From fc38a3cfe84e4e79af43f29d236748120789a286 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Fri, 23 Oct 2009 00:40:41 -0400 Subject: r600: fix render size prediction --- src/mesa/drivers/dri/r600/r700_vertprog.c | 1 + 1 file changed, 1 insertion(+) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index d12c39c9f7..65c2c3f811 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -580,6 +580,7 @@ void r700SetVertexFormat(GLcontext *ctx, const struct gl_client_array *arrays[], unBit >>= 1; ++unLoc; } + context->radeon.tcl.aos_count = context->nNumActiveAos; } void * r700GetActiveVpShaderBo(GLcontext * ctx) -- cgit v1.2.3 From 45eb9d2f6dced7654291cabb4b8dd02a695db694 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Fri, 23 Oct 2009 01:00:23 -0400 Subject: r600: remove remains of old tnl pipeline --- src/mesa/drivers/dri/r600/r600_context.h | 30 -------- src/mesa/drivers/dri/r600/r700_chip.c | 121 ++---------------------------- src/mesa/drivers/dri/r600/r700_render.c | 6 +- src/mesa/drivers/dri/r600/r700_shader.h | 1 + src/mesa/drivers/dri/r600/r700_state.c | 40 +--------- src/mesa/drivers/dri/r600/r700_vertprog.c | 103 ++++++------------------- src/mesa/drivers/dri/r600/r700_vertprog.h | 8 +- 7 files changed, 41 insertions(+), 268 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r600_context.h b/src/mesa/drivers/dri/r600/r600_context.h index 7f68820fda..394fd757d4 100644 --- a/src/mesa/drivers/dri/r600/r600_context.h +++ b/src/mesa/drivers/dri/r600/r600_context.h @@ -58,29 +58,6 @@ typedef struct r600_context context_t; #include "main/mm.h" -/************ DMA BUFFERS **************/ - -/* The blit width for texture uploads - */ -#define R600_BLIT_WIDTH_BYTES 1024 -#define R600_MAX_TEXTURE_UNITS 8 - -struct r600_texture_state { - int tc_count; /* number of incoming texture coordinates from VAP */ -}; - -/* Perhaps more if we store programs in vmem? */ -/* drm_r600_cmd_header_t->vpu->count is unsigned char */ -#define VSF_MAX_FRAGMENT_LENGTH (255*4) - -/* Can be tested with colormat currently. */ -#define VSF_MAX_FRAGMENT_TEMPS (14) - -#define STATE_R600_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0) -#define STATE_R600_TEXRECT_FACTOR (STATE_INTERNAL_DRIVER+1) - -extern int hw_tcl_on; - #define COLOR_IS_RGBA #define TAG(x) r600##x #include "tnl_dd/t_dd_vertex.h" @@ -167,9 +144,6 @@ struct r600_context { /* Vertex buffers */ - GLvector4f dummy_attrib[_TNL_ATTRIB_MAX]; - GLvector4f *temp_attrib[_TNL_ATTRIB_MAX]; - GLint nNumActiveAos; StreamDesc stream_desc[VERT_ATTRIB_MAX]; struct r700_index_buffer ind_buf; @@ -203,7 +177,6 @@ extern GLboolean r700SyncSurf(context_t *context, uint32_t write_domain, uint32_t sync_type); -extern void r700SetupStreams(GLcontext * ctx); extern void r700Start3D(context_t *context); extern void r600InitAtoms(context_t *context); extern void r700InitDraw(GLcontext *ctx); @@ -213,7 +186,4 @@ extern void r700InitDraw(GLcontext *ctx); #define RADEON_D_PLAYBACK_RAW 2 #define RADEON_D_T 3 -#define r600PackFloat32 radeonPackFloat32 -#define r600PackFloat24 radeonPackFloat24 - #endif /* __R600_CONTEXT_H__ */ diff --git a/src/mesa/drivers/dri/r600/r700_chip.c b/src/mesa/drivers/dri/r600/r700_chip.c index 3ebc53d94f..75b97c56cd 100644 --- a/src/mesa/drivers/dri/r600/r700_chip.c +++ b/src/mesa/drivers/dri/r600/r700_chip.c @@ -141,77 +141,10 @@ static void r700SendTexBorderColorState(GLcontext *ctx, struct radeon_state_atom } } +extern int getTypeSize(GLenum type); static void r700SetupVTXConstants(GLcontext * ctx, - unsigned int nStreamID, void * pAos, - unsigned int size, /* number of elements in vector */ - unsigned int stride, - unsigned int count) /* number of vectors in stream */ -{ - context_t *context = R700_CONTEXT(ctx); - struct radeon_aos * paos = (struct radeon_aos *)pAos; - BATCH_LOCALS(&context->radeon); - radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s\n", __func__); - - unsigned int uSQ_VTX_CONSTANT_WORD0_0; - unsigned int uSQ_VTX_CONSTANT_WORD1_0; - unsigned int uSQ_VTX_CONSTANT_WORD2_0 = 0; - unsigned int uSQ_VTX_CONSTANT_WORD3_0 = 0; - unsigned int uSQ_VTX_CONSTANT_WORD6_0 = 0; - - if (!paos->bo) - return; - - if ((context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV610) || - (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV620) || - (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RS780) || - (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RS880) || - (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV710)) - r700SyncSurf(context, paos->bo, RADEON_GEM_DOMAIN_GTT, 0, TC_ACTION_ENA_bit); - else - r700SyncSurf(context, paos->bo, RADEON_GEM_DOMAIN_GTT, 0, VC_ACTION_ENA_bit); - - uSQ_VTX_CONSTANT_WORD0_0 = paos->offset; - uSQ_VTX_CONSTANT_WORD1_0 = count * (size * 4) - 1; - - SETfield(uSQ_VTX_CONSTANT_WORD2_0, 0, BASE_ADDRESS_HI_shift, BASE_ADDRESS_HI_mask); /* TODO */ - SETfield(uSQ_VTX_CONSTANT_WORD2_0, stride, SQ_VTX_CONSTANT_WORD2_0__STRIDE_shift, - SQ_VTX_CONSTANT_WORD2_0__STRIDE_mask); - SETfield(uSQ_VTX_CONSTANT_WORD2_0, GetSurfaceFormat(GL_FLOAT, size, NULL), - SQ_VTX_CONSTANT_WORD2_0__DATA_FORMAT_shift, - SQ_VTX_CONSTANT_WORD2_0__DATA_FORMAT_mask); /* TODO : trace back api for initial data type, not only GL_FLOAT */ - SETfield(uSQ_VTX_CONSTANT_WORD2_0, SQ_NUM_FORMAT_SCALED, - SQ_VTX_CONSTANT_WORD2_0__NUM_FORMAT_ALL_shift, SQ_VTX_CONSTANT_WORD2_0__NUM_FORMAT_ALL_mask); - SETbit(uSQ_VTX_CONSTANT_WORD2_0, SQ_VTX_CONSTANT_WORD2_0__FORMAT_COMP_ALL_bit); - - SETfield(uSQ_VTX_CONSTANT_WORD3_0, 1, MEM_REQUEST_SIZE_shift, MEM_REQUEST_SIZE_mask); - SETfield(uSQ_VTX_CONSTANT_WORD6_0, SQ_TEX_VTX_VALID_BUFFER, - SQ_TEX_RESOURCE_WORD6_0__TYPE_shift, SQ_TEX_RESOURCE_WORD6_0__TYPE_mask); - - BEGIN_BATCH_NO_AUTOSTATE(9 + 2); - - R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_RESOURCE, 7)); - R600_OUT_BATCH((nStreamID + SQ_FETCH_RESOURCE_VS_OFFSET) * FETCH_RESOURCE_STRIDE); - R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD0_0); - R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD1_0); - R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD2_0); - R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD3_0); - R600_OUT_BATCH(0); - R600_OUT_BATCH(0); - R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD6_0); - R600_OUT_BATCH_RELOC(uSQ_VTX_CONSTANT_WORD0_0, - paos->bo, - uSQ_VTX_CONSTANT_WORD0_0, - RADEON_GEM_DOMAIN_GTT, 0, 0); - END_BATCH(); - COMMIT_BATCH(); - -} - -extern int getTypeSize(GLenum type); -static void r700SetupVTXConstants2(GLcontext * ctx, - void * pAos, - StreamDesc * pStreamDesc) + StreamDesc * pStreamDesc) { context_t *context = R700_CONTEXT(ctx); struct radeon_aos * paos = (struct radeon_aos *)pAos; @@ -295,31 +228,6 @@ static void r700SetupVTXConstants2(GLcontext * ctx, } -void r700SetupStreams(GLcontext *ctx) -{ - context_t *context = R700_CONTEXT(ctx); - struct r700_vertex_program *vp = context->selected_vp; - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *vb = &tnl->vb; - unsigned int i, j = 0; - radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s\n", __func__); - - R600_STATECHANGE(context, vtx); - - for(i=0; imesa_program->Base.InputsRead & (1 << i)) { - rcommon_emit_vector(ctx, - &context->radeon.tcl.aos[j], - vb->AttribPtr[i]->data, - vb->AttribPtr[i]->size, - vb->AttribPtr[i]->stride, - vb->Count); - j++; - } - } - context->radeon.tcl.aos_count = j; -} - static void r700SendVTXState(GLcontext *ctx, struct radeon_state_atom *atom) { context_t *context = R700_CONTEXT(ctx); @@ -343,25 +251,12 @@ static void r700SendVTXState(GLcontext *ctx, struct radeon_state_atom *atom) COMMIT_BATCH(); for(i=0; imesa_program->Base.InputsRead & (1 << i)) - { - if(1 == context->selected_vp->uiVersion) - { - /* currently aos are packed */ - r700SetupVTXConstants(ctx, - i, - (void*)(&context->radeon.tcl.aos[j]), - (unsigned int)context->radeon.tcl.aos[j].components, - (unsigned int)context->radeon.tcl.aos[j].stride * 4, - (unsigned int)context->radeon.tcl.aos[j].count); - } - else - { /* context->selected_vp->uiVersion == 2 : aos not always packed */ - r700SetupVTXConstants2(ctx, - (void*)(&context->radeon.tcl.aos[j]), - &(context->stream_desc[j])); - } - j++; + if(vp->mesa_program->Base.InputsRead & (1 << i)) + { + r700SetupVTXConstants(ctx, + (void*)(&context->radeon.tcl.aos[j]), + &(context->stream_desc[j])); + j++; } } } diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c index c2e7680eae..3e1ce9fb72 100644 --- a/src/mesa/drivers/dri/r600/r700_render.c +++ b/src/mesa/drivers/dri/r600/r700_render.c @@ -550,7 +550,7 @@ static void r700AlignDataToDword(GLcontext *ctx, attr->stride = dst_stride; } -static void r700SetupStreams2(GLcontext *ctx, const struct gl_client_array *input[], int count) +static void r700SetupStreams(GLcontext *ctx, const struct gl_client_array *input[], int count) { context_t *context = R700_CONTEXT(ctx); GLuint stride; @@ -825,7 +825,7 @@ static GLboolean r700TryDrawPrims(GLcontext *ctx, _tnl_UpdateFixedFunctionProgram(ctx); r700SetVertexFormat(ctx, arrays, max_index + 1); /* shaders need to be updated before buffers are validated */ - r700UpdateShaders2(ctx); + r700UpdateShaders(ctx); if (!r600ValidateBuffers(ctx)) return GL_FALSE; @@ -845,7 +845,7 @@ static GLboolean r700TryDrawPrims(GLcontext *ctx, + context->radeon.cmdbuf.cs->cdw; r700SetupIndexBuffer(ctx, ib); - r700SetupStreams2(ctx, arrays, max_index + 1); + r700SetupStreams(ctx, arrays, max_index + 1); radeonEmitState(radeon); diff --git a/src/mesa/drivers/dri/r600/r700_shader.h b/src/mesa/drivers/dri/r600/r700_shader.h index 997cb05aaf..c6a058617e 100644 --- a/src/mesa/drivers/dri/r600/r700_shader.h +++ b/src/mesa/drivers/dri/r600/r700_shader.h @@ -128,6 +128,7 @@ typedef struct R700_Shader //Internal void AddInstToList(TypedShaderList * plstCFInstructions, R700ShaderInstruction * pInst); +void TakeInstOutFromList(TypedShaderList * plstCFInstructions, R700ShaderInstruction * pInst); void ResolveLinks(R700_Shader *pShader); void Assemble(R700_Shader *pShader); diff --git a/src/mesa/drivers/dri/r600/r700_state.c b/src/mesa/drivers/dri/r600/r700_state.c index 3d3c8b958f..9a6a68a68c 100644 --- a/src/mesa/drivers/dri/r600/r700_state.c +++ b/src/mesa/drivers/dri/r600/r700_state.c @@ -61,12 +61,9 @@ static void r700UpdatePolygonMode(GLcontext * ctx); static void r700SetPolygonOffsetState(GLcontext * ctx, GLboolean state); static void r700SetStencilState(GLcontext * ctx, GLboolean state); -void r700UpdateShaders (GLcontext * ctx) //---------------------------------- +void r700UpdateShaders(GLcontext * ctx) { context_t *context = R700_CONTEXT(ctx); - GLvector4f dummy_attrib[_TNL_ATTRIB_MAX]; - GLvector4f *temp_attrib[_TNL_ATTRIB_MAX]; - int i; /* should only happenen once, just after context is created */ /* TODO: shouldn't we fallback to sw here? */ @@ -77,40 +74,7 @@ void r700UpdateShaders (GLcontext * ctx) //---------------------------------- r700SelectFragmentShader(ctx); - if (context->radeon.NewGLState) { - for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) { - /* mat states from state var not array for sw */ - dummy_attrib[i].stride = 0; - temp_attrib[i] = TNL_CONTEXT(ctx)->vb.AttribPtr[i]; - TNL_CONTEXT(ctx)->vb.AttribPtr[i] = &(dummy_attrib[i]); - } - - _tnl_UpdateFixedFunctionProgram(ctx); - - for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) { - TNL_CONTEXT(ctx)->vb.AttribPtr[i] = temp_attrib[i]; - } - } - - r700SelectVertexShader(ctx, 1); - r700UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS); - context->radeon.NewGLState = 0; -} - -void r700UpdateShaders2(GLcontext * ctx) -{ - context_t *context = R700_CONTEXT(ctx); - - /* should only happenen once, just after context is created */ - /* TODO: shouldn't we fallback to sw here? */ - if (!ctx->FragmentProgram._Current) { - _mesa_fprintf(stderr, "No ctx->FragmentProgram._Current!!\n"); - return; - } - - r700SelectFragmentShader(ctx); - - r700SelectVertexShader(ctx, 2); + r700SelectVertexShader(ctx); r700UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS); context->radeon.NewGLState = 0; } diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index 65c2c3f811..c84b0ac059 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -203,22 +203,11 @@ void Map_Vertex_Program(GLcontext *ctx, pAsm->number_used_registers += num_inputs; // Create VFETCH instructions for inputs - if(1 == vp->uiVersion) - { - if (GL_TRUE != Process_Vertex_Program_Vfetch_Instructions(vp, mesa_vp) ) - { - radeon_error("Calling Process_Vertex_Program_Vfetch_Instructions return error. \n"); - return; - } - } - else - { - if (GL_TRUE != Process_Vertex_Program_Vfetch_Instructions2(ctx, vp, mesa_vp) ) - { - radeon_error("Calling Process_Vertex_Program_Vfetch_Instructions2 return error. \n"); - return; - } - } + if (GL_TRUE != Process_Vertex_Program_Vfetch_Instructions2(ctx, vp, mesa_vp) ) + { + radeon_error("Calling Process_Vertex_Program_Vfetch_Instructions2 return error. \n"); + return; + } // Map Outputs pAsm->number_of_exports = Map_Vertex_Output(pAsm, mesa_vp, pAsm->number_used_registers); @@ -228,7 +217,7 @@ void Map_Vertex_Program(GLcontext *ctx, pAsm->number_used_registers += pAsm->number_of_exports; pAsm->pucOutMask = (unsigned char*) MALLOC(pAsm->number_of_exports); - + for(ui=0; uinumber_of_exports; ui++) { pAsm->pucOutMask[ui] = 0x0; @@ -245,7 +234,7 @@ void Map_Vertex_Program(GLcontext *ctx, { /* fix func t_vp uses NumTemporaries */ pAsm->number_used_registers += mesa_vp->Base.NumTemporaries; } - + pAsm->uFirstHelpReg = pAsm->number_used_registers; } @@ -300,18 +289,13 @@ GLboolean Find_Instruction_Dependencies_vp(struct r700_vertex_program *vp, } struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, - struct gl_vertex_program *mesa_vp, - GLint nVer) + struct gl_vertex_program *mesa_vp) { context_t *context = R700_CONTEXT(ctx); struct r700_vertex_program *vp; - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *vb = &tnl->vb; - unsigned int unBit; unsigned int i; vp = _mesa_calloc(sizeof(*vp)); - vp->uiVersion = nVer; vp->mesa_program = (struct gl_vertex_program *)_mesa_clone_program(ctx, &mesa_vp->Base); if (mesa_vp->IsPositionInvariant) @@ -319,29 +303,13 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, _mesa_insert_mvp_code(ctx, vp->mesa_program); } - if( 1 == nVer ) + for(i=0; inNumActiveAos; i++) { - for(i=0; imesa_program->Base.InputsRead & unBit) /* ctx->Array.ArrayObj->xxxxxxx */ - { - vp->aos_desc[i].size = vb->AttribPtr[i]->size; - vp->aos_desc[i].stride = vb->AttribPtr[i]->size * sizeof(GL_FLOAT);/* when emit array, data is packed. vb->AttribPtr[i]->stride;*/ - vp->aos_desc[i].type = GL_FLOAT; - } - } + vp->aos_desc[i].size = context->stream_desc[i].size; + vp->aos_desc[i].stride = context->stream_desc[i].stride; + vp->aos_desc[i].type = context->stream_desc[i].type; } - else - { - for(i=0; inNumActiveAos; i++) - { - vp->aos_desc[i].size = context->stream_desc[i].size; - vp->aos_desc[i].stride = context->stream_desc[i].stride; - vp->aos_desc[i].type = context->stream_desc[i].type; - } - } - + if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770) { vp->r700AsmCode.bR6xx = 1; @@ -354,14 +322,14 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, if(GL_FALSE == Find_Instruction_Dependencies_vp(vp, vp->mesa_program)) { return NULL; - } + } if(GL_FALSE == AssembleInstr(vp->mesa_program->Base.NumInstructions, - &(vp->mesa_program->Base.Instructions[0]), + &(vp->mesa_program->Base.Instructions[0]), &(vp->r700AsmCode)) ) { return NULL; - } + } if(GL_FALSE == Process_Vertex_Exports(&(vp->r700AsmCode), vp->mesa_program->Base.OutputsWritten) ) { @@ -378,14 +346,11 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, return vp; } -void r700SelectVertexShader(GLcontext *ctx, GLint nVersion) +void r700SelectVertexShader(GLcontext *ctx) { context_t *context = R700_CONTEXT(ctx); struct r700_vertex_program_cont *vpc; struct r700_vertex_program *vp; - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *vb = &tnl->vb; - unsigned int unBit; unsigned int i; GLboolean match; GLbitfield InputsRead; @@ -396,47 +361,27 @@ void r700SelectVertexShader(GLcontext *ctx, GLint nVersion) if (vpc->mesa_program.IsPositionInvariant) { InputsRead |= VERT_BIT_POS; - } - + } + for (vp = vpc->progs; vp; vp = vp->next) { - if (vp->uiVersion != nVersion ) - continue; - match = GL_TRUE; - if ( 1 == nVersion ) + match = GL_TRUE; + for(i=0; inNumActiveAos; i++) { - for(i=0; iaos_desc[i].size != context->stream_desc[i].size) { - if (vp->aos_desc[i].size != vb->AttribPtr[i]->size) - { match = GL_FALSE; break; - } } - } } - else - { - for(i=0; inNumActiveAos; i++) - { - if (vp->aos_desc[i].size != context->stream_desc[i].size) - { - match = GL_FALSE; - break; - } - } - } - if (match) + if (match) { context->selected_vp = vp; return; } } - vp = r700TranslateVertexShader(ctx, &(vpc->mesa_program), nVersion); + vp = r700TranslateVertexShader(ctx, &(vpc->mesa_program)); if(!vp) { radeon_error("Failed to translate vertex shader. \n"); diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.h b/src/mesa/drivers/dri/r600/r700_vertprog.h index f9a3e395ee..00824c29d3 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.h +++ b/src/mesa/drivers/dri/r600/r700_vertprog.h @@ -52,8 +52,7 @@ struct r700_vertex_program GLboolean translated; GLboolean loaded; - GLint uiVersion; - + void * shaderbo; ArrayDesc aos_desc[VERT_ATTRIB_MAX]; @@ -87,11 +86,10 @@ GLboolean Find_Instruction_Dependencies_vp(struct r700_vertex_program *vp, struct gl_vertex_program *mesa_vp); struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, - struct gl_vertex_program *mesa_vp, - GLint nVer); + struct gl_vertex_program *mesa_vp); /* Interface */ -extern void r700SelectVertexShader(GLcontext *ctx, GLint nVersion); +extern void r700SelectVertexShader(GLcontext *ctx); extern void r700SetVertexFormat(GLcontext *ctx, const struct gl_client_array *arrays[], int count); extern GLboolean r700SetupVertexProgram(GLcontext * ctx); -- cgit v1.2.3 From 738b394769bb95036635f7a00a1ef08890c5be63 Mon Sep 17 00:00:00 2001 From: Andre Maasikas Date: Fri, 23 Oct 2009 14:25:09 +0300 Subject: r600: for position invariant programs reading vert_pos is not always known at this point --- src/mesa/drivers/dri/r600/r700_vertprog.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index c84b0ac059..ffc6068bd8 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -515,6 +515,11 @@ void r700SetVertexFormat(GLcontext *ctx, const struct gl_client_array *arrays[], unsigned int unBit = mesa_vp->Base.InputsRead; context->nNumActiveAos = 0; + if (mesa_vp->IsPositionInvariant) + { + unBit |= VERT_BIT_POS; + } + while(unBit) { if(unBit & 1) -- cgit v1.2.3 From 3132853e1242607d5ff62785cd7dad5ef3a783d0 Mon Sep 17 00:00:00 2001 From: Richard Li Date: Tue, 17 Nov 2009 16:25:02 -0500 Subject: r600 : Initial version of glsl fc. --- src/mesa/drivers/dri/r600/r700_assembler.c | 2572 ++++++++++++++++++++++++---- src/mesa/drivers/dri/r600/r700_assembler.h | 111 +- src/mesa/drivers/dri/r600/r700_chip.c | 99 +- src/mesa/drivers/dri/r600/r700_fragprog.c | 94 +- src/mesa/drivers/dri/r600/r700_shader.c | 9 +- src/mesa/drivers/dri/r600/r700_shader.h | 3 +- src/mesa/drivers/dri/r600/r700_vertprog.c | 21 +- 7 files changed, 2507 insertions(+), 402 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index e0d7d4fa6b..4b5d40bd3a 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -38,6 +38,8 @@ #include "r700_assembler.h" +#define USE_CF_FOR_CONTINUE_BREAK 1 + BITS addrmode_PVSDST(PVSDST * pPVSDST) { return pPVSDST->addrmode0 | ((BITS)pPVSDST->addrmode1 << 1); @@ -343,6 +345,8 @@ unsigned int r700GetNumOperands(r700_AssemblerBase* pAsm) case SQ_OP2_INST_MIN: //case SQ_OP2_INST_MAX_DX10: //case SQ_OP2_INST_MIN_DX10: + case SQ_OP2_INST_SETE: + case SQ_OP2_INST_SETNE: case SQ_OP2_INST_SETGT: case SQ_OP2_INST_SETGE: case SQ_OP2_INST_PRED_SETE: @@ -398,6 +402,9 @@ int Init_r700_AssemblerBase(SHADER_PIPE_TYPE spt, r700_AssemblerBase* pAsm, R700 pAsm->number_of_exports = 0; pAsm->number_of_export_opcodes = 0; + pAsm->alu_x_opcode = 0; + + pAsm->D2.bits = 0; pAsm->D.bits = 0; pAsm->S[0].bits = 0; @@ -474,6 +481,22 @@ int Init_r700_AssemblerBase(SHADER_PIPE_TYPE spt, r700_AssemblerBase* pAsm, R700 pAsm->is_tex = GL_FALSE; pAsm->need_tex_barrier = GL_FALSE; + pAsm->subs = NULL; + pAsm->unSubArraySize = 0; + pAsm->unSubArrayPointer = 0; + pAsm->callers = NULL; + pAsm->unCallerArraySize = 0; + pAsm->unCallerArrayPointer = 0; + + pAsm->CALLSP = 0; + pAsm->CALLSTACK[0].FCSP_BeforeEntry; + pAsm->CALLSTACK[0].plstCFInstructions_local + = &(pAsm->pR700Shader->lstCFInstructions); + + SetActiveCFlist(pAsm->pR700Shader, pAsm->CALLSTACK[0].plstCFInstructions_local); + + pAsm->unCFflags = 0; + return 0; } @@ -592,6 +615,31 @@ int check_current_clause(r700_AssemblerBase* pAsm, return GL_TRUE; } +GLboolean add_cf_instruction(r700_AssemblerBase* pAsm) +{ + if(GL_FALSE == check_current_clause(pAsm, CF_OTHER_CLAUSE)) + { + return GL_FALSE; + } + + pAsm->cf_current_cf_clause_ptr = + (R700ControlFlowGenericClause*) CALLOC_STRUCT(R700ControlFlowGenericClause); + + if (pAsm->cf_current_cf_clause_ptr != NULL) + { + Init_R700ControlFlowGenericClause(pAsm->cf_current_cf_clause_ptr); + AddCFInstruction( pAsm->pR700Shader, + (R700ControlFlowInstruction *)pAsm->cf_current_cf_clause_ptr ); + } + else + { + radeon_error("Could not allocate a new VFetch CF instruction.\n"); + return GL_FALSE; + } + + return GL_TRUE; +} + GLboolean add_vfetch_instruction(r700_AssemblerBase* pAsm, R700VertexInstruction* vertex_instruction_ptr) { @@ -1153,6 +1201,7 @@ GLboolean assemble_src(r700_AssemblerBase *pAsm, case PROGRAM_LOCAL_PARAM: case PROGRAM_ENV_PARAM: case PROGRAM_STATE_VAR: + case PROGRAM_UNIFORM: if (1 == pILInst->SrcReg[src].RelAddr) { setaddrmode_PVSSRC(&(pAsm->S[fld].src), ADDR_RELATIVE_A0); @@ -1179,7 +1228,7 @@ GLboolean assemble_src(r700_AssemblerBase *pAsm, } break; default: - radeon_error("Invalid source argument type\n"); + radeon_error("Invalid source argument type : %d \n", pILInst->SrcReg[src].File); return GL_FALSE; } } @@ -1315,7 +1364,7 @@ GLboolean tex_src(r700_AssemblerBase *pAsm) case FRAG_ATTRIB_TEX0: case FRAG_ATTRIB_TEX1: case FRAG_ATTRIB_TEX2: - case FRAG_ATTRIB_TEX3: + case FRAG_ATTRIB_TEX3: case FRAG_ATTRIB_TEX4: case FRAG_ATTRIB_TEX5: case FRAG_ATTRIB_TEX6: @@ -1335,6 +1384,16 @@ GLboolean tex_src(r700_AssemblerBase *pAsm) fprintf(stderr, "FRAG_ATTRIB_VAR0 unsupported\n"); break; } + + if( (pILInst->SrcReg[0].Index >= FRAG_ATTRIB_VAR0) || + (pILInst->SrcReg[0].Index < FRAG_ATTRIB_MAX) ) + { + bValidTexCoord = GL_TRUE; + pAsm->S[0].src.reg = + pAsm->uiFP_AttributeMap[pILInst->SrcReg[0].Index]; + pAsm->S[0].src.rtype = SRC_REG_INPUT; + } + break; } } @@ -1517,6 +1576,10 @@ GLboolean assemble_alu_src(R700ALUInstruction* alu_instruction_ptr, { src_sel = pSource->reg + CFILE_REGISTER_OFFSET; } + else if (pSource->rtype == SRC_REC_LITERAL) + { + src_sel = SQ_ALU_SRC_LITERAL; + } else { radeon_error("Source (%d) register type (%d) not one of TEMP, INPUT, or CONSTANT.\n", @@ -1606,7 +1669,8 @@ GLboolean add_alu_instruction(r700_AssemblerBase* pAsm, return GL_FALSE; } - if ( pAsm->cf_current_alu_clause_ptr == NULL || + if ( pAsm->alu_x_opcode != 0 || + pAsm->cf_current_alu_clause_ptr == NULL || ( (pAsm->cf_current_alu_clause_ptr != NULL) && (pAsm->cf_current_alu_clause_ptr->m_Word1.f.count >= (GetCFMaxInstructions(pAsm->cf_current_alu_clause_ptr->m_ShaderInstType)-contiguous_slots_needed-1) ) ) ) @@ -1636,9 +1700,17 @@ GLboolean add_alu_instruction(r700_AssemblerBase* pAsm, pAsm->cf_current_alu_clause_ptr->m_Word1.f.kcache_addr0 = 0x0; pAsm->cf_current_alu_clause_ptr->m_Word1.f.kcache_addr1 = 0x0; - //cf_current_alu_clause_ptr->m_Word1.f.count = number_of_scalar_operations - 1; pAsm->cf_current_alu_clause_ptr->m_Word1.f.count = 0x0; - pAsm->cf_current_alu_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_ALU; + + if(pAsm->alu_x_opcode != 0) + { + pAsm->cf_current_alu_clause_ptr->m_Word1.f.cf_inst = pAsm->alu_x_opcode; + pAsm->alu_x_opcode = 0; + } + else + { + pAsm->cf_current_alu_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_ALU; + } pAsm->cf_current_alu_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; @@ -2358,146 +2430,711 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) return GL_TRUE; } -GLboolean next_ins(r700_AssemblerBase *pAsm) +GLboolean assemble_alu_instruction2(r700_AssemblerBase *pAsm) { - struct prog_instruction *pILInst = &(pAsm->pILInst[pAsm->uiCurInst]); + GLuint number_of_scalar_operations; + GLboolean is_single_scalar_operation; + GLuint scalar_channel_index; - if( GL_TRUE == pAsm->is_tex ) + PVSSRC * pcurrent_source; + int current_source_index; + GLuint contiguous_slots_needed; + + GLuint uNumSrc = r700GetNumOperands(pAsm); + + GLboolean bSplitInst = GL_FALSE; + + if (1 == pAsm->D.dst.math) { - if (pILInst->TexSrcTarget == TEXTURE_RECT_INDEX) { - if( GL_FALSE == assemble_tex_instruction(pAsm, GL_FALSE) ) - { - radeon_error("Error assembling TEX instruction\n"); - return GL_FALSE; - } - } else { - if( GL_FALSE == assemble_tex_instruction(pAsm, GL_TRUE) ) - { - radeon_error("Error assembling TEX instruction\n"); - return GL_FALSE; - } - } + is_single_scalar_operation = GL_TRUE; + number_of_scalar_operations = 1; } else - { //ALU - if( GL_FALSE == assemble_alu_instruction(pAsm) ) - { - radeon_error("Error assembling ALU instruction\n"); - return GL_FALSE; - } - } - - if(pAsm->D.dst.rtype == DST_REG_OUT) { - if(pAsm->D.dst.op3) - { - // There is no mask for OP3 instructions, so all channels are written - pAsm->pucOutMask[pAsm->D.dst.reg - pAsm->starting_export_register_number] = 0xF; - } - else - { - pAsm->pucOutMask[pAsm->D.dst.reg - pAsm->starting_export_register_number] - |= (unsigned char)pAsm->pILInst[pAsm->uiCurInst].DstReg.WriteMask; - } + is_single_scalar_operation = GL_FALSE; + number_of_scalar_operations = 4; } - - //reset for next inst. - pAsm->D.bits = 0; - pAsm->S[0].bits = 0; - pAsm->S[1].bits = 0; - pAsm->S[2].bits = 0; - pAsm->is_tex = GL_FALSE; - pAsm->need_tex_barrier = GL_FALSE; - return GL_TRUE; -} - -GLboolean assemble_math_function(r700_AssemblerBase* pAsm, BITS opcode) -{ - BITS tmp; - - checkop1(pAsm); - - tmp = gethelpr(pAsm); - - // opcode tmp.x, a.x - // MOV dst, tmp.x - pAsm->D.dst.opcode = opcode; - pAsm->D.dst.math = 1; - - setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); - pAsm->D.dst.rtype = DST_REG_TEMPORARY; - pAsm->D.dst.reg = tmp; - pAsm->D.dst.writex = 1; + contiguous_slots_needed = 0; - if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + if(GL_TRUE == is_reduction_opcode(&(pAsm->D)) ) { - return GL_FALSE; + contiguous_slots_needed = 4; } - if ( GL_FALSE == next_ins(pAsm) ) + initialize(pAsm); + + for (scalar_channel_index=0; + scalar_channel_index < number_of_scalar_operations; + scalar_channel_index++) { - return GL_FALSE; - } + R700ALUInstruction* alu_instruction_ptr = (R700ALUInstruction*) CALLOC_STRUCT(R700ALUInstruction); + if (alu_instruction_ptr == NULL) + { + return GL_FALSE; + } + Init_R700ALUInstruction(alu_instruction_ptr); + + //src 0 + current_source_index = 0; + pcurrent_source = &(pAsm->S[0].src); - // Now replicate result to all necessary channels in destination - pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + if (GL_FALSE == assemble_alu_src(alu_instruction_ptr, + current_source_index, + pcurrent_source, + scalar_channel_index) ) + { + return GL_FALSE; + } + + if (uNumSrc > 1) + { + // Process source 1 + current_source_index = 1; + pcurrent_source = &(pAsm->S[current_source_index].src); - if( GL_FALSE == assemble_dst(pAsm) ) - { - return GL_FALSE; - } + if (GL_FALSE == assemble_alu_src(alu_instruction_ptr, + current_source_index, + pcurrent_source, + scalar_channel_index) ) + { + return GL_FALSE; + } + } - setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); - pAsm->S[0].src.rtype = DST_REG_TEMPORARY; - pAsm->S[0].src.reg = tmp; + //other bits + alu_instruction_ptr->m_Word0.f.index_mode = SQ_INDEX_LOOP; - setswizzle_PVSSRC(&(pAsm->S[0].src), SQ_SEL_X); - noneg_PVSSRC(&(pAsm->S[0].src)); + if( (is_single_scalar_operation == GL_TRUE) + || (GL_TRUE == bSplitInst) ) + { + alu_instruction_ptr->m_Word0.f.last = 1; + } + else + { + alu_instruction_ptr->m_Word0.f.last = (scalar_channel_index == 3) ? 1 : 0; + } - if( GL_FALSE == next_ins(pAsm) ) - { - return GL_FALSE; - } + alu_instruction_ptr->m_Word0.f.pred_sel = (pAsm->D.dst.pred_inv > 0) ? 1 : 0; + if(1 == pAsm->D.dst.predicated) + { + alu_instruction_ptr->m_Word1_OP2.f.update_pred = 0x1; + alu_instruction_ptr->m_Word1_OP2.f.update_execute_mask = 0x1; + } + else + { + alu_instruction_ptr->m_Word1_OP2.f.update_pred = 0x0; + alu_instruction_ptr->m_Word1_OP2.f.update_execute_mask = 0x0; + } + + // dst + if( (pAsm->D.dst.rtype == DST_REG_TEMPORARY) || + (pAsm->D.dst.rtype == DST_REG_OUT) ) + { + alu_instruction_ptr->m_Word1.f.dst_gpr = pAsm->D.dst.reg; + } + else + { + radeon_error("Only temp destination registers supported for ALU dest regs.\n"); + return GL_FALSE; + } - return GL_TRUE; -} + alu_instruction_ptr->m_Word1.f.dst_rel = SQ_ABSOLUTE; //D.rtype -GLboolean assemble_ABS(r700_AssemblerBase *pAsm) -{ - checkop1(pAsm); + if ( is_single_scalar_operation == GL_TRUE ) + { + // Override scalar_channel_index since only one scalar value will be written + if(pAsm->D.dst.writex) + { + scalar_channel_index = 0; + } + else if(pAsm->D.dst.writey) + { + scalar_channel_index = 1; + } + else if(pAsm->D.dst.writez) + { + scalar_channel_index = 2; + } + else if(pAsm->D.dst.writew) + { + scalar_channel_index = 3; + } + } - pAsm->D.dst.opcode = SQ_OP2_INST_MAX; + alu_instruction_ptr->m_Word1.f.dst_chan = scalar_channel_index; - if( GL_FALSE == assemble_dst(pAsm) ) - { - return GL_FALSE; - } - if( GL_FALSE == assemble_src(pAsm, 0, -1) ) - { - return GL_FALSE; - } - - pAsm->S[1].bits = pAsm->S[0].bits; - flipneg_PVSSRC(&(pAsm->S[1].src)); + alu_instruction_ptr->m_Word1.f.clamp = pAsm->D2.dst2.SaturateMode; - if ( GL_FALSE == next_ins(pAsm) ) - { - return GL_FALSE; - } + if (pAsm->D.dst.op3) + { + //op3 - return GL_TRUE; -} + alu_instruction_ptr->m_Word1_OP3.f.alu_inst = pAsm->D.dst.opcode; -GLboolean assemble_ADD(r700_AssemblerBase *pAsm) -{ - if( GL_FALSE == checkop2(pAsm) ) - { - return GL_FALSE; - } + //There's 3rd src for op3 + current_source_index = 2; + pcurrent_source = &(pAsm->S[current_source_index].src); - pAsm->D.dst.opcode = SQ_OP2_INST_ADD; - + if ( GL_FALSE == assemble_alu_src(alu_instruction_ptr, + current_source_index, + pcurrent_source, + scalar_channel_index) ) + { + return GL_FALSE; + } + } + else + { + //op2 + if (pAsm->bR6xx) + { + alu_instruction_ptr->m_Word1_OP2.f6.alu_inst = pAsm->D.dst.opcode; + + alu_instruction_ptr->m_Word1_OP2.f6.src0_abs = 0x0; + alu_instruction_ptr->m_Word1_OP2.f6.src1_abs = 0x0; + + //alu_instruction_ptr->m_Word1_OP2.f6.update_execute_mask = 0x0; + //alu_instruction_ptr->m_Word1_OP2.f6.update_pred = 0x0; + switch (scalar_channel_index) + { + case 0: + alu_instruction_ptr->m_Word1_OP2.f6.write_mask = pAsm->D.dst.writex; + break; + case 1: + alu_instruction_ptr->m_Word1_OP2.f6.write_mask = pAsm->D.dst.writey; + break; + case 2: + alu_instruction_ptr->m_Word1_OP2.f6.write_mask = pAsm->D.dst.writez; + break; + case 3: + alu_instruction_ptr->m_Word1_OP2.f6.write_mask = pAsm->D.dst.writew; + break; + default: + alu_instruction_ptr->m_Word1_OP2.f6.write_mask = 1; //SQ_SEL_MASK; + break; + } + alu_instruction_ptr->m_Word1_OP2.f6.omod = SQ_ALU_OMOD_OFF; + } + else + { + alu_instruction_ptr->m_Word1_OP2.f.alu_inst = pAsm->D.dst.opcode; + + alu_instruction_ptr->m_Word1_OP2.f.src0_abs = 0x0; + alu_instruction_ptr->m_Word1_OP2.f.src1_abs = 0x0; + + //alu_instruction_ptr->m_Word1_OP2.f.update_execute_mask = 0x0; + //alu_instruction_ptr->m_Word1_OP2.f.update_pred = 0x0; + switch (scalar_channel_index) + { + case 0: + alu_instruction_ptr->m_Word1_OP2.f.write_mask = pAsm->D.dst.writex; + break; + case 1: + alu_instruction_ptr->m_Word1_OP2.f.write_mask = pAsm->D.dst.writey; + break; + case 2: + alu_instruction_ptr->m_Word1_OP2.f.write_mask = pAsm->D.dst.writez; + break; + case 3: + alu_instruction_ptr->m_Word1_OP2.f.write_mask = pAsm->D.dst.writew; + break; + default: + alu_instruction_ptr->m_Word1_OP2.f.write_mask = 1; //SQ_SEL_MASK; + break; + } + alu_instruction_ptr->m_Word1_OP2.f.omod = SQ_ALU_OMOD_OFF; + } + } + + if(GL_FALSE == add_alu_instruction(pAsm, alu_instruction_ptr, contiguous_slots_needed) ) + { + return GL_FALSE; + } + + /* + * Judge the type of current instruction, is it vector or scalar + * instruction. + */ + if (is_single_scalar_operation) + { + if(GL_FALSE == check_scalar(pAsm, alu_instruction_ptr) ) + { + return GL_FALSE; + } + } + else + { + if(GL_FALSE == check_vector(pAsm, alu_instruction_ptr) ) + { + return 1; + } + } + + contiguous_slots_needed = 0; + } + + return GL_TRUE; +} + +GLboolean assemble_alu_instruction_literal(r700_AssemblerBase *pAsm, GLfloat * pLiteral) +{ + R700ALUInstruction * alu_instruction_ptr; + R700ALUInstructionHalfLiteral * alu_instruction_ptr_hl; + R700ALUInstructionFullLiteral * alu_instruction_ptr_fl; + + GLuint number_of_scalar_operations; + GLboolean is_single_scalar_operation; + GLuint scalar_channel_index; + + GLuint contiguous_slots_needed; + GLuint lastInstruction; + GLuint not_masked[4]; + + GLuint uNumSrc = r700GetNumOperands(pAsm); + + GLboolean bSplitInst = GL_FALSE; + + number_of_scalar_operations = 0; + contiguous_slots_needed = 0; + + if(1 == pAsm->D.dst.writew) + { + lastInstruction = 3; + number_of_scalar_operations++; + not_masked[3] = 1; + } + else + { + not_masked[3] = 0; + } + if(1 == pAsm->D.dst.writez) + { + lastInstruction = 2; + number_of_scalar_operations++; + not_masked[2] = 1; + } + else + { + not_masked[2] = 0; + } + if(1 == pAsm->D.dst.writey) + { + lastInstruction = 1; + number_of_scalar_operations++; + not_masked[1] = 1; + } + else + { + not_masked[1] = 0; + } + if(1 == pAsm->D.dst.writex) + { + lastInstruction = 0; + number_of_scalar_operations++; + not_masked[0] = 1; + } + else + { + not_masked[0] = 0; + } + + if(GL_TRUE == is_reduction_opcode(&(pAsm->D)) ) + { + contiguous_slots_needed = 4; + } + else + { + contiguous_slots_needed = number_of_scalar_operations; + } + + if(1 == pAsm->D2.dst2.literal) + { + contiguous_slots_needed += 1; + } + else if(2 == pAsm->D2.dst2.literal) + { + contiguous_slots_needed += 2; + } + + initialize(pAsm); + + for (scalar_channel_index=0; scalar_channel_index < 4; scalar_channel_index++) + { + if(0 == not_masked[scalar_channel_index]) + { + continue; + } + + if(scalar_channel_index == lastInstruction) + { + switch (pAsm->D2.dst2.literal) + { + case 0: + alu_instruction_ptr = (R700ALUInstruction*) CALLOC_STRUCT(R700ALUInstruction); + if (alu_instruction_ptr == NULL) + { + return GL_FALSE; + } + Init_R700ALUInstruction(alu_instruction_ptr); + break; + case 1: + alu_instruction_ptr_hl = (R700ALUInstructionHalfLiteral*) CALLOC_STRUCT(R700ALUInstructionHalfLiteral); + if (alu_instruction_ptr_hl == NULL) + { + return GL_FALSE; + } + Init_R700ALUInstructionHalfLiteral(alu_instruction_ptr_hl, pLiteral[0], pLiteral[1]); + alu_instruction_ptr = (R700ALUInstruction*)alu_instruction_ptr_hl; + break; + case 2: + alu_instruction_ptr_fl = (R700ALUInstructionFullLiteral*) CALLOC_STRUCT(R700ALUInstructionFullLiteral); + if (alu_instruction_ptr_fl == NULL) + { + return GL_FALSE; + } + Init_R700ALUInstructionFullLiteral(alu_instruction_ptr_fl, pLiteral[0], pLiteral[1], pLiteral[2], pLiteral[3]); + alu_instruction_ptr = (R700ALUInstruction*)alu_instruction_ptr_fl; + break; + default: + break; + }; + } + else + { + alu_instruction_ptr = (R700ALUInstruction*) CALLOC_STRUCT(R700ALUInstruction); + if (alu_instruction_ptr == NULL) + { + return GL_FALSE; + } + Init_R700ALUInstruction(alu_instruction_ptr); + } + + //src 0 + if (GL_FALSE == assemble_alu_src(alu_instruction_ptr, + 0, + &(pAsm->S[0].src), + scalar_channel_index) ) + { + return GL_FALSE; + } + + if (uNumSrc > 1) + { + // Process source 1 + if (GL_FALSE == assemble_alu_src(alu_instruction_ptr, + 1, + &(pAsm->S[1].src), + scalar_channel_index) ) + { + return GL_FALSE; + } + } + + //other bits + alu_instruction_ptr->m_Word0.f.index_mode = SQ_INDEX_LOOP; + + if(scalar_channel_index == lastInstruction) + { + alu_instruction_ptr->m_Word0.f.last = 1; + } + + alu_instruction_ptr->m_Word0.f.pred_sel = 0x0; + if(1 == pAsm->D.dst.predicated) + { + alu_instruction_ptr->m_Word1_OP2.f.update_pred = 0x1; + alu_instruction_ptr->m_Word1_OP2.f.update_execute_mask = 0x1; + } + else + { + alu_instruction_ptr->m_Word1_OP2.f.update_pred = 0; + alu_instruction_ptr->m_Word1_OP2.f.update_execute_mask = 0; + } + + // dst + if( (pAsm->D.dst.rtype == DST_REG_TEMPORARY) || + (pAsm->D.dst.rtype == DST_REG_OUT) ) + { + alu_instruction_ptr->m_Word1.f.dst_gpr = pAsm->D.dst.reg; + } + else + { + radeon_error("Only temp destination registers supported for ALU dest regs.\n"); + return GL_FALSE; + } + + alu_instruction_ptr->m_Word1.f.dst_rel = SQ_ABSOLUTE; //D.rtype + + alu_instruction_ptr->m_Word1.f.dst_chan = scalar_channel_index; + + alu_instruction_ptr->m_Word1.f.clamp = pAsm->D2.dst2.SaturateMode; + + if (pAsm->D.dst.op3) + { + //op3 + alu_instruction_ptr->m_Word1_OP3.f.alu_inst = pAsm->D.dst.opcode; + + //There's 3rd src for op3 + if ( GL_FALSE == assemble_alu_src(alu_instruction_ptr, + 2, + &(pAsm->S[2].src), + scalar_channel_index) ) + { + return GL_FALSE; + } + } + else + { + //op2 + if (pAsm->bR6xx) + { + alu_instruction_ptr->m_Word1_OP2.f6.alu_inst = pAsm->D.dst.opcode; + alu_instruction_ptr->m_Word1_OP2.f6.src0_abs = 0x0; + alu_instruction_ptr->m_Word1_OP2.f6.src1_abs = 0x0; + alu_instruction_ptr->m_Word1_OP2.f6.write_mask = 1; + alu_instruction_ptr->m_Word1_OP2.f6.omod = SQ_ALU_OMOD_OFF; + } + else + { + alu_instruction_ptr->m_Word1_OP2.f.alu_inst = pAsm->D.dst.opcode; + alu_instruction_ptr->m_Word1_OP2.f.src0_abs = 0x0; + alu_instruction_ptr->m_Word1_OP2.f.src1_abs = 0x0; + alu_instruction_ptr->m_Word1_OP2.f.write_mask = 1; + alu_instruction_ptr->m_Word1_OP2.f.omod = SQ_ALU_OMOD_OFF; + } + } + + if(GL_FALSE == add_alu_instruction(pAsm, alu_instruction_ptr, contiguous_slots_needed) ) + { + return GL_FALSE; + } + + if (1 == number_of_scalar_operations) + { + if(GL_FALSE == check_scalar(pAsm, alu_instruction_ptr) ) + { + return GL_FALSE; + } + } + else + { + if(GL_FALSE == check_vector(pAsm, alu_instruction_ptr) ) + { + return GL_FALSE; + } + } + + contiguous_slots_needed -= 2; + } + + return GL_TRUE; +} + +GLboolean next_ins(r700_AssemblerBase *pAsm) +{ + struct prog_instruction *pILInst = &(pAsm->pILInst[pAsm->uiCurInst]); + + if( GL_TRUE == pAsm->is_tex ) + { + if (pILInst->TexSrcTarget == TEXTURE_RECT_INDEX) { + if( GL_FALSE == assemble_tex_instruction(pAsm, GL_FALSE) ) + { + radeon_error("Error assembling TEX instruction\n"); + return GL_FALSE; + } + } else { + if( GL_FALSE == assemble_tex_instruction(pAsm, GL_TRUE) ) + { + radeon_error("Error assembling TEX instruction\n"); + return GL_FALSE; + } + } + } + else + { //ALU + if( GL_FALSE == assemble_alu_instruction(pAsm) ) + { + radeon_error("Error assembling ALU instruction\n"); + return GL_FALSE; + } + } + + if(pAsm->D.dst.rtype == DST_REG_OUT) + { + if(pAsm->D.dst.op3) + { + // There is no mask for OP3 instructions, so all channels are written + pAsm->pucOutMask[pAsm->D.dst.reg - pAsm->starting_export_register_number] = 0xF; + } + else + { + pAsm->pucOutMask[pAsm->D.dst.reg - pAsm->starting_export_register_number] + |= (unsigned char)pAsm->pILInst[pAsm->uiCurInst].DstReg.WriteMask; + } + } + + //reset for next inst. + pAsm->D.bits = 0; + pAsm->D2.bits = 0; + pAsm->S[0].bits = 0; + pAsm->S[1].bits = 0; + pAsm->S[2].bits = 0; + pAsm->is_tex = GL_FALSE; + pAsm->need_tex_barrier = GL_FALSE; + + return GL_TRUE; +} + +GLboolean next_ins2(r700_AssemblerBase *pAsm) +{ + struct prog_instruction *pILInst = &(pAsm->pILInst[pAsm->uiCurInst]); + + //ALU + if( GL_FALSE == assemble_alu_instruction2(pAsm) ) + { + radeon_error("Error assembling ALU instruction\n"); + return GL_FALSE; + } + + if(pAsm->D.dst.rtype == DST_REG_OUT) + { + if(pAsm->D.dst.op3) + { + // There is no mask for OP3 instructions, so all channels are written + pAsm->pucOutMask[pAsm->D.dst.reg - pAsm->starting_export_register_number] = 0xF; + } + else + { + pAsm->pucOutMask[pAsm->D.dst.reg - pAsm->starting_export_register_number] + |= (unsigned char)pAsm->pILInst[pAsm->uiCurInst].DstReg.WriteMask; + } + } + + //reset for next inst. + pAsm->D.bits = 0; + pAsm->D2.bits = 0; + pAsm->S[0].bits = 0; + pAsm->S[1].bits = 0; + pAsm->S[2].bits = 0; + pAsm->is_tex = GL_FALSE; + pAsm->need_tex_barrier = GL_FALSE; + + //richard nov.16 glsl + pAsm->D2.bits = 0; + + return GL_TRUE; +} + +/* not work yet */ +GLboolean next_ins_literal(r700_AssemblerBase *pAsm, GLfloat * pLiteral) +{ + struct prog_instruction *pILInst = &(pAsm->pILInst[pAsm->uiCurInst]); + + //ALU + if( GL_FALSE == assemble_alu_instruction_literal(pAsm, pLiteral) ) + { + radeon_error("Error assembling ALU instruction\n"); + return GL_FALSE; + } + + //reset for next inst. + pAsm->D.bits = 0; + pAsm->D2.bits = 0; + pAsm->S[0].bits = 0; + pAsm->S[1].bits = 0; + pAsm->S[2].bits = 0; + pAsm->is_tex = GL_FALSE; + pAsm->need_tex_barrier = GL_FALSE; + return GL_TRUE; +} + +GLboolean assemble_math_function(r700_AssemblerBase* pAsm, BITS opcode) +{ + BITS tmp; + + checkop1(pAsm); + + tmp = gethelpr(pAsm); + + // opcode tmp.x, a.x + // MOV dst, tmp.x + + pAsm->D.dst.opcode = opcode; + pAsm->D.dst.math = 1; + + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + pAsm->D.dst.writex = 1; + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + + if ( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + // Now replicate result to all necessary channels in destination + pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + + if( GL_FALSE == assemble_dst(pAsm) ) + { + return GL_FALSE; + } + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = DST_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp; + + setswizzle_PVSSRC(&(pAsm->S[0].src), SQ_SEL_X); + noneg_PVSSRC(&(pAsm->S[0].src)); + + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + return GL_TRUE; +} + +GLboolean assemble_ABS(r700_AssemblerBase *pAsm) +{ + checkop1(pAsm); + + pAsm->D.dst.opcode = SQ_OP2_INST_MAX; + + if( GL_FALSE == assemble_dst(pAsm) ) + { + return GL_FALSE; + } + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + + pAsm->S[1].bits = pAsm->S[0].bits; + flipneg_PVSSRC(&(pAsm->S[1].src)); + + if ( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + return GL_TRUE; +} + +GLboolean assemble_ADD(r700_AssemblerBase *pAsm) +{ + if( GL_FALSE == checkop2(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = SQ_OP2_INST_ADD; + if( GL_FALSE == assemble_dst(pAsm) ) { return GL_FALSE; @@ -3809,10 +4446,78 @@ GLboolean assemble_SCS(r700_AssemblerBase *pAsm) pAsm->S[0].src.swizzlez = SQ_SEL_0; pAsm->S[0].src.swizzlew = SQ_SEL_0; - if ( GL_FALSE == next_ins(pAsm) ) - { - return GL_FALSE; - } + if ( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + return GL_TRUE; +} + +GLboolean assemble_LOGIC(r700_AssemblerBase *pAsm, BITS opcode) +{ + if( GL_FALSE == checkop2(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = opcode; + pAsm->D.dst.math = 1; + + if( GL_FALSE == assemble_dst(pAsm) ) + { + return GL_FALSE; + } + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + + if( GL_FALSE == assemble_src(pAsm, 1, -1) ) + { + return GL_FALSE; + } + + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + return GL_TRUE; +} + +GLboolean assemble_LOGIC_PRED(r700_AssemblerBase *pAsm, BITS opcode) +{ + if( GL_FALSE == checkop2(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = opcode; + pAsm->D.dst.math = 1; + pAsm->D.dst.predicated = 1; + pAsm->D2.dst2.SaturateMode = pAsm->pILInst[pAsm->uiCurInst].SaturateMode; + + if( GL_FALSE == assemble_dst(pAsm) ) + { + return GL_FALSE; + } + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + + if( GL_FALSE == assemble_src(pAsm, 1, -1) ) + { + return GL_FALSE; + } + + if( GL_FALSE == next_ins2(pAsm) ) + { + return GL_FALSE; + } return GL_TRUE; } @@ -4077,223 +4782,930 @@ GLboolean assemble_TEX(r700_AssemblerBase *pAsm) pAsm->D.dst.writez = 0; pAsm->D.dst.writew = 0; - setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); - pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; - pAsm->S[0].src.reg = tmp2; - noswizzle_PVSSRC(&(pAsm->S[0].src)); - setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE); - pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; - pAsm->S[1].src.reg = 252; // SQ_ALU_SRC_0_5 - noswizzle_PVSSRC(&(pAsm->S[1].src)); + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp2; + noswizzle_PVSSRC(&(pAsm->S[0].src)); + setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE); + pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[1].src.reg = 252; // SQ_ALU_SRC_0_5 + noswizzle_PVSSRC(&(pAsm->S[1].src)); + + next_ins(pAsm); + + /* tmp1.xy = temp2.xy */ + pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp1; + pAsm->D.dst.writex = 1; + pAsm->D.dst.writey = 1; + pAsm->D.dst.writez = 0; + pAsm->D.dst.writew = 0; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp2; + noswizzle_PVSSRC(&(pAsm->S[0].src)); + + next_ins(pAsm); + pAsm->aArgSubst[1] = tmp1; + need_barrier = GL_TRUE; + + } + + pAsm->D.dst.opcode = SQ_TEX_INST_SAMPLE; + pAsm->is_tex = GL_TRUE; + if ( GL_TRUE == need_barrier ) + { + pAsm->need_tex_barrier = GL_TRUE; + } + // Set src1 to tex unit id + pAsm->S[1].src.reg = pAsm->pILInst[pAsm->uiCurInst].TexSrcUnit; + pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; + + //No sw info from mesa compiler, so hard code here. + pAsm->S[1].src.swizzlex = SQ_SEL_X; + pAsm->S[1].src.swizzley = SQ_SEL_Y; + pAsm->S[1].src.swizzlez = SQ_SEL_Z; + pAsm->S[1].src.swizzlew = SQ_SEL_W; + + if( GL_FALSE == tex_dst(pAsm) ) + { + return GL_FALSE; + } + + if( GL_FALSE == tex_src(pAsm) ) + { + return GL_FALSE; + } + + if(pAsm->pILInst[pAsm->uiCurInst].Opcode == OPCODE_TXP) + { + /* hopefully did swizzles before */ + noswizzle_PVSSRC(&(pAsm->S[0].src)); + } + + if(pAsm->pILInst[pAsm->uiCurInst].TexSrcTarget == TEXTURE_CUBE_INDEX) + { + /* SAMPLE dst, tmp.yxwy, CUBE */ + pAsm->S[0].src.swizzlex = SQ_SEL_Y; + pAsm->S[0].src.swizzley = SQ_SEL_X; + pAsm->S[0].src.swizzlez = SQ_SEL_W; + pAsm->S[0].src.swizzlew = SQ_SEL_Y; + } + + if ( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + return GL_TRUE; +} + +GLboolean assemble_XPD(r700_AssemblerBase *pAsm) +{ + BITS tmp; + + if( GL_FALSE == checkop2(pAsm) ) + { + return GL_FALSE; + } + + tmp = gethelpr(pAsm); + + pAsm->D.dst.opcode = SQ_OP2_INST_MUL; + + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + nomask_PVSDST(&(pAsm->D.dst)); + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + + if( GL_FALSE == assemble_src(pAsm, 1, -1) ) + { + return GL_FALSE; + } + + swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_Z, SQ_SEL_X, SQ_SEL_Y, SQ_SEL_0); + swizzleagain_PVSSRC(&(pAsm->S[1].src), SQ_SEL_Y, SQ_SEL_Z, SQ_SEL_X, SQ_SEL_0); + + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = SQ_OP3_INST_MULADD; + pAsm->D.dst.op3 = 1; + + if(0xF != pAsm->pILInst[pAsm->uiCurInst].DstReg.WriteMask) + { + tmp = gethelpr(pAsm); + + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + + nomask_PVSDST(&(pAsm->D.dst)); + } + else + { + if( GL_FALSE == assemble_dst(pAsm) ) + { + return GL_FALSE; + } + } + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + + if( GL_FALSE == assemble_src(pAsm, 1, -1) ) + { + return GL_FALSE; + } + + swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_Y, SQ_SEL_Z, SQ_SEL_X, SQ_SEL_0); + swizzleagain_PVSSRC(&(pAsm->S[1].src), SQ_SEL_Z, SQ_SEL_X, SQ_SEL_Y, SQ_SEL_0); + + // result1 + (neg) result0 + setaddrmode_PVSSRC(&(pAsm->S[2].src),ADDR_ABSOLUTE); + pAsm->S[2].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[2].src.reg = tmp; + + neg_PVSSRC(&(pAsm->S[2].src)); + noswizzle_PVSSRC(&(pAsm->S[2].src)); + + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + + if(0xF != pAsm->pILInst[pAsm->uiCurInst].DstReg.WriteMask) + { + if( GL_FALSE == assemble_dst(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + + // Use tmp as source + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp; + + noneg_PVSSRC(&(pAsm->S[0].src)); + noswizzle_PVSSRC(&(pAsm->S[0].src)); + + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + } + + return GL_TRUE; +} + +GLboolean assemble_EXPORT(r700_AssemblerBase *pAsm) +{ + return GL_TRUE; +} + +GLboolean jumpToOffest(r700_AssemblerBase *pAsm, GLuint pops, GLint offset) +{ + if(GL_FALSE == add_cf_instruction(pAsm) ) + { + return GL_FALSE; + } + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = pops; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_JUMP; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + pAsm->cf_current_cf_clause_ptr->m_Word0.f.addr = pAsm->cf_current_cf_clause_ptr->m_uIndex + offset; + + return GL_TRUE; +} + +GLboolean pops(r700_AssemblerBase *pAsm, GLuint pops) +{ + if(GL_FALSE == add_cf_instruction(pAsm) ) + { + return GL_FALSE; + } + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = pops; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_POP; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + pAsm->cf_current_cf_clause_ptr->m_Word0.f.addr = pAsm->cf_current_cf_clause_ptr->m_uIndex + 1; + + return GL_TRUE; +} + +GLboolean assemble_IF(r700_AssemblerBase *pAsm) +{ + if(GL_FALSE == add_cf_instruction(pAsm) ) + { + return GL_FALSE; + } + + if(GL_TRUE != bHasElse) + { + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; + } + else + { + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 0; + } + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_JUMP; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + pAsm->FCSP++; + pAsm->fc_stack[pAsm->FCSP].type = FC_IF; + pAsm->fc_stack[pAsm->FCSP].bpush = 0; + pAsm->fc_stack[pAsm->FCSP].mid = NULL; + pAsm->fc_stack[pAsm->FCSP].midLen= 0; + pAsm->fc_stack[pAsm->FCSP].first = pAsm->cf_current_cf_clause_ptr; + + if(GL_TRUE != bHasElse) + { + pAsm->alu_x_opcode = SQ_CF_INST_ALU_POP_AFTER; + } + + pAsm->branch_depth++; + + if(pAsm->branch_depth > pAsm->max_branch_depth) + { + pAsm->max_branch_depth = pAsm->branch_depth; + } + return GL_TRUE; +} + +GLboolean assemble_ELSE(r700_AssemblerBase *pAsm) +{ + if(GL_FALSE == add_cf_instruction(pAsm) ) + { + return GL_FALSE; + } + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; /// + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; - next_ins(pAsm); + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_ELSE; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; - /* tmp1.xy = temp2.xy */ - pAsm->D.dst.opcode = SQ_OP2_INST_MOV; - setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); - pAsm->D.dst.rtype = DST_REG_TEMPORARY; - pAsm->D.dst.reg = tmp1; - pAsm->D.dst.writex = 1; - pAsm->D.dst.writey = 1; - pAsm->D.dst.writez = 0; - pAsm->D.dst.writew = 0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; - setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); - pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; - pAsm->S[0].src.reg = tmp2; - noswizzle_PVSSRC(&(pAsm->S[0].src)); + pAsm->fc_stack[pAsm->FCSP].mid = (R700ControlFlowGenericClause **)_mesa_realloc( (void *)pAsm->fc_stack[pAsm->FCSP].mid, + 0, + sizeof(R700ControlFlowGenericClause *) ); + pAsm->fc_stack[pAsm->FCSP].mid[0] = pAsm->cf_current_cf_clause_ptr; + //pAsm->fc_stack[pAsm->FCSP].unNumMid = 1; - next_ins(pAsm); - pAsm->aArgSubst[1] = tmp1; - need_barrier = GL_TRUE; + pAsm->alu_x_opcode = SQ_CF_INST_ALU_POP_AFTER; + + pAsm->fc_stack[pAsm->FCSP].first->m_Word0.f.addr = pAsm->pR700Shader->plstCFInstructions_active->uNumOfNode - 1; + + return GL_TRUE; +} + +GLboolean assemble_ENDIF(r700_AssemblerBase *pAsm) +{ + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + if(NULL == pAsm->fc_stack[pAsm->FCSP].mid) + { + /* no else in between */ + pAsm->fc_stack[pAsm->FCSP].first->m_Word0.f.addr = pAsm->pR700Shader->plstCFInstructions_active->uNumOfNode; + } + else + { + pAsm->fc_stack[pAsm->FCSP].mid[0]->m_Word0.f.addr = pAsm->pR700Shader->plstCFInstructions_active->uNumOfNode; } - pAsm->D.dst.opcode = SQ_TEX_INST_SAMPLE; - pAsm->is_tex = GL_TRUE; - if ( GL_TRUE == need_barrier ) + if(NULL != pAsm->fc_stack[pAsm->FCSP].mid) { - pAsm->need_tex_barrier = GL_TRUE; + FREE(pAsm->fc_stack[pAsm->FCSP].mid); } - // Set src1 to tex unit id - pAsm->S[1].src.reg = pAsm->pILInst[pAsm->uiCurInst].TexSrcUnit; - pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; - //No sw info from mesa compiler, so hard code here. - pAsm->S[1].src.swizzlex = SQ_SEL_X; - pAsm->S[1].src.swizzley = SQ_SEL_Y; - pAsm->S[1].src.swizzlez = SQ_SEL_Z; - pAsm->S[1].src.swizzlew = SQ_SEL_W; + if(pAsm->fc_stack[pAsm->FCSP].type != FC_IF) + { + radeon_error("if/endif in shader code are not paired. \n"); + return GL_FALSE; + } + pAsm->branch_depth--; + pAsm->FCSP--; - if( GL_FALSE == tex_dst(pAsm) ) + return GL_TRUE; +} + +GLboolean assemble_BGNLOOP(r700_AssemblerBase *pAsm) +{ + if(GL_FALSE == add_cf_instruction(pAsm) ) { return GL_FALSE; } - if( GL_FALSE == tex_src(pAsm) ) + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_LOOP_START_NO_AL; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + pAsm->FCSP++; + pAsm->fc_stack[pAsm->FCSP].type = FC_LOOP; + pAsm->fc_stack[pAsm->FCSP].bpush = 1; + pAsm->fc_stack[pAsm->FCSP].mid = NULL; + pAsm->fc_stack[pAsm->FCSP].unNumMid = 0; + pAsm->fc_stack[pAsm->FCSP].midLen = 0; + pAsm->fc_stack[pAsm->FCSP].first = pAsm->cf_current_cf_clause_ptr; + + pAsm->branch_depth++; + + if(pAsm->branch_depth > pAsm->max_branch_depth) + { + pAsm->max_branch_depth = pAsm->branch_depth; + } + return GL_TRUE; +} + +GLboolean assemble_BRK(r700_AssemblerBase *pAsm) +{ +#ifdef USE_CF_FOR_CONTINUE_BREAK + unsigned int unFCSP; + for(unFCSP=pAsm->FCSP; unFCSP>0; unFCSP--) + { + if(FC_LOOP == pAsm->fc_stack[unFCSP].type) + { + break; + } + } + if(0 == FC_LOOP) { + radeon_error("Break is not inside loop/endloop pair.\n"); return GL_FALSE; } - if(pAsm->pILInst[pAsm->uiCurInst].Opcode == OPCODE_TXP) + if(GL_FALSE == add_cf_instruction(pAsm) ) { - /* hopefully did swizzles before */ - noswizzle_PVSSRC(&(pAsm->S[0].src)); + return GL_FALSE; } - - if(pAsm->pILInst[pAsm->uiCurInst].TexSrcTarget == TEXTURE_CUBE_INDEX) + + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_LOOP_BREAK; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + pAsm->fc_stack[unFCSP].mid = (R700ControlFlowGenericClause **)_mesa_realloc( + (void *)pAsm->fc_stack[unFCSP].mid, + sizeof(R700ControlFlowGenericClause *) * pAsm->fc_stack[unFCSP].unNumMid, + sizeof(R700ControlFlowGenericClause *) * (pAsm->fc_stack[unFCSP].unNumMid + 1) ); + pAsm->fc_stack[unFCSP].mid[pAsm->fc_stack[unFCSP].unNumMid] = pAsm->cf_current_cf_clause_ptr; + pAsm->fc_stack[unFCSP].unNumMid++; + + if(GL_FALSE == add_cf_instruction(pAsm) ) { - /* SAMPLE dst, tmp.yxwy, CUBE */ - pAsm->S[0].src.swizzlex = SQ_SEL_Y; - pAsm->S[0].src.swizzley = SQ_SEL_X; - pAsm->S[0].src.swizzlez = SQ_SEL_W; - pAsm->S[0].src.swizzlew = SQ_SEL_Y; + return GL_FALSE; } + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_POP; - if ( GL_FALSE == next_ins(pAsm) ) - { - return GL_FALSE; - } + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + pAsm->cf_current_cf_clause_ptr->m_Word0.f.addr = pAsm->cf_current_cf_clause_ptr->m_uIndex + 1; +#endif //USE_CF_FOR_CONTINUE_BREAK return GL_TRUE; } -GLboolean assemble_XPD(r700_AssemblerBase *pAsm) +GLboolean assemble_CONT(r700_AssemblerBase *pAsm) { - BITS tmp; +#ifdef USE_CF_FOR_CONTINUE_BREAK + unsigned int unFCSP; + for(unFCSP=pAsm->FCSP; unFCSP>0; unFCSP--) + { + if(FC_LOOP == pAsm->fc_stack[unFCSP].type) + { + break; + } + } + if(0 == FC_LOOP) + { + radeon_error("Continue is not inside loop/endloop pair.\n"); + return GL_FALSE; + } - if( GL_FALSE == checkop2(pAsm) ) + if(GL_FALSE == add_cf_instruction(pAsm) ) { - return GL_FALSE; + return GL_FALSE; } - tmp = gethelpr(pAsm); + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; - pAsm->D.dst.opcode = SQ_OP2_INST_MUL; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_LOOP_CONTINUE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; - setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); - pAsm->D.dst.rtype = DST_REG_TEMPORARY; - pAsm->D.dst.reg = tmp; - nomask_PVSDST(&(pAsm->D.dst)); - - if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + pAsm->fc_stack[unFCSP].mid = (R700ControlFlowGenericClause **)_mesa_realloc( + (void *)pAsm->fc_stack[unFCSP].mid, + sizeof(R700ControlFlowGenericClause *) * pAsm->fc_stack[unFCSP].unNumMid, + sizeof(R700ControlFlowGenericClause *) * (pAsm->fc_stack[unFCSP].unNumMid + 1) ); + pAsm->fc_stack[unFCSP].mid[pAsm->fc_stack[unFCSP].unNumMid] = pAsm->cf_current_cf_clause_ptr; + pAsm->fc_stack[unFCSP].unNumMid++; + + if(GL_FALSE == add_cf_instruction(pAsm) ) { return GL_FALSE; } - if( GL_FALSE == assemble_src(pAsm, 1, -1) ) + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_POP; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + pAsm->cf_current_cf_clause_ptr->m_Word0.f.addr = pAsm->cf_current_cf_clause_ptr->m_uIndex + 1; + +#endif /* USE_CF_FOR_CONTINUE_BREAK */ + + return GL_TRUE; +} + +GLboolean assemble_ENDLOOP(r700_AssemblerBase *pAsm) +{ + GLuint i; + + if(GL_FALSE == add_cf_instruction(pAsm) ) { return GL_FALSE; } - - swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_Z, SQ_SEL_X, SQ_SEL_Y, SQ_SEL_0); - swizzleagain_PVSSRC(&(pAsm->S[1].src), SQ_SEL_Y, SQ_SEL_Z, SQ_SEL_X, SQ_SEL_0); - if( GL_FALSE == next_ins(pAsm) ) + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_LOOP_END; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + pAsm->cf_current_cf_clause_ptr->m_Word0.f.addr = pAsm->fc_stack[pAsm->FCSP].first->m_uIndex + 1; + pAsm->fc_stack[pAsm->FCSP].first->m_Word0.f.addr = pAsm->cf_current_cf_clause_ptr->m_uIndex + 1; + +#ifdef USE_CF_FOR_CONTINUE_BREAK + for(i=0; ifc_stack[pAsm->FCSP].unNumMid; i++) + { + pAsm->fc_stack[pAsm->FCSP].mid[i]->m_Word0.f.addr = pAsm->cf_current_cf_clause_ptr->m_uIndex; + } + if(NULL != pAsm->fc_stack[pAsm->FCSP].mid) + { + FREE(pAsm->fc_stack[pAsm->FCSP].mid); + } +#endif + + if(pAsm->fc_stack[pAsm->FCSP].type != FC_LOOP) { + radeon_error("loop/endloop in shader code are not paired. \n"); return GL_FALSE; } - pAsm->D.dst.opcode = SQ_OP3_INST_MULADD; - pAsm->D.dst.op3 = 1; + unsigned int unFCSP = 0; + if((pAsm->unCFflags & HAS_CURRENT_LOOPRET) > 0) + { + for(unFCSP=(pAsm->FCSP-1); unFCSP>pAsm->CALLSTACK[pAsm->CALLSP].FCSP_BeforeEntry; unFCSP--) + { + if(FC_LOOP == pAsm->fc_stack[unFCSP].type) + { + break; + } + } + if(unFCSP <= pAsm->CALLSTACK[pAsm->CALLSP].FCSP_BeforeEntry) + { + unFCSP = 0; - if(0xF != pAsm->pILInst[pAsm->uiCurInst].DstReg.WriteMask) - { - tmp = gethelpr(pAsm); + returnOnFlag(pAsm); + pAsm->unCFflags &= ~HAS_CURRENT_LOOPRET; + } + } - setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); - pAsm->D.dst.rtype = DST_REG_TEMPORARY; - pAsm->D.dst.reg = tmp; + pAsm->branch_depth--; + pAsm->FCSP--; - nomask_PVSDST(&(pAsm->D.dst)); + if(unFCSP > 0) + { + breakLoopOnFlag(pAsm, unFCSP); } - else + + return GL_TRUE; +} + +void add_return_inst(r700_AssemblerBase *pAsm) +{ + if(GL_FALSE == add_cf_instruction(pAsm) ) { - if( GL_FALSE == assemble_dst(pAsm) ) + return GL_FALSE; + } + //pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_RETURN; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; +} + +GLboolean assemble_BGNSUB(r700_AssemblerBase *pAsm, GLint nILindex) +{ + /* Put in sub */ + if( (pAsm->unSubArrayPointer + 1) > pAsm->unSubArraySize ) + { + pAsm->subs = (SUB_OFFSET*)_mesa_realloc( (void *)pAsm->subs, + sizeof(SUB_OFFSET) * pAsm->unSubArraySize, + sizeof(SUB_OFFSET) * (pAsm->unSubArraySize + 10) ); + if(NULL == pAsm->subs) { return GL_FALSE; } + pAsm->unSubArraySize += 10; } - if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + pAsm->subs[pAsm->unSubArrayPointer].subIL_Offset = nILindex; + pAsm->subs[pAsm->unSubArrayPointer].lstCFInstructions_local.pHead=NULL; + pAsm->subs[pAsm->unSubArrayPointer].lstCFInstructions_local.pTail=NULL; + pAsm->subs[pAsm->unSubArrayPointer].lstCFInstructions_local.uNumOfNode=0; + + pAsm->CALLSP++; + pAsm->CALLSTACK[pAsm->CALLSP].FCSP_BeforeEntry = pAsm->FCSP; + pAsm->CALLSTACK[pAsm->CALLSP].plstCFInstructions_local + = &(pAsm->subs[pAsm->unSubArrayPointer].lstCFInstructions_local); + SetActiveCFlist(pAsm->pR700Shader, + pAsm->CALLSTACK[pAsm->CALLSP].plstCFInstructions_local); + + pAsm->unSubArrayPointer++; + + /* start sub */ + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + + return GL_TRUE; +} + +GLboolean assemble_ENDSUB(r700_AssemblerBase *pAsm) +{ + pAsm->CALLSP--; + SetActiveCFlist(pAsm->pR700Shader, + pAsm->CALLSTACK[pAsm->CALLSP].plstCFInstructions_local); + + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + + return GL_TRUE; +} + +GLboolean assemble_RET(r700_AssemblerBase *pAsm) +{ + if(pAsm->CALLSP > 0) + { /* in sub */ + unsigned int unFCSP; + for(unFCSP=pAsm->FCSP; unFCSP>pAsm->CALLSTACK[pAsm->CALLSP].FCSP_BeforeEntry; unFCSP--) + { + if(FC_LOOP == pAsm->fc_stack[unFCSP].type) + { + setRetInLoopFlag(pAsm, SQ_SEL_1); + breakLoopOnFlag(pAsm, unFCSP); + pAsm->unCFflags |= LOOPRET_FLAGS; + + return GL_TRUE; + } + } + } + + add_return_inst(pAsm); + + return GL_TRUE; +} + +GLboolean assemble_CAL(r700_AssemblerBase *pAsm, + GLint nILindex, + GLuint uiNumberInsts, + struct prog_instruction *pILInst) +{ + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + + if(GL_FALSE == add_cf_instruction(pAsm) ) { return GL_FALSE; } - if( GL_FALSE == assemble_src(pAsm, 1, -1) ) + pAsm->cf_current_cf_clause_ptr->m_Word1.f.call_count = 1; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_CALL; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + /* Put in caller */ + if( (pAsm->unCallerArrayPointer + 1) > pAsm->unCallerArraySize ) { - return GL_FALSE; + pAsm->callers = (CALLER_POINTER*)_mesa_realloc( (void *)pAsm->callers, + sizeof(CALLER_POINTER) * pAsm->unCallerArraySize, + sizeof(CALLER_POINTER) * (pAsm->unCallerArraySize + 10) ); + if(NULL == pAsm->callers) + { + return GL_FALSE; + } + pAsm->unCallerArraySize += 10; } - - swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_Y, SQ_SEL_Z, SQ_SEL_X, SQ_SEL_0); - swizzleagain_PVSSRC(&(pAsm->S[1].src), SQ_SEL_Z, SQ_SEL_X, SQ_SEL_Y, SQ_SEL_0); + + pAsm->callers[pAsm->unCallerArrayPointer].subIL_Offset = nILindex; + pAsm->callers[pAsm->unCallerArrayPointer].cf_ptr = pAsm->cf_current_cf_clause_ptr; - // result1 + (neg) result0 - setaddrmode_PVSSRC(&(pAsm->S[2].src),ADDR_ABSOLUTE); - pAsm->S[2].src.rtype = SRC_REG_TEMPORARY; - pAsm->S[2].src.reg = tmp; + pAsm->unCallerArrayPointer++; - neg_PVSSRC(&(pAsm->S[2].src)); - noswizzle_PVSSRC(&(pAsm->S[2].src)); + int j; + for(j=0; junSubArrayPointer; j++) + { + if(nILindex == pAsm->subs[j].subIL_Offset) + { /* compiled before */ + pAsm->callers[pAsm->unCallerArrayPointer - 1].subDescIndex = j; + return GL_TRUE; + } + } + + pAsm->callers[pAsm->unCallerArrayPointer - 1].subDescIndex = pAsm->unSubArrayPointer; + + return AssembleInstr(nILindex, uiNumberInsts, pILInst, pAsm); +} + +GLboolean setRetInLoopFlag(r700_AssemblerBase *pAsm, GLuint flagValue) +{ + GLfloat fLiteral[2] = {0.1, 0.0}; + + pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + pAsm->D.dst.op3 = 0; + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = pAsm->flag_reg_index; + pAsm->D.dst.writex = 1; + pAsm->D.dst.writey = 0; + pAsm->D.dst.writez = 0; + pAsm->D.dst.writew = 0; + pAsm->D2.dst2.literal = 1; + pAsm->D2.dst2.SaturateMode = SATURATE_OFF; + pAsm->D.dst.predicated = 0; +#if 0 + pAsm->S[0].src.rtype = SRC_REC_LITERAL; + //pAsm->S[0].src.reg = 0; + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + noneg_PVSSRC(&(pAsm->S[0].src)); + pAsm->S[0].src.swizzlex = SQ_SEL_X; + pAsm->S[0].src.swizzley = SQ_SEL_Y; + pAsm->S[0].src.swizzlez = SQ_SEL_Z; + pAsm->S[0].src.swizzlew = SQ_SEL_W; + + if( GL_FALSE == next_ins_literal(pAsm, &(fLiteral[0])) ) + { + return GL_FALSE; + } +#else + pAsm->S[0].src.rtype = DST_REG_TEMPORARY; + pAsm->S[0].src.reg = 0; + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + noneg_PVSSRC(&(pAsm->S[0].src)); + pAsm->S[0].src.swizzlex = flagValue; + pAsm->S[0].src.swizzley = flagValue; + pAsm->S[0].src.swizzlez = flagValue; + pAsm->S[0].src.swizzlew = flagValue; - if( GL_FALSE == next_ins(pAsm) ) + if( GL_FALSE == next_ins2(pAsm) ) { return GL_FALSE; } +#endif + return GL_TRUE; +} - if(0xF != pAsm->pILInst[pAsm->uiCurInst].DstReg.WriteMask) - { - if( GL_FALSE == assemble_dst(pAsm) ) - { - return GL_FALSE; - } +GLboolean testFlag(r700_AssemblerBase *pAsm) +{ + GLfloat fLiteral[2] = {0.1, 0.0}; - pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + //Test flag + GLuint tmp = gethelpr(pAsm); + pAsm->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; - // Use tmp as source - setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); - pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; - pAsm->S[0].src.reg = tmp; + pAsm->D.dst.opcode = SQ_OP2_INST_PRED_SETE; + pAsm->D.dst.math = 1; + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + pAsm->D.dst.writex = 1; + pAsm->D.dst.writey = 0; + pAsm->D.dst.writez = 0; + pAsm->D.dst.writew = 0; + pAsm->D2.dst2.literal = 1; + pAsm->D2.dst2.SaturateMode = SATURATE_OFF; + pAsm->D.dst.predicated = 1; - noneg_PVSSRC(&(pAsm->S[0].src)); - noswizzle_PVSSRC(&(pAsm->S[0].src)); + pAsm->S[0].src.rtype = DST_REG_TEMPORARY; + pAsm->S[0].src.reg = pAsm->flag_reg_index; + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + noneg_PVSSRC(&(pAsm->S[0].src)); + pAsm->S[0].src.swizzlex = SQ_SEL_X; + pAsm->S[0].src.swizzley = SQ_SEL_Y; + pAsm->S[0].src.swizzlez = SQ_SEL_Z; + pAsm->S[0].src.swizzlew = SQ_SEL_W; +#if 0 + pAsm->S[1].src.rtype = SRC_REC_LITERAL; + //pAsm->S[1].src.reg = 0; + setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE); + noneg_PVSSRC(&(pAsm->S[1].src)); + pAsm->S[1].src.swizzlex = SQ_SEL_X; + pAsm->S[1].src.swizzley = SQ_SEL_Y; + pAsm->S[1].src.swizzlez = SQ_SEL_Z; + pAsm->S[1].src.swizzlew = SQ_SEL_W; - if( GL_FALSE == next_ins(pAsm) ) - { - return GL_FALSE; - } + if( GL_FALSE == next_ins_literal(pAsm, &(fLiteral[0])) ) + { + return GL_FALSE; + } +#else + pAsm->S[1].src.rtype = DST_REG_TEMPORARY; + pAsm->S[1].src.reg = 0; + setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE); + noneg_PVSSRC(&(pAsm->S[1].src)); + pAsm->S[1].src.swizzlex = SQ_SEL_1; + pAsm->S[1].src.swizzley = SQ_SEL_1; + pAsm->S[1].src.swizzlez = SQ_SEL_1; + pAsm->S[1].src.swizzlew = SQ_SEL_1; + + if( GL_FALSE == next_ins2(pAsm) ) + { + return GL_FALSE; } +#endif return GL_TRUE; } -GLboolean assemble_EXPORT(r700_AssemblerBase *pAsm) +GLboolean returnOnFlag(r700_AssemblerBase *pAsm) { - return GL_TRUE; -} + testFlag(pAsm); + jumpToOffest(pAsm, 1, 4); + setRetInLoopFlag(pAsm, SQ_SEL_0); + pops(pAsm, 1); + add_return_inst(pAsm); -GLboolean assemble_IF(r700_AssemblerBase *pAsm) -{ return GL_TRUE; } -GLboolean assemble_ENDIF(r700_AssemblerBase *pAsm) +GLboolean breakLoopOnFlag(r700_AssemblerBase *pAsm, GLuint unFCSP) { + testFlag(pAsm); + + //break + if(GL_FALSE == add_cf_instruction(pAsm) ) + { + return GL_FALSE; + } + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.pop_count = 1; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_const = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cond = SQ_CF_COND_ACTIVE; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.end_of_program = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.valid_pixel_mode = 0x0; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.cf_inst = SQ_CF_INST_LOOP_BREAK; + pAsm->cf_current_cf_clause_ptr->m_Word1.f.whole_quad_mode = 0x0; + + pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; + + pAsm->fc_stack[unFCSP].mid = (R700ControlFlowGenericClause **)_mesa_realloc( + (void *)pAsm->fc_stack[unFCSP].mid, + sizeof(R700ControlFlowGenericClause *) * pAsm->fc_stack[unFCSP].unNumMid, + sizeof(R700ControlFlowGenericClause *) * (pAsm->fc_stack[unFCSP].unNumMid + 1) ); + pAsm->fc_stack[unFCSP].mid[pAsm->fc_stack[unFCSP].unNumMid] = pAsm->cf_current_cf_clause_ptr; + pAsm->fc_stack[unFCSP].unNumMid++; + + pops(pAsm, 1); + return GL_TRUE; } -GLboolean AssembleInstr(GLuint uiNumberInsts, +GLboolean AssembleInstr(GLuint uiFirstInst, + GLuint uiNumberInsts, struct prog_instruction *pILInst, r700_AssemblerBase *pR700AsmCode) { GLuint i; pR700AsmCode->pILInst = pILInst; - for(i=0; iuiCurInst = i; +#ifndef USE_CF_FOR_CONTINUE_BREAK + if(OPCODE_BRK == pILInst[i+1].Opcode) + { + switch(pILInst[i].Opcode) + { + case OPCODE_SLE: + pILInst[i].Opcode = OPCODE_SGT; + break; + case OPCODE_SLT: + pILInst[i].Opcode = OPCODE_SGE; + break; + case OPCODE_SGE: + pILInst[i].Opcode = OPCODE_SLT; + break; + case OPCODE_SGT: + pILInst[i].Opcode = OPCODE_SLE; + break; + case OPCODE_SEQ: + pILInst[i].Opcode = OPCODE_SNE; + break; + case OPCODE_SNE: + pILInst[i].Opcode = OPCODE_SEQ; + break; + default: + break; + } + } +#endif + switch (pILInst[i].Opcode) { case OPCODE_ABS: @@ -4337,101 +5749,383 @@ GLboolean AssembleInstr(GLuint uiNumberInsts, return GL_FALSE; break; - case OPCODE_EX2: - if ( GL_FALSE == assemble_EX2(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_EXP: - if ( GL_FALSE == assemble_EXP(pR700AsmCode) ) - return GL_FALSE; + case OPCODE_EX2: + if ( GL_FALSE == assemble_EX2(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_EXP: + if ( GL_FALSE == assemble_EXP(pR700AsmCode) ) + return GL_FALSE; + break; + + case OPCODE_FLR: + if ( GL_FALSE == assemble_FLR(pR700AsmCode) ) + return GL_FALSE; + break; + //case OP_FLR_INT: + // if ( GL_FALSE == assemble_FLR_INT() ) + // return GL_FALSE; + // break; + + case OPCODE_FRC: + if ( GL_FALSE == assemble_FRC(pR700AsmCode) ) + return GL_FALSE; + break; + + case OPCODE_KIL: + if ( GL_FALSE == assemble_KIL(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_LG2: + if ( GL_FALSE == assemble_LG2(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_LIT: + if ( GL_FALSE == assemble_LIT(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_LRP: + if ( GL_FALSE == assemble_LRP(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_LOG: + if ( GL_FALSE == assemble_LOG(pR700AsmCode) ) + return GL_FALSE; + break; + + case OPCODE_MAD: + if ( GL_FALSE == assemble_MAD(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_MAX: + if ( GL_FALSE == assemble_MAX(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_MIN: + if ( GL_FALSE == assemble_MIN(pR700AsmCode) ) + return GL_FALSE; + break; + + case OPCODE_MOV: + if ( GL_FALSE == assemble_MOV(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_MUL: + if ( GL_FALSE == assemble_MUL(pR700AsmCode) ) + return GL_FALSE; + break; + + case OPCODE_POW: + if ( GL_FALSE == assemble_POW(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_RCP: + if ( GL_FALSE == assemble_RCP(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_RSQ: + if ( GL_FALSE == assemble_RSQ(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_SIN: + if ( GL_FALSE == assemble_SIN(pR700AsmCode) ) + return GL_FALSE; + break; + case OPCODE_SCS: + if ( GL_FALSE == assemble_SCS(pR700AsmCode) ) + return GL_FALSE; + break; + + case OPCODE_SEQ: + if(OPCODE_IF == pILInst[i+1].Opcode) + { + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETE) ) + { + return GL_FALSE; + } + } + else if(OPCODE_BRK == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_BREAK; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETE) ) + { + return GL_FALSE; + } + } + else if(OPCODE_CONT == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_CONTINUE; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETE) ) + { + return GL_FALSE; + } + } + else + { + if ( GL_FALSE == assemble_LOGIC(pR700AsmCode, SQ_OP2_INST_SETE) ) + { + return GL_FALSE; + } + } + break; + + case OPCODE_SGT: + if(OPCODE_IF == pILInst[i+1].Opcode) + { + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGT) ) + { + return GL_FALSE; + } + } + else if(OPCODE_BRK == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_BREAK; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGT) ) + { + return GL_FALSE; + } + } + else if(OPCODE_CONT == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_CONTINUE; +#endif + + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGT) ) + { + return GL_FALSE; + } + } + else + { + if ( GL_FALSE == assemble_LOGIC(pR700AsmCode, SQ_OP2_INST_SETGT) ) + { + return GL_FALSE; + } + } + break; + + case OPCODE_SGE: + if(OPCODE_IF == pILInst[i+1].Opcode) + { + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGE) ) + { + return GL_FALSE; + } + } + else if(OPCODE_BRK == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_BREAK; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGE) ) + { + return GL_FALSE; + } + } + else if(OPCODE_CONT == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_CONTINUE; +#endif + + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGE) ) + { + return GL_FALSE; + } + } + else + { + if ( GL_FALSE == assemble_SGE(pR700AsmCode) ) + { + return GL_FALSE; + } + } + break; + + /* NO LT, LE, TODO : use GE => LE, GT => LT : reverse 2 src order would be simpliest. Or use SQ_CF_COND_FALSE for SQ_CF_COND_ACTIVE.*/ + case OPCODE_SLT: + { + struct prog_src_register SrcRegSave[2]; + SrcRegSave[0] = pILInst[i].SrcReg[0]; + SrcRegSave[1] = pILInst[i].SrcReg[1]; + pILInst[i].SrcReg[0] = SrcRegSave[1]; + pILInst[i].SrcReg[1] = SrcRegSave[0]; + if(OPCODE_IF == pILInst[i+1].Opcode) + { + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGT) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + else if(OPCODE_BRK == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_BREAK; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGT) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + else if(OPCODE_CONT == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_CONTINUE; +#endif + + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGT) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + else + { + if ( GL_FALSE == assemble_LOGIC(pR700AsmCode, SQ_OP2_INST_SETGT) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + } break; - case OPCODE_FLR: - if ( GL_FALSE == assemble_FLR(pR700AsmCode) ) - return GL_FALSE; - break; - //case OP_FLR_INT: - // if ( GL_FALSE == assemble_FLR_INT() ) - // return GL_FALSE; - // break; - - case OPCODE_FRC: - if ( GL_FALSE == assemble_FRC(pR700AsmCode) ) - return GL_FALSE; - break; + case OPCODE_SLE: + { + struct prog_src_register SrcRegSave[2]; + SrcRegSave[0] = pILInst[i].SrcReg[0]; + SrcRegSave[1] = pILInst[i].SrcReg[1]; + pILInst[i].SrcReg[0] = SrcRegSave[1]; + pILInst[i].SrcReg[1] = SrcRegSave[0]; + if(OPCODE_IF == pILInst[i+1].Opcode) + { + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGE) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + else if(OPCODE_BRK == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_BREAK; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGE) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + else if(OPCODE_CONT == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_CONTINUE; +#endif - case OPCODE_KIL: - if ( GL_FALSE == assemble_KIL(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_LG2: - if ( GL_FALSE == assemble_LG2(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_LIT: - if ( GL_FALSE == assemble_LIT(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_LRP: - if ( GL_FALSE == assemble_LRP(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_LOG: - if ( GL_FALSE == assemble_LOG(pR700AsmCode) ) - return GL_FALSE; + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETGE) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + else + { + if ( GL_FALSE == assemble_LOGIC(pR700AsmCode, SQ_OP2_INST_SETGE) ) + { + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + return GL_FALSE; + } + } + pILInst[i].SrcReg[0] = SrcRegSave[0]; + pILInst[i].SrcReg[1] = SrcRegSave[1]; + } break; - case OPCODE_MAD: - if ( GL_FALSE == assemble_MAD(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_MAX: - if ( GL_FALSE == assemble_MAX(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_MIN: - if ( GL_FALSE == assemble_MIN(pR700AsmCode) ) - return GL_FALSE; - break; - - case OPCODE_MOV: - if ( GL_FALSE == assemble_MOV(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_MUL: - if ( GL_FALSE == assemble_MUL(pR700AsmCode) ) - return GL_FALSE; - break; - - case OPCODE_POW: - if ( GL_FALSE == assemble_POW(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_RCP: - if ( GL_FALSE == assemble_RCP(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_RSQ: - if ( GL_FALSE == assemble_RSQ(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_SIN: - if ( GL_FALSE == assemble_SIN(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_SCS: - if ( GL_FALSE == assemble_SCS(pR700AsmCode) ) - return GL_FALSE; - break; - - case OPCODE_SGE: - if ( GL_FALSE == assemble_SGE(pR700AsmCode) ) - return GL_FALSE; - break; - case OPCODE_SLT: - if ( GL_FALSE == assemble_SLT(pR700AsmCode) ) - return GL_FALSE; - break; + case OPCODE_SNE: + if(OPCODE_IF == pILInst[i+1].Opcode) + { + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETNE) ) + { + return GL_FALSE; + } + } + else if(OPCODE_BRK == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_BREAK; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETNE) ) + { + return GL_FALSE; + } + } + else if(OPCODE_CONT == pILInst[i+1].Opcode) + { +#ifdef USE_CF_FOR_CONTINUE_BREAK + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_PUSH_BEFORE; +#else + pR700AsmCode->alu_x_opcode = SQ_CF_INST_ALU_CONTINUE; +#endif + if ( GL_FALSE == assemble_LOGIC_PRED(pR700AsmCode, SQ_OP2_INST_PRED_SETNE) ) + { + return GL_FALSE; + } + } + else + { + if ( GL_FALSE == assemble_LOGIC(pR700AsmCode, SQ_OP2_INST_SETNE) ) + { + return GL_FALSE; + } + } + break; //case OP_STP: // if ( GL_FALSE == assemble_STP(pR700AsmCode) ) @@ -4471,24 +6165,91 @@ GLboolean AssembleInstr(GLuint uiNumberInsts, break; case OPCODE_IF : - if ( GL_FALSE == assemble_IF(pR700AsmCode) ) - return GL_FALSE; + { + GLboolean bHasElse = GL_FALSE; + + if(pILInst[pILInst[i].BranchTarget - 1].Opcode == OPCODE_ELSE) + { + bHasElse = GL_TRUE; + } + + if ( GL_FALSE == assemble_IF(pR700AsmCode, bHasElse) ) + { + return GL_FALSE; + } + } break; + case OPCODE_ELSE : - radeon_error("Not yet implemented instruction OPCODE_ELSE \n"); - //if ( GL_FALSE == assemble_BAD("ELSE") ) + if ( GL_FALSE == assemble_ELSE(pR700AsmCode) ) return GL_FALSE; break; + case OPCODE_ENDIF: if ( GL_FALSE == assemble_ENDIF(pR700AsmCode) ) return GL_FALSE; break; + case OPCODE_BGNLOOP: + if( GL_FALSE == assemble_BGNLOOP(pR700AsmCode) ) + { + return GL_FALSE; + } + break; + + case OPCODE_BRK: + if( GL_FALSE == assemble_BRK(pR700AsmCode) ) + { + return GL_FALSE; + } + break; + + case OPCODE_CONT: + if( GL_FALSE == assemble_CONT(pR700AsmCode) ) + { + return GL_FALSE; + } + break; + + case OPCODE_ENDLOOP: + if( GL_FALSE == assemble_ENDLOOP(pR700AsmCode) ) + { + return GL_FALSE; + } + break; + + case OPCODE_BGNSUB: + if( GL_FALSE == assemble_BGNSUB(pR700AsmCode, i) ) + { + return GL_FALSE; + } + break; + + case OPCODE_RET: + if( GL_FALSE == assemble_RET(pR700AsmCode) ) + { + return GL_FALSE; + } + break; + + case OPCODE_CAL: + if( GL_FALSE == assemble_CAL(pR700AsmCode, + pILInst[i].BranchTarget, + uiNumberInsts, + pILInst) ) + { + return GL_FALSE; + } + break; + //case OPCODE_EXPORT: // if ( GL_FALSE == assemble_EXPORT() ) // return GL_FALSE; // break; + case OPCODE_ENDSUB: + return assemble_ENDSUB(pR700AsmCode); + case OPCODE_END: //pR700AsmCode->uiCurInst = i; //This is to remaind that if in later exoort there is depth/stencil @@ -4505,6 +6266,116 @@ GLboolean AssembleInstr(GLuint uiNumberInsts, return GL_TRUE; } +GLboolean InitShaderProgram(r700_AssemblerBase * pAsm) +{ + setRetInLoopFlag(pAsm, SQ_SEL_0); + return GL_TRUE; +} + +GLboolean RelocProgram(r700_AssemblerBase * pAsm) +{ + GLuint i; + GLuint unCFoffset; + TypedShaderList * plstCFmain; + TypedShaderList * plstCFsub; + + R700ShaderInstruction * pInst; + R700ControlFlowGenericClause * pCFInst; + + if(0 == pAsm->unSubArrayPointer) + { + return GL_TRUE; + } + + plstCFmain = pAsm->CALLSTACK[0].plstCFInstructions_local; + unCFoffset = plstCFmain->uNumOfNode; + + /* Reloc subs */ + for(i=0; iunSubArrayPointer; i++) + { + pAsm->subs[i].unCFoffset = unCFoffset; + plstCFsub = &(pAsm->subs[i].lstCFInstructions_local); + + pInst = plstCFsub->pHead; + + /* reloc instructions */ + while(pInst) + { + if(SIT_CF_GENERIC == pInst->m_ShaderInstType) + { + pCFInst = (R700ControlFlowGenericClause *)pInst; + + switch (pCFInst->m_Word1.f.cf_inst) + { + case SQ_CF_INST_POP: + case SQ_CF_INST_JUMP: + case SQ_CF_INST_ELSE: + case SQ_CF_INST_LOOP_END: + case SQ_CF_INST_LOOP_START: + case SQ_CF_INST_LOOP_START_NO_AL: + case SQ_CF_INST_LOOP_CONTINUE: + case SQ_CF_INST_LOOP_BREAK: + pCFInst->m_Word0.f.addr += unCFoffset; + break; + default: + break; + } + } + + pInst->m_uIndex += unCFoffset; + + pInst = pInst->pNextInst; + }; + + /* Put sub into main */ + plstCFmain->pTail->pNextInst = plstCFsub->pHead; + plstCFmain->pTail = plstCFsub->pTail; + plstCFmain->uNumOfNode += plstCFsub->uNumOfNode; + + unCFoffset += plstCFsub->uNumOfNode; + } + + /* reloc callers */ + for(i=0; iunCallerArrayPointer; i++) + { + pAsm->callers[i].cf_ptr->m_Word0.f.addr + = pAsm->subs[pAsm->callers[i].subDescIndex].unCFoffset; + } + + /* remove flags init if they are not used */ + if((pAsm->unCFflags & HAS_LOOPRET) == 0) + { + R700ControlFlowALUClause * pCF_ALU; + pInst = plstCFmain->pHead; + while(pInst) + { + if(SIT_CF_ALU == pInst->m_ShaderInstType) + { + pCF_ALU = (R700ControlFlowALUClause *)pInst; + if(1 == pCF_ALU->m_Word1.f.count) + { + pCF_ALU->m_Word1.f.cf_inst = SQ_CF_INST_NOP; + } + else + { + R700ALUInstruction * pALU = pCF_ALU->m_pLinkedALUInstruction; + + pALU->m_pLinkedALUClause = NULL; + pALU = (R700ALUInstruction *)(pALU->pNextInst); + pALU->m_pLinkedALUClause = pCF_ALU; + pCF_ALU->m_pLinkedALUInstruction = pALU; + + pCF_ALU->m_Word1.f.count--; + } + break; + } + pInst = pInst->pNextInst; + }; + } + + return GL_TRUE; +} + GLboolean Process_Export(r700_AssemblerBase* pAsm, GLuint type, GLuint export_starting_index, @@ -4800,6 +6671,25 @@ GLboolean Process_Vertex_Exports(r700_AssemblerBase *pR700AsmCode, } } + for(i=VERT_RESULT_VAR0; iucVP_OutputMap[i], + GL_FALSE) ) + { + return GL_FALSE; + } + + export_starting_index++; + } + } + // At least one param should be exported if (export_count) { @@ -4833,6 +6723,16 @@ GLboolean Clean_Up_Assembler(r700_AssemblerBase *pR700AsmCode) { FREE(pR700AsmCode->pucOutMask); FREE(pR700AsmCode->pInstDeps); + + if(NULL != pR700AsmCode->subs) + { + FREE(pR700AsmCode->subs); + } + if(NULL != pR700AsmCode->callers) + { + FREE(pR700AsmCode->callers); + } + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r600/r700_assembler.h b/src/mesa/drivers/dri/r600/r700_assembler.h index c66db502a1..85d32212c0 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.h +++ b/src/mesa/drivers/dri/r600/r700_assembler.h @@ -72,7 +72,8 @@ typedef enum SrcRegisterType SRC_REG_INPUT = 1, SRC_REG_CONSTANT = 2, SRC_REG_ALT_TEMPORARY = 3, - NUMBER_OF_SRC_REG_TYPE = 4 + SRC_REC_LITERAL = 4, + NUMBER_OF_SRC_REG_TYPE = 5 } SrcRegisterType; typedef enum DstRegisterType @@ -111,6 +112,12 @@ typedef struct PVSDSTtag BITS addrmode1:1; //32 } PVSDST; +typedef struct PVSINSTtag +{ + BITS literal :2; + BITS SaturateMode :2; +} PVSINST; + typedef struct PVSSRCtag { BITS rtype:4; @@ -148,6 +155,7 @@ typedef union PVSDWORDtag { BITS bits; PVSDST dst; + PVSINST dst2; PVSSRC src; PVSMATH math; float f; @@ -263,14 +271,15 @@ enum typedef struct FC_LEVEL { - unsigned int first; ///< first fc instruction on level (if, rep, loop) - unsigned int* mid; ///< middle instructions - else or all breaks on this level - unsigned int midLen; - unsigned int type; - unsigned int cond; - unsigned int inv; - unsigned int bpush; ///< 1 if first instruction does branch stack push - int id; ///< id of bool or int variable + R700ControlFlowGenericClause * first; + R700ControlFlowGenericClause ** mid; + unsigned int unNumMid; + unsigned int midLen; + unsigned int type; + unsigned int cond; + unsigned int inv; + unsigned int bpush; ///< 1 if first instruction does branch stack push + int id; ///< id of bool or int variable } FC_LEVEL; typedef struct VTX_FETCH_METHOD @@ -279,6 +288,28 @@ typedef struct VTX_FETCH_METHOD GLuint mega_fetch_remainder; } VTX_FETCH_METHOD; +typedef struct SUB_OFFSET +{ + GLint subIL_Offset; + GLuint unCFoffset; + TypedShaderList lstCFInstructions_local; +} SUB_OFFSET; + +typedef struct CALLER_POINTER +{ + GLint subIL_Offset; + GLint subDescIndex; + R700ControlFlowGenericClause* cf_ptr; +} CALLER_POINTER; + +#define SQ_MAX_CALL_DEPTH 0x00000020 + +typedef struct CALL_LEVEL +{ + unsigned int FCSP_BeforeEntry; + TypedShaderList * plstCFInstructions_local; +} CALL_LEVEL; + typedef struct r700_AssemblerBase { R700ControlFlowSXClause* cf_last_export_ptr; @@ -294,11 +325,14 @@ typedef struct r700_AssemblerBase // No clause has been created yet CF_CLAUSE_TYPE cf_current_clause_type; + BITS alu_x_opcode; + GLuint number_of_exports; GLuint number_of_colorandz_exports; GLuint number_of_export_opcodes; PVSDWORD D; + PVSDWORD D2; PVSDWORD S[3]; unsigned int uLastPosUpdate; @@ -310,6 +344,8 @@ typedef struct r700_AssemblerBase unsigned int number_used_registers; unsigned int uUsedConsts; + unsigned int flag_reg_index; + // Fragment programs unsigned int uiFP_AttributeMap[FRAG_ATTRIB_MAX]; unsigned int uiFP_OutputMap[FRAG_RESULT_MAX]; @@ -378,6 +414,18 @@ typedef struct r700_AssemblerBase GLboolean is_tex; /* we inserted helper intructions and need barrier on next TEX ins */ GLboolean need_tex_barrier; + + SUB_OFFSET * subs; + GLuint unSubArraySize; + GLuint unSubArrayPointer; + CALLER_POINTER * callers; + GLuint unCallerArraySize; + GLuint unCallerArrayPointer; + unsigned int CALLSP; + CALL_LEVEL CALLSTACK[SQ_MAX_CALL_DEPTH]; + + GLuint unCFflags; + } r700_AssemblerBase; //Internal use @@ -446,6 +494,10 @@ GLboolean assemble_alu_src(R700ALUInstruction* alu_instruction_ptr, GLboolean add_alu_instruction(r700_AssemblerBase* pAsm, R700ALUInstruction* alu_instruction_ptr, GLuint contiguous_slots_needed); + +GLboolean add_cf_instruction(r700_AssemblerBase* pAsm); +void add_return_inst(r700_AssemblerBase *pAsm); + void get_src_properties(R700ALUInstruction* alu_instruction_ptr, int source_index, BITS* psrc_sel, @@ -467,6 +519,21 @@ GLboolean check_vector(r700_AssemblerBase* pAsm, R700ALUInstruction* alu_instruction_ptr); GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm); GLboolean next_ins(r700_AssemblerBase *pAsm); + +GLboolean next_ins2(r700_AssemblerBase *pAsm); +GLboolean assemble_alu_instruction2(r700_AssemblerBase *pAsm); + +/* TODO : merge next_ins/2/literal, assemble_alu_instruction/2/literal */ +GLboolean next_ins_literal(r700_AssemblerBase *pAsm, GLfloat * pLiteral); +GLboolean assemble_alu_instruction_literal(r700_AssemblerBase *pAsm, GLfloat * pLiteral); + +GLboolean pops(r700_AssemblerBase *pAsm, GLuint pops); +GLboolean jumpToOffest(r700_AssemblerBase *pAsm, GLuint pops, GLint offset); +GLboolean setRetInLoopFlag(r700_AssemblerBase *pAsm, GLuint flagValue); +GLboolean testFlag(r700_AssemblerBase *pAsm); +GLboolean breakLoopOnFlag(r700_AssemblerBase *pAsm, GLuint unFCSP); +GLboolean returnOnFlag(r700_AssemblerBase *pAsm); + GLboolean assemble_math_function(r700_AssemblerBase* pAsm, BITS opcode); GLboolean assemble_ABS(r700_AssemblerBase *pAsm); GLboolean assemble_ADD(r700_AssemblerBase *pAsm); @@ -497,14 +564,32 @@ GLboolean assemble_RSQ(r700_AssemblerBase *pAsm); GLboolean assemble_SIN(r700_AssemblerBase *pAsm); GLboolean assemble_SCS(r700_AssemblerBase *pAsm); GLboolean assemble_SGE(r700_AssemblerBase *pAsm); + +GLboolean assemble_LOGIC(r700_AssemblerBase *pAsm, BITS opcode); +GLboolean assemble_LOGIC_PRED(r700_AssemblerBase *pAsm, BITS opcode); + GLboolean assemble_SLT(r700_AssemblerBase *pAsm); GLboolean assemble_STP(r700_AssemblerBase *pAsm); GLboolean assemble_TEX(r700_AssemblerBase *pAsm); GLboolean assemble_XPD(r700_AssemblerBase *pAsm); GLboolean assemble_EXPORT(r700_AssemblerBase *pAsm); -GLboolean assemble_IF(r700_AssemblerBase *pAsm); +GLboolean assemble_IF(r700_AssemblerBase *pAsm, GLboolean bHasElse); +GLboolean assemble_ELSE(r700_AssemblerBase *pAsm); GLboolean assemble_ENDIF(r700_AssemblerBase *pAsm); +GLboolean assemble_BGNLOOP(r700_AssemblerBase *pAsm); +GLboolean assemble_BRK(r700_AssemblerBase *pAsm); +GLboolean assemble_COND(r700_AssemblerBase *pAsm); +GLboolean assemble_ENDLOOP(r700_AssemblerBase *pAsm); + +GLboolean assemble_BGNSUB(r700_AssemblerBase *pAsm, GLint nILindex); +GLboolean assemble_ENDSUB(r700_AssemblerBase *pAsm); +GLboolean assemble_RET(r700_AssemblerBase *pAsm); +GLboolean assemble_CAL(r700_AssemblerBase *pAsm, + GLint nILindex, + GLuint uiNumberInsts, + struct prog_instruction *pILInst); + GLboolean Process_Export(r700_AssemblerBase* pAsm, GLuint type, GLuint export_starting_index, @@ -516,12 +601,16 @@ GLboolean Move_Depth_Exports_To_Correct_Channels(r700_AssemblerBase *pAsm, //Interface -GLboolean AssembleInstr(GLuint uiNumberInsts, +GLboolean AssembleInstr(GLuint uiFirstInst, + GLuint uiNumberInsts, struct prog_instruction *pILInst, r700_AssemblerBase *pR700AsmCode); GLboolean Process_Fragment_Exports(r700_AssemblerBase *pR700AsmCode, GLbitfield OutputsWritten); GLboolean Process_Vertex_Exports(r700_AssemblerBase *pR700AsmCode, GLbitfield OutputsWritten); +GLboolean RelocProgram(r700_AssemblerBase * pAsm); +GLboolean InitShaderProgram(r700_AssemblerBase * pAsm); + int Init_r700_AssemblerBase(SHADER_PIPE_TYPE spt, r700_AssemblerBase* pAsm, R700_Shader* pShader); GLboolean Clean_Up_Assembler(r700_AssemblerBase *pR700AsmCode); diff --git a/src/mesa/drivers/dri/r600/r700_chip.c b/src/mesa/drivers/dri/r600/r700_chip.c index ec76fbcb6d..197916ac0d 100644 --- a/src/mesa/drivers/dri/r600/r700_chip.c +++ b/src/mesa/drivers/dri/r600/r700_chip.c @@ -442,68 +442,77 @@ static void r700SendRenderTargetState(GLcontext *ctx, struct radeon_state_atom * static void r700SendPSState(GLcontext *ctx, struct radeon_state_atom *atom) { - context_t *context = R700_CONTEXT(ctx); - R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); - struct radeon_bo * pbo; - BATCH_LOCALS(&context->radeon); - radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s\n", __func__); + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); + struct radeon_bo * pbo; + BATCH_LOCALS(&context->radeon); + radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s\n", __func__); - pbo = (struct radeon_bo *)r700GetActiveFpShaderBo(GL_CONTEXT(context)); + pbo = (struct radeon_bo *)r700GetActiveFpShaderBo(GL_CONTEXT(context)); - if (!pbo) - return; + if (!pbo) + return; - r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit); + r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit); - BEGIN_BATCH_NO_AUTOSTATE(3 + 2); - R600_OUT_BATCH_REGSEQ(SQ_PGM_START_PS, 1); - R600_OUT_BATCH(r700->ps.SQ_PGM_START_PS.u32All); - R600_OUT_BATCH_RELOC(r700->ps.SQ_PGM_START_PS.u32All, - pbo, - r700->ps.SQ_PGM_START_PS.u32All, - RADEON_GEM_DOMAIN_GTT, 0, 0); - END_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(3 + 2); + R600_OUT_BATCH_REGSEQ(SQ_PGM_START_PS, 1); + R600_OUT_BATCH(r700->ps.SQ_PGM_START_PS.u32All); + R600_OUT_BATCH_RELOC(r700->ps.SQ_PGM_START_PS.u32All, + pbo, + r700->ps.SQ_PGM_START_PS.u32All, + RADEON_GEM_DOMAIN_GTT, 0, 0); + END_BATCH(); - BEGIN_BATCH_NO_AUTOSTATE(9); - R600_OUT_BATCH_REGVAL(SQ_PGM_RESOURCES_PS, r700->ps.SQ_PGM_RESOURCES_PS.u32All); - R600_OUT_BATCH_REGVAL(SQ_PGM_EXPORTS_PS, r700->ps.SQ_PGM_EXPORTS_PS.u32All); - R600_OUT_BATCH_REGVAL(SQ_PGM_CF_OFFSET_PS, r700->ps.SQ_PGM_CF_OFFSET_PS.u32All); - END_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(9); + R600_OUT_BATCH_REGVAL(SQ_PGM_RESOURCES_PS, r700->ps.SQ_PGM_RESOURCES_PS.u32All); + R600_OUT_BATCH_REGVAL(SQ_PGM_EXPORTS_PS, r700->ps.SQ_PGM_EXPORTS_PS.u32All); + R600_OUT_BATCH_REGVAL(SQ_PGM_CF_OFFSET_PS, r700->ps.SQ_PGM_CF_OFFSET_PS.u32All); + END_BATCH(); - COMMIT_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(3); + R600_OUT_BATCH_REGVAL(SQ_LOOP_CONST_0, 0x01000FFF); + END_BATCH(); + + COMMIT_BATCH(); } static void r700SendVSState(GLcontext *ctx, struct radeon_state_atom *atom) { - context_t *context = R700_CONTEXT(ctx); - R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); - struct radeon_bo * pbo; - BATCH_LOCALS(&context->radeon); - radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s\n", __func__); + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); + struct radeon_bo * pbo; + BATCH_LOCALS(&context->radeon); + radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s\n", __func__); - pbo = (struct radeon_bo *)r700GetActiveVpShaderBo(GL_CONTEXT(context)); + pbo = (struct radeon_bo *)r700GetActiveVpShaderBo(GL_CONTEXT(context)); - if (!pbo) - return; + if (!pbo) + return; - r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit); + r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit); - BEGIN_BATCH_NO_AUTOSTATE(3 + 2); - R600_OUT_BATCH_REGSEQ(SQ_PGM_START_VS, 1); - R600_OUT_BATCH(r700->vs.SQ_PGM_START_VS.u32All); - R600_OUT_BATCH_RELOC(r700->vs.SQ_PGM_START_VS.u32All, - pbo, - r700->vs.SQ_PGM_START_VS.u32All, - RADEON_GEM_DOMAIN_GTT, 0, 0); - END_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(3 + 2); + R600_OUT_BATCH_REGSEQ(SQ_PGM_START_VS, 1); + R600_OUT_BATCH(r700->vs.SQ_PGM_START_VS.u32All); + R600_OUT_BATCH_RELOC(r700->vs.SQ_PGM_START_VS.u32All, + pbo, + r700->vs.SQ_PGM_START_VS.u32All, + RADEON_GEM_DOMAIN_GTT, 0, 0); + END_BATCH(); - BEGIN_BATCH_NO_AUTOSTATE(6); - R600_OUT_BATCH_REGVAL(SQ_PGM_RESOURCES_VS, r700->vs.SQ_PGM_RESOURCES_VS.u32All); - R600_OUT_BATCH_REGVAL(SQ_PGM_CF_OFFSET_VS, r700->vs.SQ_PGM_CF_OFFSET_VS.u32All); - END_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(6); + R600_OUT_BATCH_REGVAL(SQ_PGM_RESOURCES_VS, r700->vs.SQ_PGM_RESOURCES_VS.u32All); + R600_OUT_BATCH_REGVAL(SQ_PGM_CF_OFFSET_VS, r700->vs.SQ_PGM_CF_OFFSET_VS.u32All); + END_BATCH(); - COMMIT_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(3); + R600_OUT_BATCH_REGVAL((SQ_LOOP_CONST_0 + 32*4), 0x0100000F); + //R600_OUT_BATCH_REGVAL((SQ_LOOP_CONST_0 + (SQ_LOOP_CONST_vs<2)), 0x0100000F); + END_BATCH(); + + COMMIT_BATCH(); } static void r700SendFSState(GLcontext *ctx, struct radeon_state_atom *atom) diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.c b/src/mesa/drivers/dri/r600/r700_fragprog.c index ccafd433bf..21ac46e7b8 100644 --- a/src/mesa/drivers/dri/r600/r700_fragprog.c +++ b/src/mesa/drivers/dri/r600/r700_fragprog.c @@ -73,11 +73,11 @@ void Map_Fragment_Program(r700_AssemblerBase *pAsm, pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL1] = pAsm->number_used_registers++; } - unBit = 1 << FRAG_ATTRIB_FOGC; - if(mesa_fp->Base.InputsRead & unBit) - { - pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC] = pAsm->number_used_registers++; - } + unBit = 1 << FRAG_ATTRIB_FOGC; + if(mesa_fp->Base.InputsRead & unBit) + { + pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC] = pAsm->number_used_registers++; + } for(i=0; i<8; i++) { @@ -88,6 +88,62 @@ void Map_Fragment_Program(r700_AssemblerBase *pAsm, } } +/* order has been taken care of */ +#if 1 + for(i=FRAG_ATTRIB_VAR0; iBase.InputsRead & unBit) + { + pAsm->uiFP_AttributeMap[i] = pAsm->number_used_registers++; + } + } +#else + if( (mesa_fp->Base.InputsRead >> FRAG_ATTRIB_VAR0) > 0 ) + { + struct r700_vertex_program_cont *vpc = + (struct r700_vertex_program_cont *)ctx->VertexProgram._Current; + struct gl_program_parameter_list * VsVarying = vpc->mesa_program.Base.Varying; + struct gl_program_parameter_list * PsVarying = mesa_fp->Base.Varying; + struct gl_program_parameter * pVsParam; + struct gl_program_parameter * pPsParam; + GLuint j, k; + GLuint unMaxVarying = 0; + + for(i=0; iNumParameters; i++) + { + pAsm->uiFP_AttributeMap[i + FRAG_ATTRIB_VAR0] = 0; + } + + for(i=FRAG_ATTRIB_VAR0; iBase.InputsRead & unBit) + { + j = i - FRAG_ATTRIB_VAR0; + pPsParam = PsVarying->Parameters + j; + + for(k=0; kNumParameters; k++) + { + pVsParam = VsVarying->Parameters + k; + + if( strcmp(pPsParam->Name, pVsParam->Name) == 0) + { + pAsm->uiFP_AttributeMap[i] = pAsm->number_used_registers + k; + if(k > unMaxVarying) + { + unMaxVarying = k; + } + break; + } + } + } + } + + pAsm->number_used_registers += unMaxVarying + 1; + } +#endif + /* Map temporary registers (GPRs) */ pAsm->starting_temp_register_number = pAsm->number_used_registers; @@ -127,6 +183,8 @@ void Map_Fragment_Program(r700_AssemblerBase *pAsm, pAsm->pucOutMask[ui] = 0x0; } + pAsm->flag_reg_index = pAsm->number_used_registers++; + pAsm->uFirstHelpReg = pAsm->number_used_registers; } @@ -247,8 +305,11 @@ GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, { return GL_FALSE; } + + InitShaderProgram(&(fp->r700AsmCode)); - if( GL_FALSE == AssembleInstr(mesa_fp->Base.NumInstructions, + if( GL_FALSE == AssembleInstr(0, + mesa_fp->Base.NumInstructions, &(mesa_fp->Base.Instructions[0]), &(fp->r700AsmCode)) ) { @@ -260,6 +321,11 @@ GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, return GL_FALSE; } + if( GL_FALSE == RelocProgram(&(fp->r700AsmCode)) ) + { + return GL_FALSE; + } + fp->r700Shader.nRegs = (fp->r700AsmCode.number_used_registers == 0) ? 0 : (fp->r700AsmCode.number_used_registers - 1); @@ -459,6 +525,22 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) } } + for(i=FRAG_ATTRIB_VAR0; iBase.InputsRead & unBit) + { + ui = pAsm->uiFP_AttributeMap[i]; + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); + SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui, + SEMANTIC_shift, SEMANTIC_mask); + if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit) + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + else + CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + } + } + exportCount = (r700->ps.SQ_PGM_EXPORTS_PS.u32All & EXPORT_MODE_mask) / (1 << EXPORT_MODE_shift); if (r700->CB_SHADER_CONTROL.u32All != ((1 << exportCount) - 1)) { diff --git a/src/mesa/drivers/dri/r600/r700_shader.c b/src/mesa/drivers/dri/r600/r700_shader.c index 955ea4e4e1..2eed1acc2f 100644 --- a/src/mesa/drivers/dri/r600/r700_shader.c +++ b/src/mesa/drivers/dri/r600/r700_shader.c @@ -159,13 +159,18 @@ void Init_R700_Shader(R700_Shader * pShader) pShader->lstVTXInstructions.uNumOfNode=0; } +void SetActiveCFlist(R700_Shader *pShader, TypedShaderList * plstCF) +{ + pShader->plstCFInstructions_active = plstCF; +} + void AddCFInstruction(R700_Shader *pShader, R700ControlFlowInstruction *pCFInst) { R700ControlFlowSXClause* pSXClause; R700ControlFlowSMXClause* pSMXClause; - pCFInst->m_uIndex = pShader->lstCFInstructions.uNumOfNode; - AddInstToList(&(pShader->lstCFInstructions), + pCFInst->m_uIndex = pShader->plstCFInstructions_active->uNumOfNode; + AddInstToList(pShader->plstCFInstructions_active, (R700ShaderInstruction*)pCFInst); pShader->uShaderBinaryDWORDSize += GetInstructionSize(pCFInst->m_ShaderInstType); diff --git a/src/mesa/drivers/dri/r600/r700_shader.h b/src/mesa/drivers/dri/r600/r700_shader.h index c6a058617e..0599ffd901 100644 --- a/src/mesa/drivers/dri/r600/r700_shader.h +++ b/src/mesa/drivers/dri/r600/r700_shader.h @@ -109,6 +109,7 @@ typedef struct R700_Shader GLuint uStackSize; GLuint uMaxCallDepth; + TypedShaderList * plstCFInstructions_active; TypedShaderList lstCFInstructions; TypedShaderList lstALUInstructions; TypedShaderList lstTEXInstructions; @@ -132,13 +133,13 @@ void TakeInstOutFromList(TypedShaderList * plstCFInstructions, R700ShaderInstruc void ResolveLinks(R700_Shader *pShader); void Assemble(R700_Shader *pShader); - //Interface void Init_R700_Shader(R700_Shader * pShader); void AddCFInstruction(R700_Shader *pShader, R700ControlFlowInstruction *pCFInst); void AddVTXInstruction(R700_Shader *pShader, R700VertexInstruction *pVTXInst); void AddTEXInstruction(R700_Shader *pShader, R700TextureInstruction *pTEXInst); void AddALUInstruction(R700_Shader *pShader, R700ALUInstruction *pALUInst); +void SetActiveCFlist(R700_Shader *pShader, TypedShaderList * plstCF); void LoadProgram(R700_Shader *pShader); void UpdateShaderRegisters(R700_Shader *pShader); diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index ffc6068bd8..c8f72d588b 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -111,6 +111,15 @@ unsigned int Map_Vertex_Output(r700_AssemblerBase *pAsm, } } + for(i=VERT_RESULT_VAR0; iBase.OutputsWritten & unBit) + { + pAsm->ucVP_OutputMap[i] = unTotal++; + } + } + return (unTotal - unStart); } @@ -235,6 +244,8 @@ void Map_Vertex_Program(GLcontext *ctx, pAsm->number_used_registers += mesa_vp->Base.NumTemporaries; } + pAsm->flag_reg_index = pAsm->number_used_registers++; + pAsm->uFirstHelpReg = pAsm->number_used_registers; } @@ -324,7 +335,10 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, return NULL; } - if(GL_FALSE == AssembleInstr(vp->mesa_program->Base.NumInstructions, + InitShaderProgram(&(vp->r700AsmCode)); + + if(GL_FALSE == AssembleInstr(0, + vp->mesa_program->Base.NumInstructions, &(vp->mesa_program->Base.Instructions[0]), &(vp->r700AsmCode)) ) { @@ -336,6 +350,11 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, return NULL; } + if( GL_FALSE == RelocProgram(&(vp->r700AsmCode)) ) + { + return GL_FALSE; + } + vp->r700Shader.nRegs = (vp->r700AsmCode.number_used_registers == 0) ? 0 : (vp->r700AsmCode.number_used_registers - 1); -- cgit v1.2.3 From 2db72f329f35ee6e12df3ed472de2ee72cf23399 Mon Sep 17 00:00:00 2001 From: Richard Li Date: Sun, 29 Nov 2009 12:12:19 -0500 Subject: r600 : add read port allocation for uniform; mapping ps input based on vs output; fix bugs including constants updating for vs. --- src/mesa/drivers/dri/r600/r700_assembler.c | 46 +++++++++------- src/mesa/drivers/dri/r600/r700_fragprog.c | 85 +++++++++++++++++------------- src/mesa/drivers/dri/r600/r700_fragprog.h | 6 ++- src/mesa/drivers/dri/r600/r700_vertprog.c | 26 +++++++-- 4 files changed, 101 insertions(+), 62 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index ba97d3e073..309c90fdd0 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -1038,7 +1038,8 @@ GLboolean checkop2(r700_AssemblerBase* pAsm) checkop_init(pAsm); - if( (pILInst->SrcReg[0].File == PROGRAM_CONSTANT) || + if( (pILInst->SrcReg[0].File == PROGRAM_UNIFORM) || + (pILInst->SrcReg[0].File == PROGRAM_CONSTANT) || (pILInst->SrcReg[0].File == PROGRAM_LOCAL_PARAM) || (pILInst->SrcReg[0].File == PROGRAM_ENV_PARAM) || (pILInst->SrcReg[0].File == PROGRAM_STATE_VAR) ) @@ -1049,7 +1050,8 @@ GLboolean checkop2(r700_AssemblerBase* pAsm) { bSrcConst[0] = GL_FALSE; } - if( (pILInst->SrcReg[1].File == PROGRAM_CONSTANT) || + if( (pILInst->SrcReg[1].File == PROGRAM_UNIFORM) || + (pILInst->SrcReg[1].File == PROGRAM_CONSTANT) || (pILInst->SrcReg[1].File == PROGRAM_LOCAL_PARAM) || (pILInst->SrcReg[1].File == PROGRAM_ENV_PARAM) || (pILInst->SrcReg[1].File == PROGRAM_STATE_VAR) ) @@ -1082,7 +1084,8 @@ GLboolean checkop3(r700_AssemblerBase* pAsm) checkop_init(pAsm); - if( (pILInst->SrcReg[0].File == PROGRAM_CONSTANT) || + if( (pILInst->SrcReg[0].File == PROGRAM_UNIFORM) || + (pILInst->SrcReg[0].File == PROGRAM_CONSTANT) || (pILInst->SrcReg[0].File == PROGRAM_LOCAL_PARAM) || (pILInst->SrcReg[0].File == PROGRAM_ENV_PARAM) || (pILInst->SrcReg[0].File == PROGRAM_STATE_VAR) ) @@ -1093,7 +1096,8 @@ GLboolean checkop3(r700_AssemblerBase* pAsm) { bSrcConst[0] = GL_FALSE; } - if( (pILInst->SrcReg[1].File == PROGRAM_CONSTANT) || + if( (pILInst->SrcReg[1].File == PROGRAM_UNIFORM) || + (pILInst->SrcReg[1].File == PROGRAM_CONSTANT) || (pILInst->SrcReg[1].File == PROGRAM_LOCAL_PARAM) || (pILInst->SrcReg[1].File == PROGRAM_ENV_PARAM) || (pILInst->SrcReg[1].File == PROGRAM_STATE_VAR) ) @@ -1104,7 +1108,8 @@ GLboolean checkop3(r700_AssemblerBase* pAsm) { bSrcConst[1] = GL_FALSE; } - if( (pILInst->SrcReg[2].File == PROGRAM_CONSTANT) || + if( (pILInst->SrcReg[2].File == PROGRAM_UNIFORM) || + (pILInst->SrcReg[2].File == PROGRAM_CONSTANT) || (pILInst->SrcReg[2].File == PROGRAM_LOCAL_PARAM) || (pILInst->SrcReg[2].File == PROGRAM_ENV_PARAM) || (pILInst->SrcReg[2].File == PROGRAM_STATE_VAR) ) @@ -1218,7 +1223,7 @@ GLboolean assemble_src(r700_AssemblerBase *pAsm, pAsm->S[fld].src.reg = pILInst->SrcReg[src].Index; break; case PROGRAM_INPUT: - setaddrmode_PVSSRC(&(pAsm->S[fld].src), ADDR_ABSOLUTE); + setaddrmode_PVSSRC(&(pAsm->S[fld].src), ADDR_ABSOLUTE); pAsm->S[fld].src.rtype = SRC_REG_INPUT; switch (pAsm->currentShaderType) { @@ -1346,6 +1351,7 @@ GLboolean tex_src(r700_AssemblerBase *pAsm) else { switch (pILInst->SrcReg[0].File) { + case PROGRAM_UNIFORM: case PROGRAM_CONSTANT: case PROGRAM_LOCAL_PARAM: case PROGRAM_ENV_PARAM: @@ -2117,7 +2123,7 @@ GLboolean check_vector(r700_AssemblerBase* pAsm, if( is_gpr(sel) ) { if( GL_FALSE == cycle_for_vector_bank_swizzle(bank_swizzle, src, &cycle) ) - { + { return GL_FALSE; } @@ -2129,7 +2135,7 @@ GLboolean check_vector(r700_AssemblerBase* pAsm, else { if( GL_FALSE == reserve_gpr(pAsm, sel, chan, cycle) ) - { + { return GL_FALSE; } } @@ -2141,7 +2147,7 @@ GLboolean check_vector(r700_AssemblerBase* pAsm, if( is_cfile(sel) ) { if( GL_FALSE == reserve_cfile(pAsm, sel, chan) ) - { + { return GL_FALSE; } } @@ -2244,7 +2250,7 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) current_source_index, pcurrent_source, scalar_channel_index) ) - { + { return GL_FALSE; } @@ -2258,7 +2264,7 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) current_source_index, pcurrent_source, scalar_channel_index) ) - { + { return GL_FALSE; } } @@ -2287,7 +2293,7 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) alu_instruction_ptr->m_Word1.f.dst_gpr = pAsm->D.dst.reg; } else - { + { radeon_error("Only temp destination registers supported for ALU dest regs.\n"); return GL_FALSE; } @@ -2401,7 +2407,7 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) } if(GL_FALSE == add_alu_instruction(pAsm, alu_instruction_ptr, contiguous_slots_needed) ) - { + { return GL_FALSE; } @@ -2412,15 +2418,15 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) if (is_single_scalar_operation) { if(GL_FALSE == check_scalar(pAsm, alu_instruction_ptr) ) - { + { return GL_FALSE; } } else { if(GL_FALSE == check_vector(pAsm, alu_instruction_ptr) ) - { - return 1; + { + return GL_FALSE; } } @@ -2667,7 +2673,7 @@ GLboolean assemble_alu_instruction2(r700_AssemblerBase *pAsm) { if(GL_FALSE == check_vector(pAsm, alu_instruction_ptr) ) { - return 1; + return GL_FALSE; } } @@ -3642,6 +3648,7 @@ GLboolean assemble_LRP(r700_AssemblerBase *pAsm) { return GL_FALSE; } + if( GL_FALSE == assemble_src(pAsm, 2, -1) ) { return GL_FALSE; @@ -4598,6 +4605,7 @@ GLboolean assemble_TEX(r700_AssemblerBase *pAsm) switch (pAsm->pILInst[pAsm->uiCurInst].SrcReg[0].File) { + case PROGRAM_UNIFORM: case PROGRAM_CONSTANT: case PROGRAM_LOCAL_PARAM: case PROGRAM_ENV_PARAM: @@ -6867,7 +6875,7 @@ GLboolean Process_Vertex_Exports(r700_AssemblerBase *pR700AsmCode, export_starting_index++; } } - + for(i=VERT_RESULT_VAR0; iucVP_OutputMap[i], GL_FALSE) ) - { + { return GL_FALSE; } diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.c b/src/mesa/drivers/dri/r600/r700_fragprog.c index 21ac46e7b8..e9ef6c8695 100644 --- a/src/mesa/drivers/dri/r600/r700_fragprog.c +++ b/src/mesa/drivers/dri/r600/r700_fragprog.c @@ -44,12 +44,18 @@ //TODO : Validate FP input with VP output. void Map_Fragment_Program(r700_AssemblerBase *pAsm, - struct gl_fragment_program *mesa_fp) + struct gl_fragment_program *mesa_fp, + GLcontext *ctx) { unsigned int unBit; unsigned int i; GLuint ui; + /* match fp inputs with vp exports. */ + struct r700_vertex_program_cont *vpc = + (struct r700_vertex_program_cont *)ctx->VertexProgram._Current; + GLbitfield OutputsWritten = vpc->mesa_program.Base.OutputsWritten; + pAsm->number_used_registers = 0; //Input mapping : mesa_fp->Base.InputsRead set the flag, set in @@ -61,41 +67,41 @@ void Map_Fragment_Program(r700_AssemblerBase *pAsm, pAsm->uiFP_AttributeMap[FRAG_ATTRIB_WPOS] = pAsm->number_used_registers++; } - unBit = 1 << FRAG_ATTRIB_COL0; - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << VERT_RESULT_COL0; + if(OutputsWritten & unBit) { pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL0] = pAsm->number_used_registers++; } - unBit = 1 << FRAG_ATTRIB_COL1; - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << VERT_RESULT_COL1; + if(OutputsWritten & unBit) { pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL1] = pAsm->number_used_registers++; } - unBit = 1 << FRAG_ATTRIB_FOGC; - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << VERT_RESULT_FOGC; + if(OutputsWritten & unBit) { - pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC] = pAsm->number_used_registers++; + pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC] = pAsm->number_used_registers++; } for(i=0; i<8; i++) { - unBit = 1 << (FRAG_ATTRIB_TEX0 + i); - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << (VERT_RESULT_TEX0 + i); + if(OutputsWritten & unBit) { pAsm->uiFP_AttributeMap[FRAG_ATTRIB_TEX0 + i] = pAsm->number_used_registers++; } } -/* order has been taken care of */ +/* order has been taken care of */ #if 1 - for(i=FRAG_ATTRIB_VAR0; iBase.InputsRead & unBit) + if(OutputsWritten & unBit) { - pAsm->uiFP_AttributeMap[i] = pAsm->number_used_registers++; + pAsm->uiFP_AttributeMap[i-VERT_RESULT_VAR0+FRAG_ATTRIB_VAR0] = pAsm->number_used_registers++; } } #else @@ -291,7 +297,8 @@ GLboolean Find_Instruction_Dependencies_fp(struct r700_fragment_program *fp, } GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, - struct gl_fragment_program *mesa_fp) + struct gl_fragment_program *mesa_fp, + GLcontext *ctx) { GLuint number_of_colors_exported; GLboolean z_enabled = GL_FALSE; @@ -299,7 +306,7 @@ GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, //Init_Program Init_r700_AssemblerBase( SPT_FP, &(fp->r700AsmCode), &(fp->r700Shader) ); - Map_Fragment_Program(&(fp->r700AsmCode), mesa_fp); + Map_Fragment_Program(&(fp->r700AsmCode), mesa_fp, ctx); if( GL_FALSE == Find_Instruction_Dependencies_fp(fp, mesa_fp) ) { @@ -366,7 +373,7 @@ void r700SelectFragmentShader(GLcontext *ctx) } if (GL_FALSE == fp->translated) - r700TranslateFragmentShader(fp, &(fp->mesa_program)); + r700TranslateFragmentShader(fp, &(fp->mesa_program), ctx); } void * r700GetActiveFpShaderBo(GLcontext * ctx) @@ -460,6 +467,9 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) EXPORT_MODE_shift, EXPORT_MODE_mask); // emit ps input map + struct r700_vertex_program_cont *vpc = + (struct r700_vertex_program_cont *)ctx->VertexProgram._Current; + GLbitfield OutputsWritten = vpc->mesa_program.Base.OutputsWritten; unBit = 1 << FRAG_ATTRIB_WPOS; if(mesa_fp->Base.InputsRead & unBit) { @@ -473,8 +483,8 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); } - unBit = 1 << FRAG_ATTRIB_COL0; - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << VERT_RESULT_COL0; + if(OutputsWritten & unBit) { ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL0]; SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); @@ -486,8 +496,8 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); } - unBit = 1 << FRAG_ATTRIB_COL1; - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << VERT_RESULT_COL1; + if(OutputsWritten & unBit) { ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL1]; SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); @@ -499,8 +509,8 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); } - unBit = 1 << FRAG_ATTRIB_FOGC; - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << VERT_RESULT_FOGC; + if(OutputsWritten & unBit) { ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC]; SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); @@ -514,8 +524,8 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) for(i=0; i<8; i++) { - unBit = 1 << (FRAG_ATTRIB_TEX0 + i); - if(mesa_fp->Base.InputsRead & unBit) + unBit = 1 << (VERT_RESULT_TEX0 + i); + if(OutputsWritten & unBit) { ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_TEX0 + i]; SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); @@ -525,12 +535,12 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) } } - for(i=FRAG_ATTRIB_VAR0; iBase.InputsRead & unBit) + unBit = 1 << i; + if(OutputsWritten & unBit) { - ui = pAsm->uiFP_AttributeMap[i]; + ui = pAsm->uiFP_AttributeMap[i-VERT_RESULT_VAR0+FRAG_ATTRIB_VAR0]; SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui, SEMANTIC_shift, SEMANTIC_mask); @@ -538,8 +548,8 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); else CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); - } - } + } + } exportCount = (r700->ps.SQ_PGM_EXPORTS_PS.u32All & EXPORT_MODE_mask) / (1 << EXPORT_MODE_shift); if (r700->CB_SHADER_CONTROL.u32All != ((1 << exportCount) - 1)) @@ -551,7 +561,8 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) /* sent out shader constants. */ paramList = fp->mesa_program.Base.Parameters; - if(NULL != paramList) { + if(NULL != paramList) + { _mesa_load_state_parameters(ctx, paramList); if (paramList->NumParameters > R700_MAX_DX9_CONSTS) @@ -564,10 +575,10 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) unNumParamData = paramList->NumParameters; for(ui=0; uips.consts[ui][0].f32All = paramList->ParameterValues[ui][0]; - r700->ps.consts[ui][1].f32All = paramList->ParameterValues[ui][1]; - r700->ps.consts[ui][2].f32All = paramList->ParameterValues[ui][2]; - r700->ps.consts[ui][3].f32All = paramList->ParameterValues[ui][3]; + r700->ps.consts[ui][0].f32All = paramList->ParameterValues[ui][0]; + r700->ps.consts[ui][1].f32All = paramList->ParameterValues[ui][1]; + r700->ps.consts[ui][2].f32All = paramList->ParameterValues[ui][2]; + r700->ps.consts[ui][3].f32All = paramList->ParameterValues[ui][3]; } } else r700->ps.num_consts = 0; diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.h b/src/mesa/drivers/dri/r600/r700_fragprog.h index cbb108d212..843de2c029 100644 --- a/src/mesa/drivers/dri/r600/r700_fragprog.h +++ b/src/mesa/drivers/dri/r600/r700_fragprog.h @@ -49,12 +49,14 @@ struct r700_fragment_program /* Internal */ void Map_Fragment_Program(r700_AssemblerBase *pAsm, - struct gl_fragment_program *mesa_fp); + struct gl_fragment_program *mesa_fp, + GLcontext *ctx); //richard glsl nov.27 GLboolean Find_Instruction_Dependencies_fp(struct r700_fragment_program *fp, struct gl_fragment_program *mesa_fp); GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, - struct gl_fragment_program *mesa_vp); + struct gl_fragment_program *mesa_vp, + GLcontext *ctx); //richard glsl nov.27 /* Interface */ extern void r700SelectFragmentShader(GLcontext *ctx); diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index c8f72d588b..7715214da1 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -631,6 +631,14 @@ GLboolean r700SetupVertexProgram(GLcontext * ctx) paramList = vp->mesa_program->Base.Parameters; if(NULL != paramList) { + //vp->mesa_program was cloned, not updated by glsl shader api. + //_mesa_reference_program has already checked glsl shProg is ok and set ctx->VertexProgem._Current + // so, use ctx->VertexProgem._Current + struct gl_program_parameter_list *paramListOrginal = + paramListOrginal = ctx->VertexProgram._Current->Base.Parameters; + + //--------------------------- + _mesa_load_state_parameters(ctx, paramList); if (paramList->NumParameters > R700_MAX_DX9_CONSTS) @@ -643,10 +651,20 @@ GLboolean r700SetupVertexProgram(GLcontext * ctx) unNumParamData = paramList->NumParameters; for(ui=0; uivs.consts[ui][0].f32All = paramList->ParameterValues[ui][0]; - r700->vs.consts[ui][1].f32All = paramList->ParameterValues[ui][1]; - r700->vs.consts[ui][2].f32All = paramList->ParameterValues[ui][2]; - r700->vs.consts[ui][3].f32All = paramList->ParameterValues[ui][3]; + if(paramList->Parameters[ui].Type == PROGRAM_UNIFORM) + { + r700->vs.consts[ui][0].f32All = paramListOrginal->ParameterValues[ui][0]; + r700->vs.consts[ui][1].f32All = paramListOrginal->ParameterValues[ui][1]; + r700->vs.consts[ui][2].f32All = paramListOrginal->ParameterValues[ui][2]; + r700->vs.consts[ui][3].f32All = paramListOrginal->ParameterValues[ui][3]; + } + else + { + r700->vs.consts[ui][0].f32All = paramList->ParameterValues[ui][0]; + r700->vs.consts[ui][1].f32All = paramList->ParameterValues[ui][1]; + r700->vs.consts[ui][2].f32All = paramList->ParameterValues[ui][2]; + r700->vs.consts[ui][3].f32All = paramList->ParameterValues[ui][3]; + } } } else r700->vs.num_consts = 0; -- cgit v1.2.3 From a1b9c4e22a83d2125f66c3a3af3143bc0daee9a4 Mon Sep 17 00:00:00 2001 From: Richard Li Date: Sun, 29 Nov 2009 12:28:32 -0500 Subject: r600 : clena up a bit for last commit. --- src/mesa/drivers/dri/r600/r700_fragprog.h | 4 ++-- src/mesa/drivers/dri/r600/r700_vertprog.c | 10 ++++------ 2 files changed, 6 insertions(+), 8 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.h b/src/mesa/drivers/dri/r600/r700_fragprog.h index 843de2c029..e562bfa478 100644 --- a/src/mesa/drivers/dri/r600/r700_fragprog.h +++ b/src/mesa/drivers/dri/r600/r700_fragprog.h @@ -50,13 +50,13 @@ struct r700_fragment_program /* Internal */ void Map_Fragment_Program(r700_AssemblerBase *pAsm, struct gl_fragment_program *mesa_fp, - GLcontext *ctx); //richard glsl nov.27 + GLcontext *ctx); GLboolean Find_Instruction_Dependencies_fp(struct r700_fragment_program *fp, struct gl_fragment_program *mesa_fp); GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, struct gl_fragment_program *mesa_vp, - GLcontext *ctx); //richard glsl nov.27 + GLcontext *ctx); /* Interface */ extern void r700SelectFragmentShader(GLcontext *ctx); diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index 7715214da1..d3d1da7959 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -631,14 +631,12 @@ GLboolean r700SetupVertexProgram(GLcontext * ctx) paramList = vp->mesa_program->Base.Parameters; if(NULL != paramList) { - //vp->mesa_program was cloned, not updated by glsl shader api. - //_mesa_reference_program has already checked glsl shProg is ok and set ctx->VertexProgem._Current - // so, use ctx->VertexProgem._Current + /* vp->mesa_program was cloned, not updated by glsl shader api. */ + /* _mesa_reference_program has already checked glsl shProg is ok and set ctx->VertexProgem._Current */ + /* so, use ctx->VertexProgem._Current */ struct gl_program_parameter_list *paramListOrginal = paramListOrginal = ctx->VertexProgram._Current->Base.Parameters; - - //--------------------------- - + _mesa_load_state_parameters(ctx, paramList); if (paramList->NumParameters > R700_MAX_DX9_CONSTS) -- cgit v1.2.3 From 50ab51101e4c5e64ab92c260c324fb66a98851a8 Mon Sep 17 00:00:00 2001 From: Andre Maasikas Date: Fri, 4 Dec 2009 16:36:41 +0200 Subject: r600: glsl - allow specifying texture sampler via uniforms looks kinda hackish, should rethink later --- src/mesa/drivers/dri/r600/r700_assembler.c | 2 +- src/mesa/drivers/dri/r600/r700_assembler.h | 1 + src/mesa/drivers/dri/r600/r700_fragprog.c | 5 +++++ src/mesa/drivers/dri/r600/r700_vertprog.c | 4 ++++ 4 files changed, 11 insertions(+), 1 deletion(-) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index 3738edbc2f..158c5fa549 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -4840,7 +4840,7 @@ GLboolean assemble_TEX(r700_AssemblerBase *pAsm) pAsm->need_tex_barrier = GL_TRUE; } // Set src1 to tex unit id - pAsm->S[1].src.reg = pAsm->pILInst[pAsm->uiCurInst].TexSrcUnit; + pAsm->S[1].src.reg = pAsm->SamplerUnits[pAsm->pILInst[pAsm->uiCurInst].TexSrcUnit]; pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; //No sw info from mesa compiler, so hard code here. diff --git a/src/mesa/drivers/dri/r600/r700_assembler.h b/src/mesa/drivers/dri/r600/r700_assembler.h index ef1f924add..48ffef501f 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.h +++ b/src/mesa/drivers/dri/r600/r700_assembler.h @@ -416,6 +416,7 @@ typedef struct r700_AssemblerBase SHADER_PIPE_TYPE currentShaderType; struct prog_instruction * pILInst; GLuint uiCurInst; + GLubyte SamplerUnits[MAX_SAMPLERS]; GLboolean bR6xx; /* helper to decide which type of instruction to assemble */ GLboolean is_tex; diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.c b/src/mesa/drivers/dri/r600/r700_fragprog.c index 0cb9707ee6..8eb439a951 100644 --- a/src/mesa/drivers/dri/r600/r700_fragprog.c +++ b/src/mesa/drivers/dri/r600/r700_fragprog.c @@ -308,6 +308,7 @@ GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, GLuint number_of_colors_exported; GLboolean z_enabled = GL_FALSE; GLuint unBit; + int i; //Init_Program Init_r700_AssemblerBase( SPT_FP, &(fp->r700AsmCode), &(fp->r700Shader) ); @@ -320,6 +321,10 @@ GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, InitShaderProgram(&(fp->r700AsmCode)); + for(i=0; i < MAX_SAMPLERS; i++) + { + fp->r700AsmCode.SamplerUnits[i] = fp->mesa_program.Base.SamplerUnits[i]; + } if( GL_FALSE == AssembleInstr(0, mesa_fp->Base.NumInstructions, &(mesa_fp->Base.Instructions[0]), diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index d3d1da7959..759b74dc7e 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -337,6 +337,10 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, InitShaderProgram(&(vp->r700AsmCode)); + for(i=0; i < MAX_SAMPLERS; i++) + { + vp->r700AsmCode.SamplerUnits[i] = vp->mesa_program->Base.SamplerUnits[i]; + } if(GL_FALSE == AssembleInstr(0, vp->mesa_program->Base.NumInstructions, &(vp->mesa_program->Base.Instructions[0]), -- cgit v1.2.3 From 8927b72118f9433aafd0e811cfc1981215eb3c5f Mon Sep 17 00:00:00 2001 From: Richard Li Date: Wed, 9 Dec 2009 15:39:16 -0500 Subject: r600 : add pre-compile mesa shader calling interface, in order to handle complex built-in shader instructions. --- src/mesa/drivers/dri/r600/r700_assembler.c | 407 +++++++++++++++++++++++++++-- src/mesa/drivers/dri/r600/r700_assembler.h | 65 ++++- src/mesa/drivers/dri/r600/r700_fragprog.c | 25 +- src/mesa/drivers/dri/r600/r700_vertprog.c | 25 +- 4 files changed, 498 insertions(+), 24 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index aed84fc3bd..e84f524525 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -32,6 +32,7 @@ #include "main/mtypes.h" #include "main/imports.h" +#include "shader/prog_parameter.h" #include "radeon_debug.h" #include "r600_context.h" @@ -41,6 +42,39 @@ #define USE_CF_FOR_CONTINUE_BREAK 1 #define USE_CF_FOR_POP_AFTER 1 +struct prog_instruction noise1_insts[12] = { + {OPCODE_BGNSUB , {{13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {13, 0, 15, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_MOV , {{0, 0, 0, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {0, 0, 2, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_MOV , {{8, 0, 0, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {0, 0, 4, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_MOV , {{8, 0, 585, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {0, 0, 8, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_SGT , {{0, 0, 585, 0, 0, 0}, {8, 0, 1170, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {0, 1, 1, 0, 8, 1672, 0}, 1, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_IF , {{13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {13, 0, 15, 0, 7, 0, 0}, 0, 0, 0, 1, 0, 0, 0, 15, 0, 0, 0}, + {OPCODE_MOV , {{0, 0, 1755, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {0, 0, 1, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_RET , {{13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {13, 0, 15, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_ENDIF , {{13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {13, 0, 15, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_MOV , {{0, 0, 1170, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {0, 0, 1, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_RET , {{13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {13, 0, 15, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0}, + {OPCODE_ENDSUB , {{13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}, {13, 0, 1672, 0, 0, 0}}, {13, 0, 15, 0, 8, 1672, 0}, 0, 0, 0, 1, 0, 0, 0, -1, 0, 0, 0} +}; +float noise1_const[2][4] = { + {0.300000f, 0.900000f, 0.500000f, 0.300000f} +}; + +COMPILED_SUB noise1_presub = { + &(noise1_insts[0]), + 12, + 2, + 1, + 0, + &(noise1_const[0]), + SWIZZLE_X, + SWIZZLE_X, + SWIZZLE_X, + SWIZZLE_X, + {0,0,0}, + 0 +}; + BITS addrmode_PVSDST(PVSDST * pPVSDST) { return pPVSDST->addrmode0 | ((BITS)pPVSDST->addrmode1 << 1); @@ -330,14 +364,14 @@ GLuint GetSurfaceFormat(GLenum eType, GLuint nChannels, GLuint * pClient_size) return(format); } -unsigned int r700GetNumOperands(r700_AssemblerBase* pAsm) +unsigned int r700GetNumOperands(GLuint opcode, GLuint nIsOp3) { - if(pAsm->D.dst.op3) + if(nIsOp3 > 0) { return 3; } - switch (pAsm->D.dst.opcode) + switch (opcode) { case SQ_OP2_INST_ADD: case SQ_OP2_INST_KILLE: @@ -378,7 +412,7 @@ unsigned int r700GetNumOperands(r700_AssemblerBase* pAsm) return 1; default: radeon_error( - "Need instruction operand number for %x.\n", pAsm->D.dst.opcode); + "Need instruction operand number for %x.\n", opcode); }; return 3; @@ -500,6 +534,11 @@ int Init_r700_AssemblerBase(SHADER_PIPE_TYPE spt, r700_AssemblerBase* pAsm, R700 pAsm->unCFflags = 0; + pAsm->presubs = NULL; + pAsm->unPresubArraySize = 0; + pAsm->unNumPresub = 0; + pAsm->unCurNumILInsts = 0; + return 0; } @@ -2010,7 +2049,7 @@ GLboolean check_scalar(r700_AssemblerBase* pAsm, GLuint swizzle_key; - GLuint number_of_operands = r700GetNumOperands(pAsm); + GLuint number_of_operands = r700GetNumOperands(pAsm->D.dst.opcode, pAsm->D.dst.op3); for (src=0; srcD.dst.opcode, pAsm->D.dst.op3); for (src=0; srcD.dst.opcode, pAsm->D.dst.op3); //GLuint channel_swizzle, j; //GLuint chan_counter[4] = {0, 0, 0, 0}; //PVSSRC * pSource[3]; @@ -4968,7 +5007,7 @@ void add_return_inst(r700_AssemblerBase *pAsm) pAsm->cf_current_cf_clause_ptr->m_Word1.f.barrier = 0x1; } -GLboolean assemble_BGNSUB(r700_AssemblerBase *pAsm, GLint nILindex) +GLboolean assemble_BGNSUB(r700_AssemblerBase *pAsm, GLint nILindex, GLuint uiIL_Shift) { /* Put in sub */ if( (pAsm->unSubArrayPointer + 1) > pAsm->unSubArraySize ) @@ -4983,7 +5022,7 @@ GLboolean assemble_BGNSUB(r700_AssemblerBase *pAsm, GLint nILindex) pAsm->unSubArraySize += 10; } - pAsm->subs[pAsm->unSubArrayPointer].subIL_Offset = nILindex; + pAsm->subs[pAsm->unSubArrayPointer].subIL_Offset = nILindex + uiIL_Shift; pAsm->subs[pAsm->unSubArrayPointer].lstCFInstructions_local.pHead=NULL; pAsm->subs[pAsm->unSubArrayPointer].lstCFInstructions_local.pTail=NULL; pAsm->subs[pAsm->unSubArrayPointer].lstCFInstructions_local.uNumOfNode=0; @@ -5074,9 +5113,13 @@ GLboolean assemble_RET(r700_AssemblerBase *pAsm) GLboolean assemble_CAL(r700_AssemblerBase *pAsm, GLint nILindex, + GLuint uiIL_Shift, GLuint uiNumberInsts, - struct prog_instruction *pILInst) + struct prog_instruction *pILInst, + PRESUB_DESC * pPresubDesc) { + GLint uiIL_Offset; + pAsm->alu_x_opcode = SQ_CF_INST_ALU; if(GL_FALSE == add_cf_instruction(pAsm) ) @@ -5109,8 +5152,12 @@ GLboolean assemble_CAL(r700_AssemblerBase *pAsm, pAsm->unCallerArraySize += 10; } - pAsm->callers[pAsm->unCallerArrayPointer].subIL_Offset = nILindex; - pAsm->callers[pAsm->unCallerArrayPointer].cf_ptr = pAsm->cf_current_cf_clause_ptr; + uiIL_Offset = nILindex + uiIL_Shift; + pAsm->callers[pAsm->unCallerArrayPointer].subIL_Offset = uiIL_Offset; + pAsm->callers[pAsm->unCallerArrayPointer].cf_ptr = pAsm->cf_current_cf_clause_ptr; + + pAsm->callers[pAsm->unCallerArrayPointer].finale_cf_ptr = NULL; + pAsm->callers[pAsm->unCallerArrayPointer].prelude_cf_ptr = NULL; pAsm->unCallerArrayPointer++; @@ -5120,7 +5167,7 @@ GLboolean assemble_CAL(r700_AssemblerBase *pAsm, GLboolean bRet; for(j=0; junSubArrayPointer; j++) { - if(nILindex == pAsm->subs[j].subIL_Offset) + if(uiIL_Offset == pAsm->subs[j].subIL_Offset) { /* compiled before */ max = pAsm->subs[j].unStackDepthMax @@ -5138,7 +5185,7 @@ GLboolean assemble_CAL(r700_AssemblerBase *pAsm, pAsm->callers[pAsm->unCallerArrayPointer - 1].subDescIndex = pAsm->unSubArrayPointer; unSubID = pAsm->unSubArrayPointer; - bRet = AssembleInstr(nILindex, uiNumberInsts, pILInst, pAsm); + bRet = AssembleInstr(nILindex, uiIL_Shift, uiNumberInsts, pILInst, pAsm); if(GL_TRUE == bRet) { @@ -5148,6 +5195,8 @@ GLboolean assemble_CAL(r700_AssemblerBase *pAsm, { pAsm->CALLSTACK[pAsm->CALLSP].max = max; } + + pAsm->subs[unSubID].pPresubDesc = pPresubDesc; } return bRet; @@ -5313,6 +5362,7 @@ GLboolean breakLoopOnFlag(r700_AssemblerBase *pAsm, GLuint unFCSP) } GLboolean AssembleInstr(GLuint uiFirstInst, + GLuint uiIL_Shift, GLuint uiNumberInsts, struct prog_instruction *pILInst, r700_AssemblerBase *pR700AsmCode) @@ -5468,6 +5518,26 @@ GLboolean AssembleInstr(GLuint uiFirstInst, case OPCODE_MUL: if ( GL_FALSE == assemble_MUL(pR700AsmCode) ) return GL_FALSE; + break; + + case OPCODE_NOISE1: + { + callPreSub(pR700AsmCode, + GLSL_NOISE1, + &noise1_presub, + pILInst->DstReg.Index + pR700AsmCode->starting_temp_register_number, + 1); + radeon_error("noise1: not yet supported shader instruction\n"); + }; + break; + case OPCODE_NOISE2: + radeon_error("noise2: not yet supported shader instruction\n"); + break; + case OPCODE_NOISE3: + radeon_error("noise3: not yet supported shader instruction\n"); + break; + case OPCODE_NOISE4: + radeon_error("noise4: not yet supported shader instruction\n"); break; case OPCODE_POW: @@ -5653,7 +5723,7 @@ GLboolean AssembleInstr(GLuint uiFirstInst, break; case OPCODE_BGNSUB: - if( GL_FALSE == assemble_BGNSUB(pR700AsmCode, i) ) + if( GL_FALSE == assemble_BGNSUB(pR700AsmCode, i, uiIL_Shift) ) { return GL_FALSE; } @@ -5668,9 +5738,11 @@ GLboolean AssembleInstr(GLuint uiFirstInst, case OPCODE_CAL: if( GL_FALSE == assemble_CAL(pR700AsmCode, - pILInst[i].BranchTarget, + pILInst[i].BranchTarget, + uiIL_Shift, uiNumberInsts, - pILInst) ) + pILInst, + NULL) ) { return GL_FALSE; } @@ -5707,7 +5779,7 @@ GLboolean InitShaderProgram(r700_AssemblerBase * pAsm) return GL_TRUE; } -GLboolean RelocProgram(r700_AssemblerBase * pAsm) +GLboolean RelocProgram(r700_AssemblerBase * pAsm, struct gl_program * pILProg) { GLuint i; GLuint unCFoffset; @@ -5717,6 +5789,12 @@ GLboolean RelocProgram(r700_AssemblerBase * pAsm) R700ShaderInstruction * pInst; R700ControlFlowGenericClause * pCFInst; + R700ControlFlowALUClause * pCF_ALU; + R700ALUInstruction * pALU; + GLuint unConstOffset = 0; + GLuint unRegOffset; + GLuint unMinRegIndex; + plstCFmain = pAsm->CALLSTACK[0].plstCFInstructions_local; /* remove flags init if they are not used */ @@ -5762,6 +5840,11 @@ GLboolean RelocProgram(r700_AssemblerBase * pAsm) unCFoffset = plstCFmain->uNumOfNode; + if(NULL != pILProg->Parameters) + { + unConstOffset = pILProg->Parameters->NumParameters; + } + /* Reloc subs */ for(i=0; iunSubArrayPointer; i++) { @@ -5799,6 +5882,84 @@ GLboolean RelocProgram(r700_AssemblerBase * pAsm) pInst = pInst->pNextInst; }; + if(NULL != pAsm->subs[i].pPresubDesc) + { + GLuint uNumSrc; + + unMinRegIndex = pAsm->subs[i].pPresubDesc->pCompiledSub->MinRegIndex; + unRegOffset = pAsm->subs[i].pPresubDesc->maxStartReg; + unConstOffset += pAsm->subs[i].pPresubDesc->unConstantsStart; + + pInst = plstCFsub->pHead; + while(pInst) + { + if(SIT_CF_ALU == pInst->m_ShaderInstType) + { + pCF_ALU = (R700ControlFlowALUClause *)pInst; + + pALU = pCF_ALU->m_pLinkedALUInstruction; + for(int j=0; j<=pCF_ALU->m_Word1.f.count; j++) + { + pALU->m_Word1.f.dst_gpr = pALU->m_Word1.f.dst_gpr + unRegOffset - unMinRegIndex; + + if(pALU->m_Word0.f.src0_sel < SQ_ALU_SRC_GPR_SIZE) + { + pALU->m_Word0.f.src0_sel = pALU->m_Word0.f.src0_sel + unRegOffset - unMinRegIndex; + } + else if(pALU->m_Word0.f.src0_sel >= SQ_ALU_SRC_CFILE_BASE) + { + pALU->m_Word0.f.src0_sel += unConstOffset; + } + + if( ((pALU->m_Word1.val >> SQ_ALU_WORD1_OP3_ALU_INST_SHIFT) & 0x0000001F) + >= SQ_OP3_INST_MUL_LIT ) + { /* op3 : 3 srcs */ + if(pALU->m_Word1_OP3.f.src2_sel < SQ_ALU_SRC_GPR_SIZE) + { + pALU->m_Word1_OP3.f.src2_sel = pALU->m_Word1_OP3.f.src2_sel + unRegOffset - unMinRegIndex; + } + else if(pALU->m_Word1_OP3.f.src2_sel >= SQ_ALU_SRC_CFILE_BASE) + { + pALU->m_Word1_OP3.f.src2_sel += unConstOffset; + } + if(pALU->m_Word0.f.src1_sel < SQ_ALU_SRC_GPR_SIZE) + { + pALU->m_Word0.f.src1_sel = pALU->m_Word0.f.src1_sel + unRegOffset - unMinRegIndex; + } + else if(pALU->m_Word0.f.src1_sel >= SQ_ALU_SRC_CFILE_BASE) + { + pALU->m_Word0.f.src1_sel += unConstOffset; + } + } + else + { + if(pAsm->bR6xx) + { + uNumSrc = r700GetNumOperands(pALU->m_Word1_OP2.f6.alu_inst, 0); + } + else + { + uNumSrc = r700GetNumOperands(pALU->m_Word1_OP2.f.alu_inst, 0); + } + if(2 == uNumSrc) + { /* 2 srcs */ + if(pALU->m_Word0.f.src1_sel < SQ_ALU_SRC_GPR_SIZE) + { + pALU->m_Word0.f.src1_sel = pALU->m_Word0.f.src1_sel + unRegOffset - unMinRegIndex; + } + else if(pALU->m_Word0.f.src1_sel >= SQ_ALU_SRC_CFILE_BASE) + { + pALU->m_Word0.f.src1_sel += unConstOffset; + } + } + } + pALU = (R700ALUInstruction*)(pALU->pNextInst); + } + } + pInst = pInst->pNextInst; + }; + } + /* Put sub into main */ plstCFmain->pTail->pNextInst = plstCFsub->pHead; plstCFmain->pTail = plstCFsub->pTail; @@ -5812,11 +5973,216 @@ GLboolean RelocProgram(r700_AssemblerBase * pAsm) { pAsm->callers[i].cf_ptr->m_Word0.f.addr = pAsm->subs[pAsm->callers[i].subDescIndex].unCFoffset; + + if(NULL != pAsm->subs[pAsm->callers[i].subDescIndex].pPresubDesc) + { + unMinRegIndex = pAsm->subs[pAsm->callers[i].subDescIndex].pPresubDesc->pCompiledSub->MinRegIndex; + unRegOffset = pAsm->subs[pAsm->callers[i].subDescIndex].pPresubDesc->maxStartReg; + + if(NULL != pAsm->callers[i].prelude_cf_ptr) + { + pCF_ALU = (R700ControlFlowALUClause * )(pAsm->callers[i].prelude_cf_ptr); + pALU = pCF_ALU->m_pLinkedALUInstruction; + for(int j=0; j<=pCF_ALU->m_Word1.f.count; j++) + { + pALU->m_Word1.f.dst_gpr = pALU->m_Word1.f.dst_gpr + unRegOffset - unMinRegIndex; + pALU = (R700ALUInstruction*)(pALU->pNextInst); + } + } + if(NULL != pAsm->callers[i].finale_cf_ptr) + { + pCF_ALU = (R700ControlFlowALUClause * )(pAsm->callers[i].finale_cf_ptr); + pALU = pCF_ALU->m_pLinkedALUInstruction; + for(int j=0; j<=pCF_ALU->m_Word1.f.count; j++) + { + pALU->m_Word0.f.src0_sel = pALU->m_Word0.f.src0_sel + unRegOffset - unMinRegIndex; + pALU = (R700ALUInstruction*)(pALU->pNextInst); + } + } + } } return GL_TRUE; } +GLboolean callPreSub(r700_AssemblerBase* pAsm, + LOADABLE_SCRIPT_SIGNITURE scriptSigniture, + COMPILED_SUB * pCompiledSub, + GLshort uOutReg, + GLshort uNumValidSrc) +{ + /* save assemble context */ + GLuint starting_temp_register_number_save; + GLuint number_used_registers_save; + GLuint uFirstHelpReg_save; + GLuint uHelpReg_save; + GLuint uiCurInst_save; + struct prog_instruction *pILInst_save; + PRESUB_DESC * pPresubDesc; + GLboolean bRet; + int i; + + R700ControlFlowGenericClause* prelude_cf_ptr = NULL; + + /* copy srcs to presub inputs */ + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + for(i=0; iD.dst.opcode = SQ_OP2_INST_MOV; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = pCompiledSub->srcRegIndex[i]; + pAsm->D.dst.writex = 1; + pAsm->D.dst.writey = 1; + pAsm->D.dst.writez = 1; + pAsm->D.dst.writew = 1; + + if( GL_FALSE == assemble_src(pAsm, i, 0) ) + { + return GL_FALSE; + } + + next_ins(pAsm); + } + if(uNumValidSrc > 0) + { + prelude_cf_ptr = pAsm->cf_current_alu_clause_ptr; + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + } + + /* browse thro existing presubs. */ + for(i=0; iunNumPresub; i++) + { + if(pAsm->presubs[i].sptSigniture == scriptSigniture) + { + break; + } + } + + if(i == pAsm->unNumPresub) + { /* not loaded yet */ + /* save assemble context */ + number_used_registers_save = pAsm->number_used_registers; + uFirstHelpReg_save = pAsm->uFirstHelpReg; + uHelpReg_save = pAsm->uHelpReg; + starting_temp_register_number_save = pAsm->starting_temp_register_number; + pILInst_save = pAsm->pILInst; + uiCurInst_save = pAsm->uiCurInst; + + /* alloc in presub */ + if( (pAsm->unNumPresub + 1) > pAsm->unPresubArraySize ) + { + pAsm->presubs = (PRESUB_DESC*)_mesa_realloc( (void *)pAsm->presubs, + sizeof(PRESUB_DESC) * pAsm->unPresubArraySize, + sizeof(PRESUB_DESC) * (pAsm->unPresubArraySize + 4) ); + if(NULL == pAsm->presubs) + { + radeon_error("No memeory to allocate built in shader function description structures. \n"); + return GL_FALSE; + } + pAsm->unPresubArraySize += 4; + } + + pPresubDesc = &(pAsm->presubs[i]); + pPresubDesc->sptSigniture = scriptSigniture; + + /* constants offsets need to be final resolved at reloc. */ + if(0 == pAsm->unNumPresub) + { + pPresubDesc->unConstantsStart = 0; + } + else + { + pPresubDesc->unConstantsStart = pAsm->presubs[i-1].unConstantsStart + + pAsm->presubs[i-1].pCompiledSub->NumParameters; + } + + pPresubDesc->pCompiledSub = pCompiledSub; + + pPresubDesc->subIL_Shift = pAsm->unCurNumILInsts; + pPresubDesc->maxStartReg = uFirstHelpReg_save; + pAsm->unCurNumILInsts += pCompiledSub->NumInstructions; + + pAsm->unNumPresub++; + + /* setup new assemble context */ + pAsm->starting_temp_register_number = 0; + pAsm->number_used_registers = pCompiledSub->NumTemporaries; + pAsm->uFirstHelpReg = pAsm->number_used_registers; + pAsm->uHelpReg = pAsm->uFirstHelpReg; + + bRet = assemble_CAL(pAsm, + 0, + pPresubDesc->subIL_Shift, + pCompiledSub->NumInstructions, + pCompiledSub->Instructions, + pPresubDesc); + + + pPresubDesc->number_used_registers = pAsm->number_used_registers; + + /* restore assemble context */ + pAsm->number_used_registers = number_used_registers_save; + pAsm->uFirstHelpReg = uFirstHelpReg_save; + pAsm->uHelpReg = uHelpReg_save; + pAsm->starting_temp_register_number = starting_temp_register_number_save; + pAsm->pILInst = pILInst_save; + pAsm->uiCurInst = uiCurInst_save; + } + else + { /* was loaded */ + pPresubDesc = &(pAsm->presubs[i]); + + bRet = assemble_CAL(pAsm, + 0, + pPresubDesc->subIL_Shift, + pCompiledSub->NumInstructions, + pCompiledSub->Instructions, + pPresubDesc); + } + + if(GL_FALSE == bRet) + { + radeon_error("Shader presub assemble failed. \n"); + } + else + { + /* copy presub output to real dst */ + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + + if( GL_FALSE == assemble_dst(pAsm) ) + { + return GL_FALSE; + } + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = pCompiledSub->dstRegIndex; + pAsm->S[0].src.swizzlex = pCompiledSub->outputSwizzleX; + pAsm->S[0].src.swizzley = pCompiledSub->outputSwizzleY; + pAsm->S[0].src.swizzlez = pCompiledSub->outputSwizzleZ; + pAsm->S[0].src.swizzlew = pCompiledSub->outputSwizzleW; + + next_ins(pAsm); + + pAsm->callers[pAsm->unCallerArrayPointer - 1].finale_cf_ptr = pAsm->cf_current_alu_clause_ptr; + pAsm->callers[pAsm->unCallerArrayPointer - 1].prelude_cf_ptr = prelude_cf_ptr; + pAsm->alu_x_opcode = SQ_CF_INST_ALU; + } + + if( (pPresubDesc->number_used_registers + pAsm->uFirstHelpReg) > pAsm->number_used_registers ) + { + pAsm->number_used_registers = pPresubDesc->number_used_registers + pAsm->uFirstHelpReg; + } + if(pAsm->uFirstHelpReg > pPresubDesc->maxStartReg) + { + pPresubDesc->maxStartReg = pAsm->uFirstHelpReg; + } + + return bRet; +} + GLboolean Process_Export(r700_AssemblerBase* pAsm, GLuint type, GLuint export_starting_index, @@ -6174,6 +6540,11 @@ GLboolean Clean_Up_Assembler(r700_AssemblerBase *pR700AsmCode) FREE(pR700AsmCode->callers); } + if(NULL != pR700AsmCode->presubs) + { + FREE(pR700AsmCode->presubs); + } + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r600/r700_assembler.h b/src/mesa/drivers/dri/r600/r700_assembler.h index 6dc44017eb..6ef945dfda 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.h +++ b/src/mesa/drivers/dri/r600/r700_assembler.h @@ -34,6 +34,45 @@ #include "r700_shaderinst.h" #include "r700_shader.h" +typedef enum LOADABLE_SCRIPT_SIGNITURE +{ + GLSL_NOISE1 = 0x10000001, + GLSL_NOISE2 = 0x10000002, + GLSL_NOISE3 = 0x10000003, + GLSL_NOISE4 = 0x10000004 +}LOADABLE_SCRIPT_SIGNITURE; + +typedef struct COMPILED_SUB +{ + struct prog_instruction *Instructions; + GLuint NumInstructions; + GLuint NumTemporaries; + GLuint NumParameters; + GLuint MinRegIndex; + GLfloat (*ParameterValues)[4]; + GLbyte outputSwizzleX; + GLbyte outputSwizzleY; + GLbyte outputSwizzleZ; + GLbyte outputSwizzleW; + GLshort srcRegIndex[3]; + GLushort dstRegIndex; +}COMPILED_SUB; + +typedef struct PRESUB_DESCtag +{ + LOADABLE_SCRIPT_SIGNITURE sptSigniture; + GLint subIL_Shift; + struct prog_src_register InReg[3]; + struct prog_dst_register OutReg; + + GLushort maxStartReg; + GLushort number_used_registers; + + GLuint unConstantsStart; + + COMPILED_SUB * pCompiledSub; +} PRESUB_DESC; + typedef enum SHADER_PIPE_TYPE { SPT_VP = 0, @@ -296,6 +335,7 @@ typedef struct SUB_OFFSET GLint subIL_Offset; GLuint unCFoffset; GLuint unStackDepthMax; + PRESUB_DESC * pPresubDesc; TypedShaderList lstCFInstructions_local; } SUB_OFFSET; @@ -304,6 +344,9 @@ typedef struct CALLER_POINTER GLint subIL_Offset; GLint subDescIndex; R700ControlFlowGenericClause* cf_ptr; + + R700ControlFlowGenericClause* prelude_cf_ptr; + R700ControlFlowGenericClause* finale_cf_ptr; } CALLER_POINTER; #define SQ_MAX_CALL_DEPTH 0x00000020 @@ -437,6 +480,11 @@ typedef struct r700_AssemblerBase GLuint unCFflags; + PRESUB_DESC * presubs; + GLuint unPresubArraySize; + GLuint unNumPresub; + GLuint unCurNumILInsts; + } r700_AssemblerBase; //Internal use @@ -458,7 +506,7 @@ BITS is_depth_component_exported(OUT_FRAGMENT_FMT_0* pFPOutFmt) ; GLboolean is_reduction_opcode(PVSDWORD * dest); GLuint GetSurfaceFormat(GLenum eType, GLuint nChannels, GLuint * pClient_size); -unsigned int r700GetNumOperands(r700_AssemblerBase* pAsm); +unsigned int r700GetNumOperands(GLuint opcode, GLuint nIsOp3); GLboolean IsTex(gl_inst_opcode Opcode); GLboolean IsAlu(gl_inst_opcode Opcode); @@ -585,13 +633,15 @@ GLboolean assemble_BRK(r700_AssemblerBase *pAsm); GLboolean assemble_COND(r700_AssemblerBase *pAsm); GLboolean assemble_ENDLOOP(r700_AssemblerBase *pAsm); -GLboolean assemble_BGNSUB(r700_AssemblerBase *pAsm, GLint nILindex); +GLboolean assemble_BGNSUB(r700_AssemblerBase *pAsm, GLint nILindex, GLuint uiIL_Shift); GLboolean assemble_ENDSUB(r700_AssemblerBase *pAsm); GLboolean assemble_RET(r700_AssemblerBase *pAsm); GLboolean assemble_CAL(r700_AssemblerBase *pAsm, GLint nILindex, + GLuint uiIL_Offest, GLuint uiNumberInsts, - struct prog_instruction *pILInst); + struct prog_instruction *pILInst, + PRESUB_DESC * pPresubDesc); GLboolean Process_Export(r700_AssemblerBase* pAsm, GLuint type, @@ -602,16 +652,23 @@ GLboolean Process_Export(r700_AssemblerBase* pAsm, GLboolean Move_Depth_Exports_To_Correct_Channels(r700_AssemblerBase *pAsm, BITS depth_channel_select); +GLboolean callPreSub(r700_AssemblerBase* pAsm, + LOADABLE_SCRIPT_SIGNITURE scriptSigniture, + /* struct prog_instruction ** pILInstParent, */ + COMPILED_SUB * pCompiledSub, + GLshort uOutReg, + GLshort uNumValidSrc); //Interface GLboolean AssembleInstr(GLuint uiFirstInst, + GLuint uiIL_Shift, GLuint uiNumberInsts, struct prog_instruction *pILInst, r700_AssemblerBase *pR700AsmCode); GLboolean Process_Fragment_Exports(r700_AssemblerBase *pR700AsmCode, GLbitfield OutputsWritten); GLboolean Process_Vertex_Exports(r700_AssemblerBase *pR700AsmCode, GLbitfield OutputsWritten); -GLboolean RelocProgram(r700_AssemblerBase * pAsm); +GLboolean RelocProgram(r700_AssemblerBase * pAsm, struct gl_program * pILProg); GLboolean InitShaderProgram(r700_AssemblerBase * pAsm); int Init_r700_AssemblerBase(SHADER_PIPE_TYPE spt, r700_AssemblerBase* pAsm, R700_Shader* pShader); diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.c b/src/mesa/drivers/dri/r600/r700_fragprog.c index 8eb439a951..d15f013710 100644 --- a/src/mesa/drivers/dri/r600/r700_fragprog.c +++ b/src/mesa/drivers/dri/r600/r700_fragprog.c @@ -325,7 +325,11 @@ GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, { fp->r700AsmCode.SamplerUnits[i] = fp->mesa_program.Base.SamplerUnits[i]; } + + fp->r700AsmCode.unCurNumILInsts = mesa_fp->Base.NumInstructions; + if( GL_FALSE == AssembleInstr(0, + 0, mesa_fp->Base.NumInstructions, &(mesa_fp->Base.Instructions[0]), &(fp->r700AsmCode)) ) @@ -338,7 +342,7 @@ GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp, return GL_FALSE; } - if( GL_FALSE == RelocProgram(&(fp->r700AsmCode)) ) + if( GL_FALSE == RelocProgram(&(fp->r700AsmCode), &(mesa_fp->Base)) ) { return GL_FALSE; } @@ -620,6 +624,25 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) } else r700->ps.num_consts = 0; + COMPILED_SUB * pCompiledSub; + GLuint uj; + GLuint unConstOffset = r700->ps.num_consts; + for(ui=0; uiunNumPresub; ui++) + { + pCompiledSub = pAsm->presubs[ui].pCompiledSub; + + r700->ps.num_consts += pCompiledSub->NumParameters; + + for(uj=0; ujNumParameters; uj++) + { + r700->ps.consts[uj + unConstOffset][0].f32All = pCompiledSub->ParameterValues[uj][0]; + r700->ps.consts[uj + unConstOffset][1].f32All = pCompiledSub->ParameterValues[uj][1]; + r700->ps.consts[uj + unConstOffset][2].f32All = pCompiledSub->ParameterValues[uj][2]; + r700->ps.consts[uj + unConstOffset][3].f32All = pCompiledSub->ParameterValues[uj][3]; + } + unConstOffset += pCompiledSub->NumParameters; + } + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index 759b74dc7e..90fac078ff 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -341,7 +341,11 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, { vp->r700AsmCode.SamplerUnits[i] = vp->mesa_program->Base.SamplerUnits[i]; } + + vp->r700AsmCode.unCurNumILInsts = vp->mesa_program->Base.NumInstructions; + if(GL_FALSE == AssembleInstr(0, + 0, vp->mesa_program->Base.NumInstructions, &(vp->mesa_program->Base.Instructions[0]), &(vp->r700AsmCode)) ) @@ -354,7 +358,7 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, return NULL; } - if( GL_FALSE == RelocProgram(&(vp->r700AsmCode)) ) + if( GL_FALSE == RelocProgram(&(vp->r700AsmCode), &(vp->mesa_program->Base)) ) { return GL_FALSE; } @@ -671,5 +675,24 @@ GLboolean r700SetupVertexProgram(GLcontext * ctx) } else r700->vs.num_consts = 0; + COMPILED_SUB * pCompiledSub; + GLuint uj; + GLuint unConstOffset = r700->vs.num_consts; + for(ui=0; uir700AsmCode.unNumPresub; ui++) + { + pCompiledSub = vp->r700AsmCode.presubs[ui].pCompiledSub; + + r700->vs.num_consts += pCompiledSub->NumParameters; + + for(uj=0; ujNumParameters; uj++) + { + r700->vs.consts[uj + unConstOffset][0].f32All = pCompiledSub->ParameterValues[uj][0]; + r700->vs.consts[uj + unConstOffset][1].f32All = pCompiledSub->ParameterValues[uj][1]; + r700->vs.consts[uj + unConstOffset][2].f32All = pCompiledSub->ParameterValues[uj][2]; + r700->vs.consts[uj + unConstOffset][3].f32All = pCompiledSub->ParameterValues[uj][3]; + } + unConstOffset += pCompiledSub->NumParameters; + } + return GL_TRUE; } -- cgit v1.2.3 From e04a818606f5639d9d8df4ebe501abb496bf847f Mon Sep 17 00:00:00 2001 From: Andre Maasikas Date: Tue, 29 Dec 2009 14:47:01 +0200 Subject: r600: support vertex_array_bgra Use vertex program key mechanism and swizzle during vertex fetch - is there a better way? --- src/mesa/drivers/dri/r600/r600_context.h | 1 + src/mesa/drivers/dri/r600/r700_assembler.c | 20 ++++++++++++++++---- src/mesa/drivers/dri/r600/r700_assembler.h | 1 + src/mesa/drivers/dri/r600/r700_vertprog.c | 8 ++++++-- src/mesa/drivers/dri/r600/r700_vertprog.h | 1 + 5 files changed, 25 insertions(+), 6 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_vertprog.c') diff --git a/src/mesa/drivers/dri/r600/r600_context.h b/src/mesa/drivers/dri/r600/r600_context.h index 394fd757d4..94662ab547 100644 --- a/src/mesa/drivers/dri/r600/r600_context.h +++ b/src/mesa/drivers/dri/r600/r600_context.h @@ -108,6 +108,7 @@ typedef struct StreamDesc GLint size; //number of data element GLenum type; //data element type GLsizei stride; + GLenum format; // GL_RGBA,GLBGRA struct radeon_bo *bo; GLint bo_offset; diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index e464c6191c..1ff89e18ea 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -891,6 +891,7 @@ GLboolean assemble_vfetch_instruction2(r700_AssemblerBase* pAsm, GLubyte element, GLuint _signed, GLboolean normalize, + GLenum format, VTX_FETCH_METHOD * pFetchMethod) { GLuint client_size_inbyte; @@ -939,10 +940,21 @@ GLboolean assemble_vfetch_instruction2(r700_AssemblerBase* pAsm, vfetch_instruction_ptr->m_Word0.f.src_sel_x = SQ_SEL_X; vfetch_instruction_ptr->m_Word0.f.mega_fetch_count = mega_fetch_count; - vfetch_instruction_ptr->m_Word1.f.dst_sel_x = (size < 1) ? SQ_SEL_0 : SQ_SEL_X; - vfetch_instruction_ptr->m_Word1.f.dst_sel_y = (size < 2) ? SQ_SEL_0 : SQ_SEL_Y; - vfetch_instruction_ptr->m_Word1.f.dst_sel_z = (size < 3) ? SQ_SEL_0 : SQ_SEL_Z; - vfetch_instruction_ptr->m_Word1.f.dst_sel_w = (size < 4) ? SQ_SEL_1 : SQ_SEL_W; + if(format == GL_BGRA) + { + vfetch_instruction_ptr->m_Word1.f.dst_sel_x = (size < 1) ? SQ_SEL_0 : SQ_SEL_Z; + vfetch_instruction_ptr->m_Word1.f.dst_sel_y = (size < 2) ? SQ_SEL_0 : SQ_SEL_Y; + vfetch_instruction_ptr->m_Word1.f.dst_sel_z = (size < 3) ? SQ_SEL_0 : SQ_SEL_X; + vfetch_instruction_ptr->m_Word1.f.dst_sel_w = (size < 4) ? SQ_SEL_1 : SQ_SEL_W; + } + else + { + vfetch_instruction_ptr->m_Word1.f.dst_sel_x = (size < 1) ? SQ_SEL_0 : SQ_SEL_X; + vfetch_instruction_ptr->m_Word1.f.dst_sel_y = (size < 2) ? SQ_SEL_0 : SQ_SEL_Y; + vfetch_instruction_ptr->m_Word1.f.dst_sel_z = (size < 3) ? SQ_SEL_0 : SQ_SEL_Z; + vfetch_instruction_ptr->m_Word1.f.dst_sel_w = (size < 4) ? SQ_SEL_1 : SQ_SEL_W; + + } vfetch_instruction_ptr->m_Word1.f.use_const_fields = 1; vfetch_instruction_ptr->m_Word1.f.data_format = data_format; diff --git a/src/mesa/drivers/dri/r600/r700_assembler.h b/src/mesa/drivers/dri/r600/r700_assembler.h index dbd9860f7d..86342b814f 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.h +++ b/src/mesa/drivers/dri/r600/r700_assembler.h @@ -531,6 +531,7 @@ GLboolean assemble_vfetch_instruction2(r700_AssemblerBase* pAsm, GLubyte element, GLuint _signed, GLboolean normalize, + GLenum format, VTX_FETCH_METHOD * pFetchMethod); GLboolean cleanup_vfetch_instructions(r700_AssemblerBase* pAsm); GLuint gethelpr(r700_AssemblerBase* pAsm); diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index 90fac078ff..782f151f5a 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -188,7 +188,8 @@ GLboolean Process_Vertex_Program_Vfetch_Instructions2( context->stream_desc[i].size, context->stream_desc[i].element, context->stream_desc[i]._signed, - context->stream_desc[i].normalize, + context->stream_desc[i].normalize, + context->stream_desc[i].format, &vtxFetchMethod); } @@ -319,6 +320,7 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx, vp->aos_desc[i].size = context->stream_desc[i].size; vp->aos_desc[i].stride = context->stream_desc[i].stride; vp->aos_desc[i].type = context->stream_desc[i].type; + vp->aos_desc[i].format = context->stream_desc[i].format; } if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770) @@ -395,7 +397,8 @@ void r700SelectVertexShader(GLcontext *ctx) match = GL_TRUE; for(i=0; inNumActiveAos; i++) { - if (vp->aos_desc[i].size != context->stream_desc[i].size) + if (vp->aos_desc[i].size != context->stream_desc[i].size || + vp->aos_desc[i].format != context->stream_desc[i].format) { match = GL_FALSE; break; @@ -498,6 +501,7 @@ static void r700TranslateAttrib(GLcontext *ctx, GLuint unLoc, int count, const s pStreamDesc->size = input->Size; pStreamDesc->dst_loc = context->nNumActiveAos; pStreamDesc->element = unLoc; + pStreamDesc->format = input->Format; switch (pStreamDesc->type) { //GetSurfaceFormat diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.h b/src/mesa/drivers/dri/r600/r700_vertprog.h index 00824c29d3..645c9ac84a 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.h +++ b/src/mesa/drivers/dri/r600/r700_vertprog.h @@ -39,6 +39,7 @@ typedef struct ArrayDesc //TEMP GLint size; //number of data element GLenum type; //data element type GLsizei stride; + GLenum format; //GL_RGBA or GL_BGRA } ArrayDesc; struct r700_vertex_program -- cgit v1.2.3