*/
#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#if defined(__i386__) || defined(__386__)
+#include "pipe/tgsi/exec/tgsi_sse2.h"
+#endif
#include "draw_private.h"
#include "draw_context.h"
-#include "draw_vertex.h"
-#include "pipe/tgsi/exec/tgsi_core.h"
+#include "x86/rtasm/x86sse.h"
+#include "pipe/llvm/gallivm.h"
+
+
+#define DBG_VS 0
+
static INLINE unsigned
-compute_clipmask(float cx, float cy, float cz, float cw)
+compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
{
unsigned mask = 0;
-
- if (-cx + cw < 0) mask |= CLIP_RIGHT_BIT;
- if ( cx + cw < 0) mask |= CLIP_LEFT_BIT;
- if (-cy + cw < 0) mask |= CLIP_TOP_BIT;
- if ( cy + cw < 0) mask |= CLIP_BOTTOM_BIT;
- if (-cz + cw < 0) mask |= CLIP_FAR_BIT;
- if ( cz + cw < 0) mask |= CLIP_NEAR_BIT;
+ 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);
+ }
return mask;
}
-
-
-#if !defined(XSTDCALL)
-#if defined(WIN32)
-#define XSTDCALL __stdcall
-#else
-#define XSTDCALL
-#endif
-#endif
-
-#if defined(USE_X86_ASM) || defined(SLANG_X86)
-typedef void (XSTDCALL *sse2_function)(
+typedef void (XSTDCALL *codegen_function) (
const struct tgsi_exec_vector *input,
struct tgsi_exec_vector *output,
float (*constant)[4],
struct tgsi_exec_vector *temporary );
-#endif
+
/**
* Transform vertices with the current vertex program/shader
unsigned elts[4], unsigned count,
struct vertex_header *vOut[])
{
- struct tgsi_exec_machine machine;
+ struct tgsi_exec_machine *machine = &draw->machine;
unsigned int j;
ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
const float *trans = draw->viewport.translate;
assert(count <= 4);
- assert(draw->vertex_shader.outputs_written & (1 << TGSI_ATTRIB_POS));
-
-#ifdef DEBUG
- memset( &machine, 0, sizeof( machine ) );
-#endif
-
- /* init machine state */
- tgsi_exec_machine_init(&machine,
- draw->vertex_shader.tokens,
- PIPE_MAX_SAMPLERS,
- NULL /*samplers*/ );
+ assert(draw->vertex_shader->state->output_semantic_name[0]
+ == TGSI_SEMANTIC_POSITION);
/* Consts does not require 16 byte alignment. */
- machine.Consts = (float (*)[4]) draw->mapped_constants;
-
- machine.Inputs = ALIGN16_ASSIGN(inputs);
- machine.Outputs = ALIGN16_ASSIGN(outputs);
+ machine->Consts = (float (*)[4]) draw->user.constants;
- draw_vertex_fetch( draw, &machine, elts, count );
+ machine->Inputs = ALIGN16_ASSIGN(inputs);
+ machine->Outputs = ALIGN16_ASSIGN(outputs);
+ draw_vertex_fetch( draw, machine, elts, count );
/* run shader */
- if( draw->vertex_shader.executable != NULL ) {
-#if defined(USE_X86_ASM) || defined(SLANG_X86)
- sse2_function func = (sse2_function) draw->vertex_shader.executable;
+#if defined(__i386__) || defined(__386__)
+ if (draw->use_sse) {
+ /* SSE */
+ /* cast away const */
+ struct draw_vertex_shader *shader
+ = (struct draw_vertex_shader *)draw->vertex_shader;
+ codegen_function func
+ = (codegen_function) x86_get_func( &shader->sse2_program );
func(
- machine.Inputs,
- machine.Outputs,
- machine.Consts,
- machine.Temps );
-#else
- assert( 0 );
-#endif
+ machine->Inputs,
+ machine->Outputs,
+ machine->Consts,
+ machine->Temps );
}
- else {
- tgsi_exec_machine_run( &machine );
+ else
+#endif
+ {
+ /* 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: */
- 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];
+ /* 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.
+ */
+ 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];
- vOut[j]->clipmask = compute_clipmask(x, y, z, w) | draw->user_clipmask;
+ vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
vOut[j]->edgeflag = 1;
/* divide by w */
vOut[j]->data[0][2] = z * scale[2] + trans[2];
vOut[j]->data[0][3] = w;
- /* remaining attributes are packed into sequential post-transform
+#if DBG_VS
+ printf("output[%d]win: %f %f %f %f\n", j,
+ vOut[j]->data[0][0],
+ vOut[j]->data[0][1],
+ vOut[j]->data[0][2],
+ vOut[j]->data[0][3]);
+#endif
+ /* Remaining attributes are packed into sequential post-transform
* vertex attrib slots.
*/
- for (slot = 1; slot < draw->vertex_info.num_attribs; 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];
+ 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];
+#if DBG_VS
+ printf("output[%d][%d]: %f %f %f %f\n", j, slot,
+ vOut[j]->data[slot][0],
+ vOut[j]->data[slot][1],
+ vOut[j]->data[slot][2],
+ vOut[j]->data[slot][3]);
+#endif
}
} /* loop over vertices */
}
/**
+ * Run the vertex shader on all vertices in the vertex queue.
* Called by the draw module when the vertx cache needs to be flushed.
- * This involves running the vertex shader.
*/
-void draw_vertex_shader_queue_flush( struct draw_context *draw )
+void
+draw_vertex_shader_queue_flush(struct draw_context *draw)
{
unsigned i, j;
+// fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
+#ifdef MESA_LLVM
+ if (draw->vertex_shader->llvm_prog) {
+ draw_vertex_shader_queue_flush_llvm(draw);
+ return;
+ }
+#endif
+
/* run vertex shader on vertex cache entries, four per invokation */
for (i = 0; i < draw->vs.queue_nr; i += 4) {
struct vertex_header *dests[4];
draw->vs.queue_nr = 0;
}
+
+struct draw_vertex_shader *
+draw_create_vertex_shader(struct draw_context *draw,
+ const struct pipe_shader_state *shader)
+{
+ struct draw_vertex_shader *vs;
+
+ vs = CALLOC_STRUCT( draw_vertex_shader );
+ if (vs == NULL) {
+ return NULL;
+ }
+
+ vs->state = shader;
+
+#ifdef MESA_LLVM
+ vs->llvm_prog = gallivm_from_tgsi(shader->tokens, GALLIVM_VS);
+ draw->engine = gallivm_global_cpu_engine();
+ if (!draw->engine) {
+ draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
+ }
+ else {
+ gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
+ }
+#elif defined(__i386__) || defined(__386__)
+ if (draw->use_sse) {
+ /* cast-away const */
+ struct pipe_shader_state *sh = (struct pipe_shader_state *) shader;
+
+ x86_init_func( &vs->sse2_program );
+ tgsi_emit_sse2( (struct tgsi_token *) sh->tokens, &vs->sse2_program );
+ }
+#endif
+
+ return vs;
+}
+
+
+void
+draw_bind_vertex_shader(struct draw_context *draw,
+ struct draw_vertex_shader *dvs)
+{
+ draw_flush(draw);
+ draw->vertex_shader = dvs;
+
+ draw->num_vs_outputs = dvs->state->num_outputs;
+
+ /* specify the fragment program to interpret/execute */
+ tgsi_exec_machine_init(&draw->machine,
+ draw->vertex_shader->state->tokens,
+ PIPE_MAX_SAMPLERS,
+ NULL /*samplers*/ );
+}
+
+
+void
+draw_delete_vertex_shader(struct draw_context *draw,
+ struct draw_vertex_shader *dvs)
+{
+#if defined(__i386__) || defined(__386__)
+ x86_release_func( (struct x86_function *) &dvs->sse2_program );
+#endif
+
+ FREE( dvs );
+}