_key *key = p->state; GLuint i; @@ -1377,215 +1053,228 @@ load_texunit_sources( struct texenv_fragment_program *p, GLuint unit ) } /** - * Generate instructions for loading bump map textures. + * Applies the fog calculations. + * + * This is basically like the ARB_fragment_prorgam fog options. Note + * that ffvertex_prog.c produces fogcoord for us when + * GL_FOG_COORDINATE_EXT is set to GL_FRAGMENT_DEPTH_EXT. */ -static GLboolean -load_texunit_bumpmap( struct texenv_fragment_program *p, GLuint unit ) +static ir_rvalue * +emit_fog_instructions(texenv_fragment_program *p, + ir_rvalue *fragcolor) { - const struct state_key *key = p->state; - GLuint bumpedUnitNr = key->unit[unit].OptRGB[1].Source - SRC_TEXTURE0; - struct ureg texcDst, bumpMapRes; - struct ureg constdudvcolor = register_const4f(p, 0.0, 0.0, 0.0, 1.0); - struct ureg texcSrc = register_input(p, FRAG_ATTRIB_TEX0 + bumpedUnitNr); - struct ureg rotMat0 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_0, unit ); - struct ureg rotMat1 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_1, unit ); - - load_texenv_source( p, unit + SRC_TEXTURE0, unit ); - - bumpMapRes = get_source(p, key->unit[unit].OptRGB[0].Source, unit); - texcDst = get_tex_temp( p ); - p->texcoord_tex[bumpedUnitNr] = texcDst; - - /* Apply rot matrix and add coords to be available in next phase. - * dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1) - * note only 2 coords are affected the rest are left unchanged (mul by 0) + struct state_key *key = p->state; + ir_rvalue *f, *temp; + ir_variable *params, *oparams; + ir_variable *fogcoord; + + /* Temporary storage for the whole fog result. Fog calculations + * only affect rgb so we're hanging on to the .a value of fragcolor + * this way. */ - emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0, - swizzle1(bumpMapRes, SWIZZLE_X), rotMat0, texcSrc ); - emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0, - swizzle1(bumpMapRes, SWIZZLE_Y), rotMat1, texcDst ); + ir_variable *fog_result = p->make_temp(glsl_type::vec4_type, "fog_result"); + p->emit(assign(fog_result, fragcolor)); - /* Move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish - * enough to access this later, should optimize away. - */ - emit_arith( p, OPCODE_MOV, bumpMapRes, WRITEMASK_XYZW, 0, - constdudvcolor, undef, undef ); + fragcolor = swizzle_xyz(fog_result); - return GL_TRUE; -} + oparams = p->shader->symbols->get_variable("gl_FogParamsOptimizedMESA"); + assert(oparams); + fogcoord = p->shader->symbols->get_variable("gl_FogFragCoord"); + assert(fogcoord); + params = p->shader->symbols->get_variable("gl_Fog"); + assert(params); + f = new(p->mem_ctx) ir_dereference_variable(fogcoord); -/** - * Generate a new fragment program which implements the context's - * current texture env/combine mode. - */ -static void -create_new_program(struct gl_context *ctx, struct state_key *key, - struct gl_fragment_program *program) -{ - struct prog_instruction instBuffer[MAX_INSTRUCTIONS]; - struct texenv_fragment_program p; - GLuint unit; - struct ureg cf, out; - int i; + ir_variable *f_var = p->make_temp(glsl_type::float_type, "fog_factor"); - memset(&p, 0, sizeof(p)); - p.state = key; - p.program = program; + switch (key->fog_mode) { + case FOG_LINEAR: + /* f = (end - z) / (end - start) + * + * gl_MesaFogParamsOptimized gives us (-1 / (end - start)) and + * (end / (end - start)) so we can generate a single MAD. + */ + f = add(mul(f, swizzle_x(oparams)), swizzle_y(oparams)); + break; + case FOG_EXP: + /* f = e^(-(density * fogcoord)) + * + * gl_MesaFogParamsOptimized gives us density/ln(2) so we can + * use EXP2 which is generally the native instruction without + * having to do any further math on the fog density uniform. + */ + f = mul(f, swizzle_z(oparams)); + f = new(p->mem_ctx) ir_expression(ir_unop_neg, f); + f = new(p->mem_ctx) ir_expression(ir_unop_exp2, f); + break; + case FOG_EXP2: + /* f = e^(-(density * fogcoord)^2) + * + * gl_MesaFogParamsOptimized gives us density/sqrt(ln(2)) so we + * can do this like FOG_EXP but with a squaring after the + * multiply by density. + */ + ir_variable *temp_var = p->make_temp(glsl_type::float_type, "fog_temp"); + p->emit(assign(temp_var, mul(f, swizzle_w(oparams)))); - /* During code generation, use locally-allocated instruction buffer, - * then alloc dynamic storage below. - */ - p.program->Base.Instructions = instBuffer; - p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB; - p.program->Base.String = NULL; - p.program->Base.NumTexIndirections = 1; /* is this right? */ - p.program->Base.NumTexInstructions = 0; - p.program->Base.NumAluInstructions = 0; - p.program->Base.NumInstructions = 0; - p.program->Base.NumTemporaries = 0; - p.program->Base.NumParameters = 0; - p.program->Base.NumAttributes = 0; - p.program->Base.NumAddressRegs = 0; - p.program->Base.Parameters = _mesa_new_parameter_list(); - p.program->Base.InputsRead = 0x0; - - if (key->num_draw_buffers == 1) - p.program->Base.OutputsWritten = 1 << FRAG_RESULT_COLOR; - else { - for (i = 0; i < key->num_draw_buffers; i++) - p.program->Base.OutputsWritten |= (1 << (FRAG_RESULT_DATA0 + i)); + f = mul(temp_var, temp_var); + f = new(p->mem_ctx) ir_expression(ir_unop_neg, f); + f = new(p->mem_ctx) ir_expression(ir_unop_exp2, f); + break; } - for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) { - p.src_texture[unit] = undef; - p.texcoord_tex[unit] = undef; - } + p->emit(assign(f_var, saturate(f))); - p.src_previous = undef; - p.half = undef; - p.zero = undef; - p.one = undef; + f = sub(new(p->mem_ctx) ir_constant(1.0f), f_var); + temp = new(p->mem_ctx) ir_dereference_variable(params); + temp = new(p->mem_ctx) ir_dereference_record(temp, "color"); + temp = mul(swizzle_xyz(temp), f); - p.last_tex_stage = 0; - release_temps(ctx, &p); + p->emit(assign(fog_result, add(temp, mul(fragcolor, f_var)), WRITEMASK_XYZ)); - if (key->enabled_units && key->num_draw_buffers) { - GLboolean needbumpstage = GL_FALSE; + return new(p->mem_ctx) ir_dereference_variable(fog_result); +} - /* Zeroth pass - bump map textures first */ - for (unit = 0; unit < key->nr_enabled_units; unit++) - if (key->unit[unit].enabled && - key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) { - needbumpstage = GL_TRUE; - load_texunit_bumpmap( &p, unit ); - } - if (needbumpstage) - p.program->Base.NumTexIndirections++; +static void +emit_instructions(texenv_fragment_program *p) +{ + struct state_key *key = p->state; + GLuint unit; + if (key->enabled_units) { /* First pass - to support texture_env_crossbar, first identify * all referenced texture sources and emit texld instructions * for each: */ for (unit = 0; unit < key->nr_enabled_units; unit++) if (key->unit[unit].enabled) { - load_texunit_sources( &p, unit ); - p.last_tex_stage = unit; + load_texunit_sources(p, unit); } /* Second pass - emit combine instructions to build final color: */ - for (unit = 0; unit < key->nr_enabled_units; unit++) + for (unit = 0; unit < key->nr_enabled_units; unit++) { if (key->unit[unit].enabled) { - p.src_previous = emit_texenv( &p, unit ); - reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */ - release_temps(ctx, &p); /* release all temps */ + p->src_previous = emit_texenv(p, unit); } + } } - cf = get_source( &p, SRC_PREVIOUS, 0 ); - - for (i = 0; i < key->num_draw_buffers; i++) { - if (key->num_draw_buffers == 1) - out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_COLOR ); - else { - out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_DATA0 + i ); + ir_rvalue *cf = get_source(p, SRC_PREVIOUS, 0); + + if (key->separate_specular) { + ir_variable *spec_result = p->make_temp(glsl_type::vec4_type, + "specular_add"); + p->emit(assign(spec_result, cf)); + + ir_rvalue *secondary; + if (p->state->inputs_available & VARYING_BIT_COL1) { + ir_variable *var = + p->shader->symbols->get_variable("gl_SecondaryColor"); + assert(var); + secondary = swizzle_xyz(var); + } else { + secondary = swizzle_xyz(get_current_attrib(p, VERT_ATTRIB_COLOR1)); } - if (key->separate_specular) { - /* Emit specular add. - */ - struct ureg s = register_input(&p, FRAG_ATTRIB_COL1); - emit_arith( &p, OPCODE_ADD, out, WRITEMASK_XYZ, 0, cf, s, undef ); - emit_arith( &p, OPCODE_MOV, out, WRITEMASK_W, 0, cf, undef, undef ); - } - else if (memcmp(&cf, &out, sizeof(cf)) != 0) { - /* Will wind up in here if no texture enabled or a couple of - * other scenarios (GL_REPLACE for instance). - */ - emit_arith( &p, OPCODE_MOV, out, WRITEMASK_XYZW, 0, cf, undef, undef ); - } + p->emit(assign(spec_result, add(swizzle_xyz(spec_result), secondary), + WRITEMASK_XYZ)); + + cf = new(p->mem_ctx) ir_dereference_variable(spec_result); } - /* Finish up: - */ - emit_arith( &p, OPCODE_END, undef, WRITEMASK_XYZW, 0, undef, undef, undef); if (key->fog_enabled) { - /* Pull fog mode from struct gl_context, the value in the state key is - * a reduced value and not what is expected in FogOption - */ - p.program->FogOption = ctx->Fog.Mode; - p.program->Base.InputsRead |= FRAG_BIT_FOGC; - } - else { - p.program->FogOption = GL_NONE; + cf = emit_fog_instructions(p, cf); } - if (p.program->Base.NumTexIndirections > ctx->Const.FragmentProgram.MaxTexIndirections) - program_error(&p, "Exceeded max nr indirect texture lookups"); + ir_variable *frag_color = p->shader->symbols->get_variable("gl_FragColor"); + assert(frag_color); + p->emit(assign(frag_color, cf)); +} - if (p.program->Base.NumTexInstructions > ctx->Const.FragmentProgram.MaxTexInstructions) - program_error(&p, "Exceeded max TEX instructions"); +/** + * Generate a new fragment program which implements the context's + * current texture env/combine mode. + */ +static struct gl_shader_program * +create_new_program(struct gl_context *ctx, struct state_key *key) +{ + texenv_fragment_program p; + unsigned int unit; + _mesa_glsl_parse_state *state; + + p.mem_ctx = ralloc_context(NULL); + p.shader = ctx->Driver.NewShader(ctx, 0, GL_FRAGMENT_SHADER); + p.shader->ir = new(p.shader) exec_list; + state = new(p.shader) _mesa_glsl_parse_state(ctx, MESA_SHADER_FRAGMENT, + p.shader); + p.shader->symbols = state->symbols; + p.top_instructions = p.shader->ir; + p.instructions = p.shader->ir; + p.state = key; + p.shader_program = ctx->Driver.NewShaderProgram(ctx, 0); - if (p.program->Base.NumAluInstructions > ctx->Const.FragmentProgram.MaxAluInstructions) - program_error(&p, "Exceeded max ALU instructions"); + /* Tell the linker to ignore the fact that we're building a + * separate shader, in case we're in a GLES2 context that would + * normally reject that. The real problem is that we're building a + * fixed function program in a GLES2 context at all, but that's a + * big mess to clean up. + */ + p.shader_program->SeparateShader = GL_TRUE; - ASSERT(p.program->Base.NumInstructions <= MAX_INSTRUCTIONS); + state->language_version = 130; + state->es_shader = false; + if (_mesa_is_gles(ctx) && ctx->Extensions.OES_EGL_image_external) + state->OES_EGL_image_external_enable = true; + _mesa_glsl_initialize_types(state); + _mesa_glsl_initialize_variables(p.instructions, state); - /* Allocate final instruction array */ - p.program->Base.Instructions - = _mesa_alloc_instructions(p.program->Base.NumInstructions); - if (!p.program->Base.Instructions) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, - "generating tex env program"); - return; + for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) { + p.src_texture[unit] = NULL; + p.texcoord_tex[unit] = NULL; } - _mesa_copy_instructions(p.program->Base.Instructions, instBuffer, - p.program->Base.NumInstructions); - if (key->num_draw_buffers && p.program->FogOption) { - _mesa_append_fog_code(ctx, p.program, GL_FALSE); - p.program->FogOption = GL_NONE; - } + p.src_previous = NULL; + ir_function *main_f = new(p.mem_ctx) ir_function("main"); + p.emit(main_f); + state->symbols->add_function(main_f); - /* Notify driver the fragment program has (actually) changed. - */ - if (ctx->Driver.ProgramStringNotify) { - GLboolean ok = ctx->Driver.ProgramStringNotify(ctx, - GL_FRAGMENT_PROGRAM_ARB, - &p.program->Base); - /* Driver should be able to handle any texenv programs as long as - * the driver correctly reported max number of texture units correctly, - * etc. - */ - ASSERT(ok); - (void) ok; /* silence unused var warning */ - } + ir_function_signature *main_sig = + new(p.mem_ctx) ir_function_signature(glsl_type::void_type); + main_sig->is_defined = true; + main_f->add_signature(main_sig); - if (DISASSEM) { - _mesa_print_program(&p.program->Base); - printf("\n"); - } + p.instructions = &main_sig->body; + if (key->num_draw_buffers) + emit_instructions(&p); + + validate_ir_tree(p.shader->ir); + + const struct gl_shader_compiler_options *options = + &ctx->Const.ShaderCompilerOptions[MESA_SHADER_FRAGMENT]; + + while (do_common_optimization(p.shader->ir, false, false, options, + ctx->Const.NativeIntegers)) + ; + reparent_ir(p.shader->ir, p.shader->ir); + + p.shader->CompileStatus = true; + p.shader->Version = state->language_version; + p.shader->uses_builtin_functions = state->uses_builtin_functions; + p.shader_program->Shaders = + (gl_shader **)malloc(sizeof(*p.shader_program->Shaders)); + p.shader_program->Shaders[0] = p.shader; + p.shader_program->NumShaders = 1; + + _mesa_glsl_link_shader(ctx, p.shader_program); + + if (!p.shader_program->LinkStatus) + _mesa_problem(ctx, "Failed to link fixed function fragment shader: %s\n", + p.shader_program->InfoLog); + + ralloc_free(p.mem_ctx); + return p.shader_program; } extern "C" { @@ -1594,30 +1283,27 @@ extern "C" { * Return a fragment program which implements the current * fixed-function texture, fog and color-sum operations. */ -struct gl_fragment_program * +struct gl_shader_program * _mesa_get_fixed_func_fragment_program(struct gl_context *ctx) { - struct gl_fragment_program *prog; + struct gl_shader_program *shader_program; struct state_key key; GLuint keySize; - + keySize = make_state_key(ctx, &key); - - prog = (struct gl_fragment_program *) + + shader_program = (struct gl_shader_program *) _mesa_search_program_cache(ctx->FragmentProgram.Cache, &key, keySize); - if (!prog) { - prog = (struct gl_fragment_program *) - ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0); - - create_new_program(ctx, &key, prog); + if (!shader_program) { + shader_program = create_new_program(ctx, &key); - _mesa_program_cache_insert(ctx, ctx->FragmentProgram.Cache, - &key, keySize, &prog->Base); + _mesa_shader_cache_insert(ctx, ctx->FragmentProgram.Cache, + &key, keySize, shader_program); } - return prog; + return shader_program; } }