X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Ftnl%2Ft_vb_program.c;h=c35eaf1538e637d66c965644d5e69003d12f5ccb;hb=882cd6c839e56a3eceb8edf62f83893f6b531d35;hp=f11ac616f16f7cb9005201d1b903b4d802c4f79b;hpb=a23e668d40b10236fcf5ca83fdbda93ab36fa86f;p=mesa.git diff --git a/src/mesa/tnl/t_vb_program.c b/src/mesa/tnl/t_vb_program.c index f11ac616f16..c35eaf1538e 100644 --- a/src/mesa/tnl/t_vb_program.c +++ b/src/mesa/tnl/t_vb_program.c @@ -1,8 +1,8 @@ /* * Mesa 3-D graphics library - * Version: 6.5 + * Version: 7.1 * - * Copyright (C) 1999-2006 Brian Paul All Rights Reserved. + * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -25,28 +25,25 @@ /** * \file tnl/t_vb_program.c - * \brief Pipeline stage for executing NVIDIA vertex programs. + * \brief Pipeline stage for executing vertex programs. * \author Brian Paul, Keith Whitwell */ -#include "glheader.h" -#include "api_noop.h" -#include "colormac.h" -#include "context.h" -#include "dlist.h" -#include "hash.h" -#include "light.h" -#include "macros.h" -#include "imports.h" -#include "simple_list.h" -#include "mtypes.h" -#include "program_instruction.h" -#include "nvvertexec.h" -#include "nvprogram.h" +#include "main/glheader.h" +#include "main/colormac.h" +#include "main/context.h" +#include "main/macros.h" +#include "main/imports.h" +#include "shader/prog_instruction.h" +#include "shader/prog_statevars.h" +#include "shader/prog_execute.h" +#include "swrast/s_context.h" +#include "swrast/s_texfilter.h" -#include "t_context.h" -#include "t_pipeline.h" +#include "tnl/tnl.h" +#include "tnl/t_context.h" +#include "tnl/t_pipeline.h" @@ -55,7 +52,7 @@ */ struct vp_stage_data { /** The results of running the vertex program go into these arrays. */ - GLvector4f attribs[15]; + GLvector4f results[VERT_RESULT_MAX]; GLvector4f ndcCoords; /**< normalized device coords */ GLubyte *clipmask; /**< clip flags */ @@ -66,6 +63,229 @@ struct vp_stage_data { #define VP_STAGE_DATA(stage) ((struct vp_stage_data *)(stage->privatePtr)) +static void +userclip( GLcontext *ctx, + GLvector4f *clip, + GLubyte *clipmask, + GLubyte *clipormask, + GLubyte *clipandmask ) +{ + GLuint p; + + for (p = 0; p < ctx->Const.MaxClipPlanes; p++) { + if (ctx->Transform.ClipPlanesEnabled & (1 << p)) { + GLuint nr, i; + const GLfloat a = ctx->Transform._ClipUserPlane[p][0]; + const GLfloat b = ctx->Transform._ClipUserPlane[p][1]; + const GLfloat c = ctx->Transform._ClipUserPlane[p][2]; + const GLfloat d = ctx->Transform._ClipUserPlane[p][3]; + GLfloat *coord = (GLfloat *)clip->data; + GLuint stride = clip->stride; + GLuint count = clip->count; + + for (nr = 0, i = 0 ; i < count ; i++) { + GLfloat dp = (coord[0] * a + + coord[1] * b + + coord[2] * c + + coord[3] * d); + + if (dp < 0) { + nr++; + clipmask[i] |= CLIP_USER_BIT; + } + + STRIDE_F(coord, stride); + } + + if (nr > 0) { + *clipormask |= CLIP_USER_BIT; + if (nr == count) { + *clipandmask |= CLIP_USER_BIT; + return; + } + } + } + } +} + + +static GLboolean +do_ndc_cliptest(GLcontext *ctx, struct vp_stage_data *store) +{ + TNLcontext *tnl = TNL_CONTEXT(ctx); + struct vertex_buffer *VB = &tnl->vb; + /* Cliptest and perspective divide. Clip functions must clear + * the clipmask. + */ + store->ormask = 0; + store->andmask = CLIP_FRUSTUM_BITS; + + if (tnl->NeedNdcCoords) { + VB->NdcPtr = + _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr, + &store->ndcCoords, + store->clipmask, + &store->ormask, + &store->andmask ); + } + else { + VB->NdcPtr = NULL; + _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr, + NULL, + store->clipmask, + &store->ormask, + &store->andmask ); + } + + if (store->andmask) { + /* All vertices are outside the frustum */ + return GL_FALSE; + } + + /* Test userclip planes. This contributes to VB->ClipMask. + */ + /** XXX NEW_SLANG _Enabled ??? */ + if (ctx->Transform.ClipPlanesEnabled && (!ctx->VertexProgram._Enabled || + ctx->VertexProgram.Current->IsPositionInvariant)) { + userclip( ctx, + VB->ClipPtr, + store->clipmask, + &store->ormask, + &store->andmask ); + + if (store->andmask) { + return GL_FALSE; + } + } + + VB->ClipAndMask = store->andmask; + VB->ClipOrMask = store->ormask; + VB->ClipMask = store->clipmask; + + return GL_TRUE; +} + + +/** + * XXX the texture sampling code in this module is a bit of a hack. + * The texture sampling code is in swrast, though it doesn't have any + * real dependencies on the rest of swrast. It should probably be + * moved into main/ someday. + */ +static void +vp_fetch_texel(GLcontext *ctx, const GLfloat texcoord[4], GLfloat lambda, + GLuint unit, GLfloat color[4]) +{ + SWcontext *swrast = SWRAST_CONTEXT(ctx); + + /* XXX use a float-valued TextureSample routine here!!! */ + swrast->TextureSample[unit](ctx, ctx->Texture.Unit[unit]._Current, + 1, (const GLfloat (*)[4]) texcoord, + &lambda, (GLfloat (*)[4]) color); +} + + +/** + * Called via ctx->Driver.ProgramStringNotify() after a new vertex program + * string has been parsed. + */ +void +_tnl_program_string(GLcontext *ctx, GLenum target, struct gl_program *program) +{ + /* No-op. + * If we had derived anything from the program that was private to this + * stage we'd recompute/validate it here. + */ +} + + +/** + * Initialize virtual machine state prior to executing vertex program. + */ +static void +init_machine(GLcontext *ctx, struct gl_program_machine *machine) +{ + /* Input registers get initialized from the current vertex attribs */ + MEMCPY(machine->VertAttribs, ctx->Current.Attrib, + MAX_VERTEX_GENERIC_ATTRIBS * 4 * sizeof(GLfloat)); + + if (ctx->VertexProgram._Current->IsNVProgram) { + GLuint i; + /* Output/result regs are initialized to [0,0,0,1] */ + for (i = 0; i < MAX_NV_VERTEX_PROGRAM_OUTPUTS; i++) { + ASSIGN_4V(machine->Outputs[i], 0.0F, 0.0F, 0.0F, 1.0F); + } + /* Temp regs are initialized to [0,0,0,0] */ + for (i = 0; i < MAX_NV_VERTEX_PROGRAM_TEMPS; i++) { + ASSIGN_4V(machine->Temporaries[i], 0.0F, 0.0F, 0.0F, 0.0F); + } + for (i = 0; i < MAX_VERTEX_PROGRAM_ADDRESS_REGS; i++) { + ASSIGN_4V(machine->AddressReg[i], 0, 0, 0, 0); + } + } + + machine->NumDeriv = 0; + + /* init condition codes */ + machine->CondCodes[0] = COND_EQ; + machine->CondCodes[1] = COND_EQ; + machine->CondCodes[2] = COND_EQ; + machine->CondCodes[3] = COND_EQ; + + /* init call stack */ + machine->StackDepth = 0; + + machine->FetchTexelLod = vp_fetch_texel; + machine->FetchTexelDeriv = NULL; /* not used by vertex programs */ + + machine->Samplers = ctx->VertexProgram._Current->Base.SamplerUnits; +} + + +/** + * Map the texture images which the vertex program will access (if any). + */ +static void +map_textures(GLcontext *ctx, const struct gl_vertex_program *vp) +{ + GLuint u; + + if (!ctx->Driver.MapTexture) + return; + + for (u = 0; u < ctx->Const.MaxVertexTextureImageUnits; u++) { + if (vp->Base.TexturesUsed[u]) { + /* Note: _Current *should* correspond to the target indicated + * in TexturesUsed[u]. + */ + ctx->Driver.MapTexture(ctx, ctx->Texture.Unit[u]._Current); + } + } +} + + +/** + * Unmap the texture images which were used by the vertex program (if any). + */ +static void +unmap_textures(GLcontext *ctx, const struct gl_vertex_program *vp) +{ + GLuint u; + + if (!ctx->Driver.MapTexture) + return; + + for (u = 0; u < ctx->Const.MaxVertexTextureImageUnits; u++) { + if (vp->Base.TexturesUsed[u]) { + /* Note: _Current *should* correspond to the target indicated + * in TexturesUsed[u]. + */ + ctx->Driver.UnmapTexture(ctx, ctx->Texture.Unit[u]._Current); + } + } +} + + /** * This function executes vertex programs */ @@ -75,24 +295,36 @@ run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage ) TNLcontext *tnl = TNL_CONTEXT(ctx); struct vp_stage_data *store = VP_STAGE_DATA(stage); struct vertex_buffer *VB = &tnl->vb; - struct gl_vertex_program *program = ctx->VertexProgram.Current; - struct vp_machine machine; - GLuint i; + struct gl_vertex_program *program = ctx->VertexProgram._Current; + struct gl_program_machine machine; + GLuint outputs[VERT_RESULT_MAX], numOutputs; + GLuint i, j; - if (ctx->ShaderObjects._VertexShaderPresent) + if (!program) return GL_TRUE; - if (!ctx->VertexProgram._Enabled || - !program->IsNVProgram) - return GL_TRUE; + if (program->IsNVProgram) { + _mesa_load_tracked_matrices(ctx); + } + else { + /* ARB program or vertex shader */ + _mesa_load_state_parameters(ctx, program->Base.Parameters); + } + + /* make list of outputs to save some time below */ + numOutputs = 0; + for (i = 0; i < VERT_RESULT_MAX; i++) { + if (program->Base.OutputsWritten & (1 << i)) { + outputs[numOutputs++] = i; + } + } - /* load program parameter registers (they're read-only) */ - _mesa_init_vp_per_primitive_registers(ctx); + map_textures(ctx, program); for (i = 0; i < VB->Count; i++) { GLuint attr; - _mesa_init_vp_per_vertex_registers(ctx, &machine); + init_machine(ctx, &machine); #if 0 printf("Input %d: %f, %f, %f, %f\n", i, @@ -119,101 +351,117 @@ run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage ) const GLuint size = VB->AttribPtr[attr]->size; const GLuint stride = VB->AttribPtr[attr]->stride; const GLfloat *data = (GLfloat *) (ptr + stride * i); - COPY_CLEAN_4V(machine.Inputs[attr], size, data); + COPY_CLEAN_4V(machine.VertAttribs[attr], size, data); } } /* execute the program */ - ASSERT(program); - _mesa_exec_vertex_program(ctx, &machine, program); + _mesa_execute_program(ctx, &program->Base, &machine); + + /* copy the output registers into the VB->attribs arrays */ + for (j = 0; j < numOutputs; j++) { + const GLuint attr = outputs[j]; + COPY_4V(store->results[attr].data[i], machine.Outputs[attr]); + } +#if 0 + printf("HPOS: %f %f %f %f\n", + machine.Outputs[0][0], + machine.Outputs[0][1], + machine.Outputs[0][2], + machine.Outputs[0][3]); +#endif + } + + unmap_textures(ctx, program); - /* Fixup fog an point size results if needed */ + /* Fixup fog and point size results if needed */ + if (program->IsNVProgram) { if (ctx->Fog.Enabled && (program->Base.OutputsWritten & (1 << VERT_RESULT_FOGC)) == 0) { - machine.Outputs[VERT_RESULT_FOGC][0] = 1.0; + for (i = 0; i < VB->Count; i++) { + store->results[VERT_RESULT_FOGC].data[i][0] = 1.0; + } } if (ctx->VertexProgram.PointSizeEnabled && (program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) == 0) { - machine.Outputs[VERT_RESULT_PSIZ][0] = ctx->Point.Size; + for (i = 0; i < VB->Count; i++) { + store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size; + } } + } - /* copy the output registers into the VB->attribs arrays */ - /* XXX (optimize) could use a conditional and smaller loop limit here */ - for (attr = 0; attr < 15; attr++) { - COPY_4V(store->attribs[attr].data[i], machine.Outputs[attr]); + if (program->IsPositionInvariant) { + /* We need the exact same transform as in the fixed function path here + * to guarantee invariance, depending on compiler optimization flags + * results could be different otherwise. + */ + VB->ClipPtr = TransformRaw( &store->results[0], + &ctx->_ModelProjectMatrix, + VB->AttribPtr[0] ); + + /* Drivers expect this to be clean to element 4... + */ + switch (VB->ClipPtr->size) { + case 1: + /* impossible */ + case 2: + _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 ); + /* fall-through */ + case 3: + _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 ); + /* fall-through */ + case 4: + break; } } + else { + /* Setup the VB pointers so that the next pipeline stages get + * their data from the right place (the program output arrays). + */ + VB->ClipPtr = &store->results[VERT_RESULT_HPOS]; + VB->ClipPtr->size = 4; + VB->ClipPtr->count = VB->Count; + } - /* Setup the VB pointers so that the next pipeline stages get - * their data from the right place (the program output arrays). - */ - VB->ClipPtr = &store->attribs[VERT_RESULT_HPOS]; - VB->ClipPtr->size = 4; - VB->ClipPtr->count = VB->Count; - VB->ColorPtr[0] = &store->attribs[VERT_RESULT_COL0]; - VB->ColorPtr[1] = &store->attribs[VERT_RESULT_BFC0]; - VB->SecondaryColorPtr[0] = &store->attribs[VERT_RESULT_COL1]; - VB->SecondaryColorPtr[1] = &store->attribs[VERT_RESULT_BFC1]; - VB->FogCoordPtr = &store->attribs[VERT_RESULT_FOGC]; - - VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->attribs[VERT_RESULT_COL0]; - VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->attribs[VERT_RESULT_COL1]; - VB->AttribPtr[VERT_ATTRIB_FOG] = &store->attribs[VERT_RESULT_FOGC]; - VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->attribs[VERT_RESULT_PSIZ]; + VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0]; + VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0]; + VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1]; + VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1]; + VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC]; + + VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0]; + VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1]; + VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC]; + VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ]; for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) { VB->TexCoordPtr[i] = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i] - = &store->attribs[VERT_RESULT_TEX0 + i]; + = &store->results[VERT_RESULT_TEX0 + i]; } - /* Cliptest and perspective divide. Clip functions must clear - * the clipmask. - */ - store->ormask = 0; - store->andmask = CLIP_FRUSTUM_BITS; - - if (tnl->NeedNdcCoords) { - VB->NdcPtr = - _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr, - &store->ndcCoords, - store->clipmask, - &store->ormask, - &store->andmask ); - } - else { - VB->NdcPtr = NULL; - _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr, - NULL, - store->clipmask, - &store->ormask, - &store->andmask ); + for (i = 0; i < ctx->Const.MaxVarying; i++) { + if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) { + /* Note: varying results get put into the generic attributes */ + VB->AttribPtr[VERT_ATTRIB_GENERIC0+i] + = &store->results[VERT_RESULT_VAR0 + i]; + } } - if (store->andmask) /* All vertices are outside the frustum */ - return GL_FALSE; - - /* This is where we'd do clip testing against the user-defined - * clipping planes, but they're not supported by vertex programs. + /* Perform NDC and cliptest operations: */ - - VB->ClipOrMask = store->ormask; - VB->ClipMask = store->clipmask; - - return GL_TRUE; + return do_ndc_cliptest(ctx, store); } - - /** * Called the first time stage->run is called. In effect, don't * allocate data until the first time the stage is run. */ -static GLboolean init_vp( GLcontext *ctx, - struct tnl_pipeline_stage *stage ) +static GLboolean +init_vp(GLcontext *ctx, struct tnl_pipeline_stage *stage) { TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_buffer *VB = &(tnl->vb); @@ -227,10 +475,9 @@ static GLboolean init_vp( GLcontext *ctx, return GL_FALSE; /* Allocate arrays of vertex output values */ - /* XXX change '15' to a named constant */ - for (i = 0; i < 15; i++) { - _mesa_vector4f_alloc( &store->attribs[i], 0, size, 32 ); - store->attribs[i].size = 4; + for (i = 0; i < VERT_RESULT_MAX; i++) { + _mesa_vector4f_alloc( &store->results[i], 0, size, 32 ); + store->results[i].size = 4; } /* a few other misc allocations */ @@ -241,13 +488,11 @@ static GLboolean init_vp( GLcontext *ctx, } - - - /** * Destructor for this pipeline stage. */ -static void dtr( struct tnl_pipeline_stage *stage ) +static void +dtr(struct tnl_pipeline_stage *stage) { struct vp_stage_data *store = VP_STAGE_DATA(stage); @@ -256,7 +501,7 @@ static void dtr( struct tnl_pipeline_stage *stage ) /* free the vertex program result arrays */ for (i = 0; i < VERT_RESULT_MAX; i++) - _mesa_vector4f_free( &store->attribs[i] ); + _mesa_vector4f_free( &store->results[i] ); /* free misc arrays */ _mesa_vector4f_free( &store->ndcCoords ); @@ -267,6 +512,17 @@ static void dtr( struct tnl_pipeline_stage *stage ) } } + +static void +validate_vp_stage(GLcontext *ctx, struct tnl_pipeline_stage *stage) +{ + if (ctx->VertexProgram._Current) { + _swrast_update_texture_samplers(ctx); + } +} + + + /** * Public description of this pipeline stage. */ @@ -276,6 +532,6 @@ const struct tnl_pipeline_stage _tnl_vertex_program_stage = NULL, /* private_data */ init_vp, /* create */ dtr, /* destroy */ - NULL, /* validate */ + validate_vp_stage, /* validate */ run_vp /* run -- initially set to ctr */ };