* Brian Paul
*/
-#include "pipe/p_util.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
#include "pipe/p_shader_tokens.h"
#include "draw_private.h"
#include "draw_context.h"
#include "draw_vs.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_scan.h"
-static INLINE unsigned
-compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
-{
- unsigned mask = 0;
- unsigned i;
- /* Do the hardwired planes first:
- */
- if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
- if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
- if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
- if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
- if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
- if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
-
- /* Followed by any remaining ones:
- */
- for (i = 6; i < nr; i++) {
- if (dot4(clip, plane[i]) < 0)
- mask |= (1<<i);
- }
+struct exec_vertex_shader {
+ struct draw_vertex_shader base;
+ struct tgsi_exec_machine *machine;
+ const struct tgsi_token *machine_tokens;
+};
- return mask;
+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 )
{
- /* specify the vertex program to interpret/execute */
- tgsi_exec_machine_bind_shader(&draw->machine,
- shader->state->tokens,
- PIPE_MAX_SAMPLERS,
- NULL /*samplers*/ );
+ struct exec_vertex_shader *evs = exec_vertex_shader(shader);
- draw_update_vertex_fetch( draw );
+ /* Specify the vertex program to interpret/execute.
+ * Avoid rebinding when possible.
+ */
+ if (evs->machine_tokens != shader->state.tokens) {
+ tgsi_exec_machine_bind_shader(evs->machine,
+ shader->state.tokens,
+ PIPE_MAX_SAMPLERS,
+ NULL /*samplers*/ );
+ evs->machine_tokens = shader->state.tokens;
+ }
}
-/**
- * Transform vertices with the current vertex program/shader
- * Up to four vertices can be shaded at a time.
- * \param vbuffer the input vertex data
- * \param elts indexes of four input vertices
- * \param count number of vertices to shade [1..4]
- * \param vOut array of pointers to four output vertices
+
+
+/* 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( struct draw_vertex_shader *shader,
- struct draw_context *draw,
- const unsigned *elts,
- unsigned count,
- struct vertex_header *vOut[] )
+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 tgsi_exec_machine *machine = &draw->machine;
- unsigned int j;
-
- ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
- ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
- const float *scale = draw->viewport.scale;
- const float *trans = draw->viewport.translate;
+ struct exec_vertex_shader *evs = exec_vertex_shader(shader);
+ struct tgsi_exec_machine *machine = evs->machine;
+ unsigned int i, j;
+ unsigned slot;
- assert(count <= 4);
- assert(draw->vertex_shader->info.output_semantic_name[0]
- == TGSI_SEMANTIC_POSITION);
+ machine->Consts = constants;
- machine->Consts = (float (*)[4]) draw->user.constants;
- machine->Inputs = ALIGN16_ASSIGN(inputs);
- machine->Outputs = ALIGN16_ASSIGN(outputs);
+ for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
+ unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
- draw->vertex_fetch.fetch_func( draw, machine, elts, count );
-
- /* run interpreter */
- tgsi_exec_machine_run( machine );
-
-
- /* store machine results */
- for (j = 0; j < count; j++) {
- unsigned slot;
- float x, y, z, w;
-
- /* Handle attr[0] (position) specially:
- *
- * XXX: Computing the clipmask should be done in the vertex
- * program as a set of DP4 instructions appended to the
- * user-provided code.
+ /* Swizzle inputs.
*/
- x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
- y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
- z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
- w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
-
- if (!draw->rasterizer->bypass_clipping) {
- vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
- vOut[j]->edgeflag = 1;
-
- /* divide by w */
- w = 1.0f / w;
- x *= w;
- y *= w;
- z *= w;
-
- /* Viewport mapping */
- vOut[j]->data[0][0] = x * scale[0] + trans[0];
- vOut[j]->data[0][1] = y * scale[1] + trans[1];
- vOut[j]->data[0][2] = z * scale[2] + trans[2];
- vOut[j]->data[0][3] = w;
- }
- else {
- vOut[j]->clipmask = 0;
- vOut[j]->edgeflag = 1;
- vOut[j]->data[0][0] = x;
- vOut[j]->data[0][1] = y;
- vOut[j]->data[0][2] = z;
- vOut[j]->data[0][3] = w;
- }
-
- /* Remaining attributes are packed into sequential post-transform
- * vertex attrib slots.
+ for (j = 0; j < max_vertices; j++) {
+#if 0
+ debug_printf("%d) Input vert:\n", i + j);
+ for (slot = 0; slot < shader->info.num_inputs; slot++) {
+ debug_printf("\t%d: %f %f %f %f\n", slot,
+ input[slot][0],
+ input[slot][1],
+ input[slot][2],
+ input[slot][3]);
+ }
+#endif
+
+ 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];
+ }
+
+ input = (const float (*)[4])((const char *)input + input_stride);
+ }
+
+ /* run interpreter */
+ tgsi_exec_machine_run( machine );
+
+ /* Unswizzle all output results.
*/
- for (slot = 1; slot < draw->num_vs_outputs; slot++) {
- vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
- vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
- vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
- vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
- }
- } /* loop over vertices */
+ 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];
+
+ }
+
+#if 0
+ debug_printf("%d) Post xform vert:\n", i + j);
+ for (slot = 0; slot < shader->info.num_outputs; slot++) {
+ debug_printf("\t%d: %f %f %f %f\n", slot,
+ output[slot][0],
+ output[slot][1],
+ output[slot][2],
+ output[slot][3]);
+ }
+#endif
+
+ output = (float (*)[4])((char *)output + output_stride);
+ }
+
+ }
}
+
static void
vs_exec_delete( struct draw_vertex_shader *dvs )
{
+ FREE((void*) dvs->state.tokens);
FREE( 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 );
if (vs == NULL)
return NULL;
- vs->state = state;
- vs->prepare = vs_exec_prepare;
- vs->run = vs_exec_run;
- vs->delete = vs_exec_delete;
+ /* we make a private copy of the tokens */
+ vs->base.state.tokens = tgsi_dup_tokens(state->tokens);
+ if (!vs->base.state.tokens) {
+ FREE(vs);
+ return NULL;
+ }
+
+ tgsi_scan_shader(state->tokens, &vs->base.info);
+
+ vs->base.draw = draw;
+ vs->base.prepare = vs_exec_prepare;
+ vs->base.run_linear = vs_exec_run_linear;
+ vs->base.delete = vs_exec_delete;
+ vs->base.create_varient = draw_vs_varient_generic;
+ vs->machine = &draw->vs.machine;
- return vs;
+ return &vs->base;
}