diff options
author | Brian Paul <[email protected]> | 2011-01-15 10:24:08 -0700 |
---|---|---|
committer | Brian Paul <[email protected]> | 2011-01-15 10:24:08 -0700 |
commit | 652901e95b4ed406293d0e1fabee857c054119b1 (patch) | |
tree | a4b40b134f7c6491cd8d9fe37590e99428d8fab5 /src/gallium/auxiliary | |
parent | 21001d2ba79be26f565a6795a907ec266564f4cd (diff) | |
parent | b7c38734c9d85f9dad1796d97690be2d9c55c397 (diff) |
Merge branch 'draw-instanced'
Conflicts:
src/gallium/auxiliary/draw/draw_llvm.c
src/gallium/drivers/llvmpipe/lp_state_fs.c
src/glsl/ir_set_program_inouts.cpp
src/mesa/tnl/t_vb_program.c
Diffstat (limited to 'src/gallium/auxiliary')
-rw-r--r-- | src/gallium/auxiliary/draw/draw_llvm.c | 21 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vs_exec.c | 6 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vs_sse.c | 6 | ||||
-rw-r--r-- | src/gallium/auxiliary/gallivm/lp_bld_tgsi.h | 8 | ||||
-rw-r--r-- | src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c | 73 | ||||
-rw-r--r-- | src/gallium/auxiliary/tgsi/tgsi_exec.c | 19 | ||||
-rw-r--r-- | src/gallium/auxiliary/tgsi/tgsi_exec.h | 7 | ||||
-rw-r--r-- | src/gallium/auxiliary/tgsi/tgsi_ppc.c | 4 | ||||
-rw-r--r-- | src/gallium/auxiliary/tgsi/tgsi_scan.c | 19 | ||||
-rw-r--r-- | src/gallium/auxiliary/tgsi/tgsi_scan.h | 4 | ||||
-rw-r--r-- | src/gallium/auxiliary/tgsi/tgsi_sse2.c | 54 |
11 files changed, 209 insertions, 12 deletions
diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c index 943ec44bcde..41269ee869d 100644 --- a/src/gallium/auxiliary/draw/draw_llvm.c +++ b/src/gallium/auxiliary/draw/draw_llvm.c @@ -437,6 +437,7 @@ generate_vs(struct draw_llvm *llvm, LLVMBuilderRef builder, LLVMValueRef (*outputs)[NUM_CHANNELS], const LLVMValueRef (*inputs)[NUM_CHANNELS], + LLVMValueRef system_values_array, LLVMValueRef context_ptr, struct lp_build_sampler_soa *draw_sampler) { @@ -468,6 +469,7 @@ generate_vs(struct draw_llvm *llvm, vs_type, NULL /*struct lp_build_mask_context *mask*/, consts_ptr, + system_values_array, NULL /*pos*/, inputs, outputs, @@ -1118,7 +1120,9 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) LLVMValueRef start, end, count, stride, step, io_itr; LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr; LLVMValueRef instance_id; + LLVMValueRef system_values_array; struct draw_context *draw = llvm->draw; + const struct tgsi_shader_info *vs_info = &draw->vs.vertex_shader->info; unsigned i, j; struct lp_build_context bld; struct lp_build_loop_state lp_loop; @@ -1179,6 +1183,9 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) lp_build_context_init(&bld, llvm->gallivm, lp_type_int(32)); + system_values_array = lp_build_system_values_array(gallivm, vs_info, + instance_id, NULL); + end = lp_build_add(&bld, start, count); step = lp_build_const_int32(gallivm, max_vertices); @@ -1233,6 +1240,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) builder, outputs, ptr_aos, + system_values_array, context_ptr, sampler); @@ -1263,8 +1271,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) /* store clipmask in vertex header and positions in data */ convert_to_aos(gallivm, io, outputs, clipmask, - draw->vs.vertex_shader->info.num_outputs, - max_vertices); + vs_info->num_outputs, max_vertices); } lp_build_loop_end_cond(&lp_loop, end, step, LLVMIntUGE); @@ -1315,7 +1322,9 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian LLVMValueRef fetch_elts, fetch_count, stride, step, io_itr; LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr; LLVMValueRef instance_id; + LLVMValueRef system_values_array; struct draw_context *draw = llvm->draw; + const struct tgsi_shader_info *vs_info = &draw->vs.vertex_shader->info; unsigned i, j; struct lp_build_context bld; struct lp_build_loop_state lp_loop; @@ -1376,6 +1385,10 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian lp_build_context_init(&bld, gallivm, lp_type_int(32)); + system_values_array = lp_build_system_values_array(gallivm, vs_info, + instance_id, NULL); + + step = lp_build_const_int32(gallivm, max_vertices); /* code generated texture sampling */ @@ -1438,6 +1451,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian builder, outputs, ptr_aos, + system_values_array, context_ptr, sampler); @@ -1471,8 +1485,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian * and transformed positions in data */ convert_to_aos(gallivm, io, outputs, clipmask, - draw->vs.vertex_shader->info.num_outputs, - max_vertices); + vs_info->num_outputs, max_vertices); } lp_build_loop_end_cond(&lp_loop, fetch_count, step, LLVMIntUGE); diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c index 667eb507855..c41d7c42a01 100644 --- a/src/gallium/auxiliary/draw/draw_vs_exec.c +++ b/src/gallium/auxiliary/draw/draw_vs_exec.c @@ -99,6 +99,12 @@ vs_exec_run_linear( struct draw_vertex_shader *shader, tgsi_exec_set_constant_buffers(machine, PIPE_MAX_CONSTANT_BUFFERS, constants, const_size); + if (shader->info.uses_instanceid) { + unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_INSTANCEID]; + assert(i < Elements(machine->SystemValue)); + machine->SystemValue[i][0] = shader->draw->instance_id; + } + for (i = 0; i < count; i += MAX_TGSI_VERTICES) { unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i); diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index dee7c0da9b6..d55b9b08070 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -71,6 +71,12 @@ vs_sse_prepare( struct draw_vertex_shader *base, struct tgsi_exec_machine *machine = shader->machine; machine->Samplers = draw->vs.samplers; + + if (base->info.uses_instanceid) { + unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_INSTANCEID]; + assert(i < Elements(machine->SystemValue)); + machine->SystemValue[i][0] = base->draw->instance_id; + } } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h index 40186befb9f..9713d100484 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h @@ -180,6 +180,7 @@ lp_build_tgsi_soa(struct gallivm_state *gallivm, struct lp_type type, struct lp_build_mask_context *mask, LLVMValueRef consts_ptr, + LLVMValueRef system_values_array, const LLVMValueRef *pos, const LLVMValueRef (*inputs)[4], LLVMValueRef (*outputs)[4], @@ -199,4 +200,11 @@ lp_build_tgsi_aos(struct gallivm_state *gallivm, const struct tgsi_shader_info *info); +LLVMValueRef +lp_build_system_values_array(struct gallivm_state *gallivm, + const struct tgsi_shader_info *info, + LLVMValueRef instance_id, + LLVMValueRef facing); + + #endif /* LP_BLD_TGSI_H */ diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c index 1b5a8a5903b..d1585c8e2b7 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c @@ -157,6 +157,8 @@ struct lp_build_tgsi_soa_context */ LLVMValueRef inputs_array; + LLVMValueRef system_values_array; + const struct tgsi_shader_info *info; /** bitmask indicating which register files are accessed indirectly */ unsigned indirect_files; @@ -759,6 +761,23 @@ emit_fetch( } break; + case TGSI_FILE_SYSTEM_VALUE: + assert(!reg->Register.Indirect); + { + LLVMValueRef index; /* index into the system value array */ + LLVMValueRef scalar, scalar_ptr; + + index = lp_build_const_int32(gallivm, + reg->Register.Index * 4 + swizzle); + + scalar_ptr = LLVMBuildGEP(builder, bld->system_values_array, + &index, 1, ""); + scalar = LLVMBuildLoad(builder, scalar_ptr, ""); + + res = lp_build_broadcast_scalar(&bld->base, scalar); + } + break; + default: assert(0 && "invalid src register in emit_fetch()"); return bld->base.undef; @@ -2322,6 +2341,7 @@ lp_build_tgsi_soa(struct gallivm_state *gallivm, struct lp_type type, struct lp_build_mask_context *mask, LLVMValueRef consts_ptr, + LLVMValueRef system_values_array, const LLVMValueRef *pos, const LLVMValueRef (*inputs)[NUM_CHANNELS], LLVMValueRef (*outputs)[NUM_CHANNELS], @@ -2411,6 +2431,8 @@ lp_build_tgsi_soa(struct gallivm_state *gallivm, } } + bld.system_values_array = system_values_array; + tgsi_parse_init( &parse, tokens ); while( !tgsi_parse_end_of_tokens( &parse ) ) { @@ -2512,3 +2534,54 @@ lp_build_tgsi_soa(struct gallivm_state *gallivm, FREE( bld.instructions ); } + +/** + * Build up the system values array out of individual values such as + * the instance ID, front-face, primitive ID, etc. The shader info is + * used to determine which system values are needed and where to put + * them in the system values array. + * + * XXX only instance ID is implemented at this time. + * + * The system values register file is similar to the constants buffer. + * Example declaration: + * DCL SV[0], INSTANCEID + * Example instruction: + * MOVE foo, SV[0].xxxx; + * + * \return LLVM float array (interpreted as float [][4]) + */ +LLVMValueRef +lp_build_system_values_array(struct gallivm_state *gallivm, + const struct tgsi_shader_info *info, + LLVMValueRef instance_id, + LLVMValueRef facing) +{ + LLVMValueRef size = lp_build_const_int32(gallivm, 4 * info->num_system_values); + LLVMTypeRef float_t = LLVMFloatTypeInContext(gallivm->context); + LLVMValueRef array = lp_build_array_alloca(gallivm, float_t, + size, "sysvals_array"); + unsigned i; + + for (i = 0; i < info->num_system_values; i++) { + LLVMValueRef index = lp_build_const_int32(gallivm, i * 4); + LLVMValueRef ptr, value; + + switch (info->system_value_semantic_name[i]) { + case TGSI_SEMANTIC_INSTANCEID: + /* convert instance ID from int to float */ + value = LLVMBuildSIToFP(gallivm->builder, instance_id, float_t, + "sysval_instanceid"); + break; + case TGSI_SEMANTIC_FACE: + /* fall-through */ + default: + assert(0 && "unexpected semantic in build_system_values_array()"); + } + + ptr = LLVMBuildGEP(gallivm->builder, array, &index, 1, ""); + LLVMBuildStore(gallivm->builder, value, ptr); + } + + return array; +} diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c index 7892a67f04c..35b27423513 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.c +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c @@ -1038,7 +1038,6 @@ fetch_src_file_channel(const struct tgsi_exec_machine *mach, break; case TGSI_FILE_INPUT: - case TGSI_FILE_SYSTEM_VALUE: for (i = 0; i < QUAD_SIZE; i++) { /* if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { @@ -1053,6 +1052,15 @@ fetch_src_file_channel(const struct tgsi_exec_machine *mach, } break; + case TGSI_FILE_SYSTEM_VALUE: + /* XXX no swizzling at this point. Will be needed if we put + * gl_FragCoord, for example, in a sys value register. + */ + for (i = 0; i < QUAD_SIZE; i++) { + chan->f[i] = mach->SystemValue[index->i[i]][0]; + } + break; + case TGSI_FILE_TEMPORARY: for (i = 0; i < QUAD_SIZE; i++) { assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); @@ -1907,8 +1915,7 @@ exec_declaration(struct tgsi_exec_machine *mach, const struct tgsi_full_declaration *decl) { if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { - if (decl->Declaration.File == TGSI_FILE_INPUT || - decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) { + if (decl->Declaration.File == TGSI_FILE_INPUT) { uint first, last, mask; first = decl->Range.First; @@ -1921,6 +1928,7 @@ exec_declaration(struct tgsi_exec_machine *mach, * ureg code to emit the right UsageMask value (WRITEMASK_X). * Then, we could remove the tgsi_exec_machine::Face field. */ + /* XXX make FACE a system value */ if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) { uint i; @@ -1962,8 +1970,13 @@ exec_declaration(struct tgsi_exec_machine *mach, } } } + + if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) { + mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First; + } } + typedef void (* micro_op)(union tgsi_exec_channel *dst); static void diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h index b5ebbfbfaab..6c204c73714 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.h +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h @@ -31,6 +31,7 @@ #include "pipe/p_compiler.h" #include "pipe/p_state.h" +#include "pipe/p_shader_tokens.h" #if defined __cplusplus extern "C" { @@ -181,6 +182,8 @@ struct tgsi_sampler /* The maximum total number of vertices */ #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS) +#define TGSI_MAX_MISC_INPUTS 8 + /** function call/activation record */ struct tgsi_call_record { @@ -228,6 +231,10 @@ struct tgsi_exec_machine struct tgsi_exec_vector Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS]; struct tgsi_exec_vector Outputs[TGSI_MAX_TOTAL_VERTICES]; + /* System values */ + unsigned SysSemanticToIndex[TGSI_SEMANTIC_COUNT]; + float SystemValue[TGSI_MAX_MISC_INPUTS][4]; + struct tgsi_exec_vector *Addrs; struct tgsi_exec_vector *Predicates; diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.c b/src/gallium/auxiliary/tgsi/tgsi_ppc.c index 3521847b619..537a0f6c5e4 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_ppc.c +++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.c @@ -294,7 +294,6 @@ emit_fetch(struct gen_context *gen, case TGSI_SWIZZLE_W: switch (reg->Register.File) { case TGSI_FILE_INPUT: - case TGSI_FILE_SYSTEM_VALUE: { int offset = (reg->Register.Index * 4 + swizzle) * 16; int offset_reg = emit_li_offset(gen, offset); @@ -302,6 +301,9 @@ emit_fetch(struct gen_context *gen, ppc_lvx(gen->f, dst_vec, gen->inputs_reg, offset_reg); } break; + case TGSI_FILE_SYSTEM_VALUE: + assert(!"unhandled system value in tgsi_ppc.c"); + break; case TGSI_FILE_TEMPORARY: if (is_ppc_vec_temporary(reg)) { /* use PPC vec register */ diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.c b/src/gallium/auxiliary/tgsi/tgsi_scan.c index 6585da3e838..83c6ac75e54 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_scan.c +++ b/src/gallium/auxiliary/tgsi/tgsi_scan.c @@ -143,7 +143,7 @@ tgsi_scan_shader(const struct tgsi_token *tokens, info->file_count[file]++; info->file_max[file] = MAX2(info->file_max[file], (int)reg); - if (file == TGSI_FILE_INPUT || file == TGSI_FILE_SYSTEM_VALUE) { + if (file == TGSI_FILE_INPUT) { info->input_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name; info->input_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index; info->input_interpolate[reg] = (ubyte)fulldecl->Declaration.Interpolate; @@ -151,6 +151,23 @@ tgsi_scan_shader(const struct tgsi_token *tokens, info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Declaration.CylindricalWrap; info->num_inputs++; } + else if (file == TGSI_FILE_SYSTEM_VALUE) { + unsigned index = fulldecl->Range.First; + unsigned semName = fulldecl->Semantic.Name; + + info->system_value_semantic_name[index] = semName; + info->num_system_values = MAX2(info->num_system_values, + index + 1); + + /* + info->system_value_semantic_name[info->num_system_values++] = + fulldecl->Semantic.Name; + */ + + if (fulldecl->Semantic.Name == TGSI_SEMANTIC_INSTANCEID) { + info->uses_instanceid = TRUE; + } + } else if (file == TGSI_FILE_OUTPUT) { info->output_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name; info->output_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index; diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.h b/src/gallium/auxiliary/tgsi/tgsi_scan.h index 104097fbc03..53ab3d509dd 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_scan.h +++ b/src/gallium/auxiliary/tgsi/tgsi_scan.h @@ -51,6 +51,9 @@ struct tgsi_shader_info ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; /**< TGSI_SEMANTIC_x */ ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; + ubyte num_system_values; + ubyte system_value_semantic_name[PIPE_MAX_SHADER_INPUTS]; + uint file_mask[TGSI_FILE_COUNT]; /**< bitmask of declared registers */ uint file_count[TGSI_FILE_COUNT]; /**< number of declared registers */ int file_max[TGSI_FILE_COUNT]; /**< highest index of declared registers */ @@ -64,6 +67,7 @@ struct tgsi_shader_info boolean writes_stencil; /**< does fragment shader write stencil value? */ boolean writes_edgeflag; /**< vertex shader outputs edgeflag */ boolean uses_kill; /**< KIL or KILP instruction used? */ + boolean uses_instanceid; /** * Bitmask indicating which register files are accessed with diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c index 086d983a73a..3f2cda860e0 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c +++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c @@ -163,6 +163,14 @@ get_immediate_base( void ) reg_DX ); } +static struct x86_reg +get_system_value_base( void ) +{ + return x86_make_disp( + get_machine_base(), + Offset(struct tgsi_exec_machine, SystemValue) ); +} + /** * Data access helpers. @@ -229,6 +237,16 @@ get_temp( } static struct x86_reg +get_system_value( + unsigned vec, + unsigned chan ) +{ + return x86_make_disp( + get_system_value_base(), /* base */ + (vec * 4 + chan) * 4 ); /* byte offset from base */ +} + +static struct x86_reg get_coef( unsigned vec, unsigned chan, @@ -423,6 +441,30 @@ emit_tempf( } /** + * Copy a system value to xmm register + * \param xmm the destination xmm register + * \param vec the source system value register + * \param chan src channel to fetch (X, Y, Z or W) + */ +static void +emit_system_value( + struct x86_function *func, + unsigned xmm, + unsigned vec, + unsigned chan ) +{ + sse_movss( + func, + make_xmm( xmm ), + get_system_value( vec, chan ) ); + sse_shufps( + func, + make_xmm( xmm ), + make_xmm( xmm ), + SHUF( 0, 0, 0, 0 ) ); +} + +/** * Load an xmm register with an input attrib coefficient (a0, dadx or dady) * \param xmm the destination xmm register * \param vec the src input/attribute coefficient index @@ -1281,8 +1323,15 @@ emit_fetch( swizzle ); break; - case TGSI_FILE_INPUT: case TGSI_FILE_SYSTEM_VALUE: + emit_system_value( + func, + xmm, + reg->Register.Index, + swizzle ); + break; + + case TGSI_FILE_INPUT: emit_inputf( func, xmm, @@ -2636,8 +2685,7 @@ emit_declaration( struct x86_function *func, struct tgsi_full_declaration *decl ) { - if( decl->Declaration.File == TGSI_FILE_INPUT || - decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE ) { + if( decl->Declaration.File == TGSI_FILE_INPUT ) { unsigned first, last, mask; unsigned i, j; |