#include "tgsi/util/tgsi_dump.h"
#include "draw_context.h"
-#include "draw_private.h"
+#include "draw_vs.h"
/*
#include "pipe/p_shader_tokens.h"
#include "draw_context.h"
-#include "draw_private.h"
+#include "draw_vs.h"
#ifndef IS_NEGATIVE
#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 )
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;
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
#include "pipe/p_util.h"
#include "pipe/p_shader_tokens.h"
-#include "draw_private.h"
+#include "draw_vs.h"
/** subclass of draw_stage */
struct draw_pt_middle_end;
struct draw_pt_front_end;
+struct draw_vertex_shader;
#define MAX_SHADER_VERTICES 128
#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.
#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.
*
#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"
#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 {
#include "pipe/p_util.h"
#include "pipe/p_shader_tokens.h"
-#include "draw_private.h"
+#include "draw_vs.h"
#include "draw_context.h"
vs = draw_create_vs_exec( draw, shader );
}
}
- assert(vs);
-
- tgsi_scan_shader(shader->tokens, &vs->info);
+ assert(vs);
return vs;
}
{
draw->vertex_shader = dvs;
draw->num_vs_outputs = dvs->info.num_outputs;
-
- tgsi_exec_machine_init(&draw->machine);
-
dvs->prepare( dvs, draw );
}
else {
#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);
return mask;
}
+#define MAX_TGSI_VERTICES 4
+
#endif
#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 )
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 */
+/* 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 )
{
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;
}
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;
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 )
{
/* 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;
{
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];
};
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 */
*/
shader->func(machine->Inputs,
machine->Outputs,
- machine->Consts,
+ (float (*)[4])machine->Consts,
machine->Temps,
shader->immediates);
}
}
+/* 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 )
/* 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 );
#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 {
projects / profile / ivi / mesa.git / commitdiff