summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/draw/draw_vs_llvm.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/auxiliary/draw/draw_vs_llvm.c')
-rw-r--r--src/gallium/auxiliary/draw/draw_vs_llvm.c138
1 files changed, 45 insertions, 93 deletions
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index 73076d2467..dcada66514 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -47,6 +47,7 @@
struct draw_llvm_vertex_shader {
struct draw_vertex_shader base;
struct gallivm_prog *llvm_prog;
+ struct tgsi_exec_machine *machine;
};
@@ -54,121 +55,68 @@ static void
vs_llvm_prepare( struct draw_vertex_shader *base,
struct draw_context *draw )
{
- draw_update_vertex_fetch( draw );
}
-/**
- * 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
- */
-static boolean
-vs_llvm_run( struct draw_vertex_shader *base,
- struct draw_context *draw,
- const unsigned *elts,
- unsigned count,
- void *vOut )
+
+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;
- unsigned int clipped = 0;
-
- ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_MAX_ATTRIBS);
- ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_MAX_ATTRIBS);
- const float *scale = draw->viewport.scale;
- const float *trans = draw->viewport.translate;
-
-
- assert(count <= 4);
- assert(draw->vertex_shader->state->output_semantic_name[0]
- == TGSI_SEMANTIC_POSITION);
+ struct tgsi_exec_machine *machine = shader->machine;
+ unsigned int i, j;
+ unsigned slot;
- /* Consts does not require 16 byte alignment. */
- machine->Consts = (float (*)[4]) draw->user.constants;
- machine->Inputs = ALIGN16_ASSIGN(inputs);
- if (draw->rasterizer->bypass_vs) {
- /* outputs are just the inputs */
- machine->Outputs = machine->Inputs;
- }
- else {
- machine->Outputs = ALIGN16_ASSIGN(outputs);
- }
+ 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];
+ }
- draw->vertex_fetch.fetch_func( draw, machine, elts, count );
+ input = (const float (*)[4])((const char *)input + input_stride);
+ }
- if (!draw->rasterizer->bypass_vs) {
/* run shader */
gallivm_cpu_vs_exec(shader->llvm_prog,
machine->Inputs,
machine->Outputs,
- machine->Consts,
+ (float (*)[4]) constants,
machine->Temps);
- }
- /* store machine results */
- for (j = 0; j < count; j++) {
- unsigned slot;
- float x, y, z, w;
-
- 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);
- clipped += vOut[j]->clipmask;
-
- /* divide by w */
- w = 1.0f / w;
- x *= w;
- y *= w;
- z *= w;
- }
- else {
- vOut[j]->clipmask = 0;
- }
- vOut[j]->edgeflag = 1;
- vOut[j]->vertex_id = UNDEFINED_VERTEX_ID;
-
- if (!draw->identity_viewport) {
- /* 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]->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.
+ /* 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];
+ 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];
+ }
+ output = (float (*)[4])((char *)output + output_stride);
}
- } /* loop over vertices */
- return clipped != 0;
+ }
}
+
+
static void
vs_llvm_delete( struct draw_vertex_shader *base )
{
@@ -198,15 +146,19 @@ 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(vs->base.state.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;
+ vs->machine = &draw->machine;
{
struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);
gallivm_ir_set_layout(ir, GALLIVM_SOA);
gallivm_ir_set_components(ir, 4);
- gallivm_ir_fill_from_tgsi(ir, vs->base.state->tokens);
+ gallivm_ir_fill_from_tgsi(ir, vs->base.state.tokens);
vs->llvm_prog = gallivm_ir_compile(ir);
gallivm_ir_delete(ir);
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-08-26 12:06:08 +0000