diff options
author | Keith Whitwell <[email protected]> | 2008-04-17 14:20:00 +0100 |
---|---|---|
committer | Keith Whitwell <[email protected]> | 2008-04-17 14:23:13 +0100 |
commit | 280bcff1fa200b790d8712946a4ffbaa47a67433 (patch) | |
tree | e9d8ebf9e24e8bee0687a3ca283dcff10787008f /src/gallium | |
parent | 938ec19210c5b4e19dcb2b606c9ade415f2c1f84 (diff) |
draw: add vertex shader run_linear function
Diffstat (limited to 'src/gallium')
-rw-r--r-- | src/gallium/auxiliary/draw/draw_aapoint.c | 2 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_clip.c | 2 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_context.c | 17 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_flatshade.c | 2 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_private.h | 28 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c | 2 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c | 2 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_twoside.c | 2 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vertex_fetch.c | 2 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vertex_shader.c | 7 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vs.h | 40 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vs_exec.c | 95 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vs_llvm.c | 66 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_vs_sse.c | 76 | ||||
-rw-r--r-- | src/gallium/auxiliary/draw/draw_wide_point.c | 2 |
15 files changed, 286 insertions, 59 deletions
diff --git a/src/gallium/auxiliary/draw/draw_aapoint.c b/src/gallium/auxiliary/draw/draw_aapoint.c index fcebe3e7a0c..e84d380e500 100644 --- a/src/gallium/auxiliary/draw/draw_aapoint.c +++ b/src/gallium/auxiliary/draw/draw_aapoint.c @@ -48,7 +48,7 @@ #include "tgsi/util/tgsi_dump.h" #include "draw_context.h" -#include "draw_private.h" +#include "draw_vs.h" /* diff --git a/src/gallium/auxiliary/draw/draw_clip.c b/src/gallium/auxiliary/draw/draw_clip.c index e24c5d80322..0ac3a240e5e 100644 --- a/src/gallium/auxiliary/draw/draw_clip.c +++ b/src/gallium/auxiliary/draw/draw_clip.c @@ -36,7 +36,7 @@ #include "pipe/p_shader_tokens.h" #include "draw_context.h" -#include "draw_private.h" +#include "draw_vs.h" #ifndef IS_NEGATIVE diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c index cdcbfdc3236..9b2dcc0b572 100644 --- a/src/gallium/auxiliary/draw/draw_context.c +++ b/src/gallium/auxiliary/draw/draw_context.c @@ -33,8 +33,8 @@ #include "pipe/p_util.h" #include "draw_context.h" -#include "draw_private.h" #include "draw_vbuf.h" +#include "draw_vs.h" struct draw_context *draw_create( void ) @@ -108,6 +108,8 @@ struct draw_context *draw_create( void ) draw_vertex_cache_invalidate( draw ); draw_set_mapped_element_buffer( draw, 0, NULL ); + tgsi_exec_machine_init(&draw->machine); + if (!draw_pt_init( draw )) goto fail; @@ -460,3 +462,16 @@ boolean draw_get_edgeflag( struct draw_context *draw, else return 1; } + + +#if 0 +/* Crufty init function. Fix me. + */ +boolean draw_init_machine( struct draw_context *draw ) +{ + ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_MAX_ATTRIBS); + ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_MAX_ATTRIBS); + machine->Inputs = ALIGN16_ASSIGN(inputs); + machine->Outputs = ALIGN16_ASSIGN(outputs); +} +#endif diff --git a/src/gallium/auxiliary/draw/draw_flatshade.c b/src/gallium/auxiliary/draw/draw_flatshade.c index af2cb05c989..54baa1fbc98 100644 --- a/src/gallium/auxiliary/draw/draw_flatshade.c +++ b/src/gallium/auxiliary/draw/draw_flatshade.c @@ -30,7 +30,7 @@ #include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" -#include "draw_private.h" +#include "draw_vs.h" /** subclass of draw_stage */ diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 359a8b8a3db..7953dbb7d9e 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -55,6 +55,7 @@ struct gallivm_cpu_engine; struct draw_pt_middle_end; struct draw_pt_front_end; +struct draw_vertex_shader; #define MAX_SHADER_VERTICES 128 @@ -132,33 +133,6 @@ struct draw_stage #define VCACHE_OVERFLOW 4 #define VS_QUEUE_LENGTH (VCACHE_SIZE + VCACHE_OVERFLOW + 1) /* can never fill up */ -/** - * Private version of the compiled vertex_shader - */ -struct draw_vertex_shader { - - /* This member will disappear shortly: - */ - struct pipe_shader_state state; - - struct tgsi_shader_info info; - - void (*prepare)( struct draw_vertex_shader *shader, - struct draw_context *draw ); - - /* Run the shader - this interface will get cleaned up in the - * future: - */ - boolean (*run)( struct draw_vertex_shader *shader, - struct draw_context *draw, - const unsigned *elts, - unsigned count, - void *out, - unsigned vertex_size); - - - void (*delete)( struct draw_vertex_shader * ); -}; /* Internal function for vertex fetch. diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c index db6d6c76ea7..79548d41563 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c @@ -32,9 +32,9 @@ #include "pipe/p_util.h" #include "draw/draw_context.h" -#include "draw/draw_private.h" #include "draw/draw_vertex.h" #include "draw/draw_pt.h" +#include "draw/draw_vs.h" /* The simplest 'middle end' in the new vertex code. * diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c index a66e8b0ba22..cf37493eb62 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c @@ -27,10 +27,10 @@ #include "pipe/p_util.h" #include "draw/draw_context.h" -#include "draw/draw_private.h" #include "draw/draw_vbuf.h" #include "draw/draw_vertex.h" #include "draw/draw_pt.h" +#include "draw/draw_vs.h" #include "translate/translate.h" diff --git a/src/gallium/auxiliary/draw/draw_twoside.c b/src/gallium/auxiliary/draw/draw_twoside.c index 3debaac2822..01d905c1534 100644 --- a/src/gallium/auxiliary/draw/draw_twoside.c +++ b/src/gallium/auxiliary/draw/draw_twoside.c @@ -31,7 +31,7 @@ #include "pipe/p_util.h" #include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" -#include "draw_private.h" +#include "draw_vs.h" struct twoside_stage { diff --git a/src/gallium/auxiliary/draw/draw_vertex_fetch.c b/src/gallium/auxiliary/draw/draw_vertex_fetch.c index 9041041006e..0bc2fcb4245 100644 --- a/src/gallium/auxiliary/draw/draw_vertex_fetch.c +++ b/src/gallium/auxiliary/draw/draw_vertex_fetch.c @@ -32,7 +32,7 @@ #include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" -#include "draw_private.h" +#include "draw_vs.h" #include "draw_context.h" diff --git a/src/gallium/auxiliary/draw/draw_vertex_shader.c b/src/gallium/auxiliary/draw/draw_vertex_shader.c index 8572a6d40c3..c0c17c116ea 100644 --- a/src/gallium/auxiliary/draw/draw_vertex_shader.c +++ b/src/gallium/auxiliary/draw/draw_vertex_shader.c @@ -95,10 +95,8 @@ draw_create_vertex_shader(struct draw_context *draw, vs = draw_create_vs_exec( draw, shader ); } } - assert(vs); - - tgsi_scan_shader(shader->tokens, &vs->info); + assert(vs); return vs; } @@ -113,9 +111,6 @@ draw_bind_vertex_shader(struct draw_context *draw, { draw->vertex_shader = dvs; draw->num_vs_outputs = dvs->info.num_outputs; - - tgsi_exec_machine_init(&draw->machine); - dvs->prepare( dvs, draw ); } else { diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 33ce1e335e0..e88d00e2479 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -35,10 +35,46 @@ #include "draw_private.h" -struct draw_vertex_shader; struct draw_context; struct pipe_shader_state; +/** + * Private version of the compiled vertex_shader + */ +struct draw_vertex_shader { + + /* This member will disappear shortly: + */ + struct pipe_shader_state state; + + struct tgsi_shader_info info; + + void (*prepare)( struct draw_vertex_shader *shader, + struct draw_context *draw ); + + /* Run the shader - this interface will get cleaned up in the + * future: + */ + boolean (*run)( struct draw_vertex_shader *shader, + struct draw_context *draw, + const unsigned *elts, + unsigned count, + void *out, + unsigned vertex_size); + + void (*run_linear)( struct draw_vertex_shader *shader, + const float (*input)[4], + float (*output)[4], + const float (*constants)[4], + unsigned count, + unsigned input_stride, + unsigned output_stride ); + + + void (*delete)( struct draw_vertex_shader * ); +}; + + struct draw_vertex_shader * draw_create_vs_exec(struct draw_context *draw, const struct pipe_shader_state *templ); @@ -79,5 +115,7 @@ compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr) return mask; } +#define MAX_TGSI_VERTICES 4 + #endif diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c index 5c88c2e24e3..0e05b797159 100644 --- a/src/gallium/auxiliary/draw/draw_vs_exec.c +++ b/src/gallium/auxiliary/draw/draw_vs_exec.c @@ -40,8 +40,20 @@ #include "tgsi/util/tgsi_parse.h" -#define MAX_TGSI_VERTICES 4 +struct exec_vertex_shader { + struct draw_vertex_shader base; + struct tgsi_exec_machine *machine; +}; + +static struct exec_vertex_shader *exec_vertex_shader( struct draw_vertex_shader *vs ) +{ + return (struct exec_vertex_shader *)vs; +} + + +/* Not required for run_linear. + */ static void vs_exec_prepare( struct draw_vertex_shader *shader, struct draw_context *draw ) @@ -81,10 +93,9 @@ vs_exec_run( struct draw_vertex_shader *shader, const float *scale = draw->viewport.scale; const float *trans = draw->viewport.translate; - assert(draw->vertex_shader->info.output_semantic_name[0] - == TGSI_SEMANTIC_POSITION); + assert(shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION); - machine->Consts = (float (*)[4]) draw->user.constants; + machine->Consts = (const float (*)[4]) draw->user.constants; machine->Inputs = ALIGN16_ASSIGN(inputs); if (draw->rasterizer->bypass_vs) { /* outputs are just the inputs */ @@ -179,6 +190,64 @@ vs_exec_run( struct draw_vertex_shader *shader, +/* Simplified vertex shader interface for the pt paths. Given the + * complexity of code-generating all the above operations together, + * it's time to try doing all the other stuff separately. + */ +static void +vs_exec_run_linear( struct draw_vertex_shader *shader, + const float (*input)[4], + float (*output)[4], + const float (*constants)[4], + unsigned count, + unsigned input_stride, + unsigned output_stride ) +{ + struct exec_vertex_shader *evs = exec_vertex_shader(shader); + struct tgsi_exec_machine *machine = evs->machine; + unsigned int i, j; + unsigned slot; + + machine->Consts = constants; + + for (i = 0; i < count; i += MAX_TGSI_VERTICES) { + unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i); + + /* Swizzle inputs. + */ + for (j = 0; j < max_vertices; j++) { + for (slot = 0; slot < shader->info.num_inputs; slot++) { + machine->Inputs[slot].xyzw[0].f[j] = input[slot][0]; + machine->Inputs[slot].xyzw[1].f[j] = input[slot][1]; + machine->Inputs[slot].xyzw[2].f[j] = input[slot][2]; + machine->Inputs[slot].xyzw[3].f[j] = input[slot][3]; + } + } + + /* run interpreter */ + tgsi_exec_machine_run( machine ); + + /* Unswizzle all output results. + */ + for (j = 0; j < max_vertices; j++) { + for (slot = 0; slot < shader->info.num_outputs; slot++) { + output[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; + output[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; + output[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; + output[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; + } + } + + /* Advance input, output pointers: + */ + input = (const float (*)[4])((const char *)input + input_stride); + output = (float (*)[4])((char *)output + output_stride); + } +} + + + + static void vs_exec_delete( struct draw_vertex_shader *dvs ) { @@ -191,17 +260,23 @@ struct draw_vertex_shader * draw_create_vs_exec(struct draw_context *draw, const struct pipe_shader_state *state) { - struct draw_vertex_shader *vs = CALLOC_STRUCT( draw_vertex_shader ); + struct exec_vertex_shader *vs = CALLOC_STRUCT( exec_vertex_shader ); uint nt = tgsi_num_tokens(state->tokens); if (vs == NULL) return NULL; /* we make a private copy of the tokens */ - vs->state.tokens = mem_dup(state->tokens, nt * sizeof(state->tokens[0])); - vs->prepare = vs_exec_prepare; - vs->run = vs_exec_run; - vs->delete = vs_exec_delete; + vs->base.state.tokens = mem_dup(state->tokens, nt * sizeof(state->tokens[0])); + tgsi_scan_shader(state->tokens, &vs->base.info); + + + vs->base.prepare = vs_exec_prepare; + vs->base.run = vs_exec_run; + vs->base.run_linear = vs_exec_run_linear; + vs->base.delete = vs_exec_delete; + vs->machine = &draw->machine; + - return vs; + return &vs->base; } diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c index 73076d24674..d0baca715e6 100644 --- a/src/gallium/auxiliary/draw/draw_vs_llvm.c +++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c @@ -88,8 +88,7 @@ vs_llvm_run( struct draw_vertex_shader *base, assert(count <= 4); - assert(draw->vertex_shader->state->output_semantic_name[0] - == TGSI_SEMANTIC_POSITION); + assert(base->state->output_semantic_name[0] == TGSI_SEMANTIC_POSITION); /* Consts does not require 16 byte alignment. */ machine->Consts = (float (*)[4]) draw->user.constants; @@ -169,6 +168,65 @@ vs_llvm_run( struct draw_vertex_shader *base, return clipped != 0; } + + + +static void +vs_llvm_run_linear( struct draw_vertex_shader *base, + const float (*input)[4], + float (*output)[4], + const float (*constants)[4], + unsigned count, + unsigned input_stride, + unsigned output_stride ) +{ + struct draw_llvm_vertex_shader *shader = + (struct draw_llvm_vertex_shader *)base; + + struct tgsi_exec_machine *machine = &draw->machine; + unsigned int j; + + + for (i = 0; i < count; i += MAX_TGSI_VERTICES) { + unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i); + + /* Swizzle inputs. + */ + for (j = 0; j < max_vertices; j++) { + for (slot = 0; slot < draw->num_vs_inputs; slot++) { + machine->Inputs[slot].xyzw[0].f[j] = input[slot][0]; + machine->Inputs[slot].xyzw[1].f[j] = input[slot][1]; + machine->Inputs[slot].xyzw[2].f[j] = input[slot][2]; + machine->Inputs[slot].xyzw[3].f[j] = input[slot][3]; + } + } + + /* run shader */ + gallivm_cpu_vs_exec(shader->llvm_prog, + machine->Inputs, + machine->Outputs, + (float (*)[4]) constants, + machine->Temps); + + + /* Unswizzle all output results + */ + for (slot = 1; slot < draw->num_vs_outputs; slot++) { + output[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; + output[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; + output[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; + output[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; + } + + /* Advance input, output pointers: + */ + input = (const float (*)[4])((const char *)input + input_stride); + output = (float (*)[4])((char *)output + output_stride); + } +} + + + static void vs_llvm_delete( struct draw_vertex_shader *base ) { @@ -198,8 +256,12 @@ draw_create_vs_llvm(struct draw_context *draw, /* we make a private copy of the tokens */ vs->base.state.tokens = mem_dup(templ->tokens, nt * sizeof(templ->tokens[0])); + + tgsi_scan_shader(shader->tokens, &vs->base.info); + vs->base.prepare = vs_llvm_prepare; vs->base.run = vs_llvm_run; + vs->base.run_linear = vs_llvm_run_linear; vs->base.delete = vs_llvm_delete; { diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index ee0a3105b99..873ecfdc5d3 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -58,7 +58,11 @@ typedef void (XSTDCALL *codegen_function) ( struct draw_sse_vertex_shader { struct draw_vertex_shader base; struct x86_function sse2_program; + codegen_function func; + + struct tgsi_exec_machine *machine; + float immediates[TGSI_EXEC_NUM_IMMEDIATES][4]; }; @@ -96,11 +100,10 @@ vs_sse_run( struct draw_vertex_shader *base, const float *scale = draw->viewport.scale; const float *trans = draw->viewport.translate; - assert(draw->vertex_shader->info.output_semantic_name[0] - == TGSI_SEMANTIC_POSITION); + assert(base->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION); /* Consts does not require 16 byte alignment. */ - machine->Consts = (float (*)[4]) draw->user.constants; + machine->Consts = (const float (*)[4]) draw->user.constants; machine->Inputs = ALIGN16_ASSIGN(inputs); if (draw->rasterizer->bypass_vs) { /* outputs are just the inputs */ @@ -124,7 +127,7 @@ vs_sse_run( struct draw_vertex_shader *base, */ shader->func(machine->Inputs, machine->Outputs, - machine->Consts, + (float (*)[4])machine->Consts, machine->Temps, shader->immediates); } @@ -200,6 +203,67 @@ vs_sse_run( struct draw_vertex_shader *base, } +/* Simplified vertex shader interface for the pt paths. Given the + * complexity of code-generating all the above operations together, + * it's time to try doing all the other stuff separately. + */ +static void +vs_sse_run_linear( struct draw_vertex_shader *base, + const float (*input)[4], + float (*output)[4], + const float (*constants)[4], + unsigned count, + unsigned input_stride, + unsigned output_stride ) +{ + struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base; + struct tgsi_exec_machine *machine = shader->machine; + unsigned int i, j; + unsigned slot; + + for (i = 0; i < count; i += MAX_TGSI_VERTICES) { + unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i); + + /* Swizzle inputs. + */ + for (j = 0; j < max_vertices; j++) { + for (slot = 0; slot < base->info.num_inputs; slot++) { + machine->Inputs[slot].xyzw[0].f[j] = input[slot][0]; + machine->Inputs[slot].xyzw[1].f[j] = input[slot][1]; + machine->Inputs[slot].xyzw[2].f[j] = input[slot][2]; + machine->Inputs[slot].xyzw[3].f[j] = input[slot][3]; + } + } + + /* run compiled shader + */ + shader->func(machine->Inputs, + machine->Outputs, + (float (*)[4])constants, + machine->Temps, + shader->immediates); + + + /* Unswizzle all output results. + */ + for (j = 0; j < max_vertices; j++) { + for (slot = 0; slot < base->info.num_outputs; slot++) { + output[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; + output[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; + output[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; + output[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; + } + } + + /* Advance input, output pointers: + */ + input = (const float (*)[4])((const char *)input + input_stride); + output = (float (*)[4])((char *)output + output_stride); + } +} + + + static void vs_sse_delete( struct draw_vertex_shader *base ) @@ -229,8 +293,12 @@ draw_create_vs_sse(struct draw_context *draw, /* we make a private copy of the tokens */ vs->base.state.tokens = mem_dup(templ->tokens, nt * sizeof(templ->tokens[0])); + + tgsi_scan_shader(templ->tokens, &vs->base.info); + vs->base.prepare = vs_sse_prepare; vs->base.run = vs_sse_run; + vs->base.run_linear = vs_sse_run_linear; vs->base.delete = vs_sse_delete; x86_init_func( &vs->sse2_program ); diff --git a/src/gallium/auxiliary/draw/draw_wide_point.c b/src/gallium/auxiliary/draw/draw_wide_point.c index 6fc7c9fcd7a..0fa7c9313c1 100644 --- a/src/gallium/auxiliary/draw/draw_wide_point.c +++ b/src/gallium/auxiliary/draw/draw_wide_point.c @@ -31,7 +31,7 @@ #include "pipe/p_util.h" #include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" -#include "draw_private.h" +#include "draw_vs.h" struct widepoint_stage { |