fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-03-04 10:20:33 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Redo the way we pass arguments to the llvm.

Browse source 
simply pass aligned arrays, they should cast to vectors without
any problems. also remove unnecessary memset
This commit is contained in:
Zack Rusin 2007-11-07 13:26:45 -05:00
parent 9d6e6e86d9
commit 6dc4e6ae15
5 changed files with 438 additions and 876 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
src/mesa/pipe/draw/draw_vertex_shader_llvm.c
View file

@ -115,13 +115,12 @@ void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw)
unsigned i;
struct vertex_header *dests[VS_QUEUE_LENGTH];
float inputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4];
float outputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4];
float inputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4] ALIGN16_ATTRIB;
float outputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4] ALIGN16_ATTRIB;
float (*consts)[4] = (float (*)[4]) draw->user.constants;
struct gallivm_prog *prog = draw->vertex_shader->llvm_prog;
const float *scale = draw->viewport.scale;
const float *trans = draw->viewport.translate;
/* fetch the inputs */
for (i = 0; i < draw->vs.queue_nr; ++i) {
unsigned elt = draw->vs.queue[i].elt;
@ -135,6 +134,7 @@ void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw)
draw->vertex_shader->state->num_inputs,
draw->vertex_info.num_attribs - 2);
/* store machine results */
for (int i = 0; i < draw->vs.queue_nr; ++i) {
unsigned slot;
@ -158,7 +158,6 @@ void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw)
vOut->clipmask = compute_clipmask(vOut->clip, draw->plane, draw->nr_planes);
vOut->edgeflag = 1;
/* divide by w */
w = 1.0f / w;
x *= w;

1280
src/mesa/pipe/llvm/llvm_base_shader.cpp
View file

File diff suppressed because it is too large Load diff

20
src/mesa/pipe/llvm/llvm_entry.c
View file

@ -68,6 +68,7 @@ compute_clipmask(float4 clip, float4 (*plane), unsigned nr)
return mask;
}
inline void collect_results(float4 *results, struct vertex_header *vOut,
float4 *planes, int nr_planes,
float4 scale, float4 trans,
@ -76,7 +77,6 @@ inline void collect_results(float4 *results, struct vertex_header *vOut,
/* store results */
unsigned slot;
float x, y, z, w;
/* Handle attr[0] (position) specially:
*/
float4 res0 = results[0];
@ -85,7 +85,6 @@ inline void collect_results(float4 *results, struct vertex_header *vOut,
y = clip[1] = res0.y;
z = clip[2] = res0.z;
w = clip[3] = res0.w;
vOut->clipmask = compute_clipmask(res0, planes, nr_planes);
vOut->edgeflag = 1;
@ -176,23 +175,20 @@ struct ShaderInput
extern void execute_shader(struct ShaderInput *input);
void run_vertex_shader(float (*ainputs)[16][4],
float (*dests)[16][4],
void run_vertex_shader(float4 (*inputs)[16],
float4 (*results)[16],
float (*aconsts)[4],
int num_vertices,
int num_inputs,
int num_attribs,
int num_consts)
{
float4 inputs[16*32*4][16];
float4 consts[32];
float4 results[16*32*4][16];
float4 temps[128];//MAX_PROGRAM_TEMPS
struct ShaderInput args;
/*printf("XXX LLVM run_vertex_shader vertices = %d, inputs = %d, attribs = %d, consts = %d\n",
num_vertices, num_inputs, num_attribs, num_consts);*/
from_array(inputs, ainputs, num_vertices, num_inputs);
from_consts(consts, aconsts, num_consts);
args.consts = consts;
args.temps = temps;
@ -200,7 +196,6 @@ void run_vertex_shader(float (*ainputs)[16][4],
args.dests = results[i];
args.inputs = inputs[i];
execute_shader(&args);
to_array(dests[i], args.dests, num_attribs);
}
}
@ -227,22 +222,19 @@ struct tgsi_sampler
int run_fragment_shader(float x, float y,
float (*dests)[16][4],
float (*ainputs)[16][4],
float4 (*results)[16],
float4 (*inputs)[16],
int num_inputs,
float (*aconsts)[4],
int num_consts,
struct tgsi_sampler *samplers)
{
float4 inputs[4][16];
float4 consts[32];
float4 results[4][16];
float4 temps[128];//MAX_PROGRAM_TEMPS
struct ShaderInput args;
int mask = 0;
args.kilmask = 0;
from_array(inputs, ainputs, 4, num_inputs);
from_consts(consts, aconsts, num_consts);
args.consts = consts;
args.temps = temps;
@ -254,8 +246,6 @@ int run_fragment_shader(float x, float y,
args.kilmask = 0;
execute_shader(&args);
args.kilmask = mask | (args.kilmask << i);
to_array(dests[i], args.dests, 2);
}
return ~args.kilmask;
}

2
src/mesa/pipe/p_compiler.h
View file

@ -73,9 +73,11 @@ typedef unsigned long long uint64;
#if defined __GNUC__
#define ALIGN16_DECL(TYPE, NAME, SIZE) TYPE NAME##___aligned[SIZE] __attribute__(( aligned( 16 ) ))
#define ALIGN16_ASSIGN(NAME) NAME##___aligned
#define ALIGN16_ATTRIB __attribute__(( aligned( 16 ) ))
#else
#define ALIGN16_DECL(TYPE, NAME, SIZE) TYPE NAME##___unaligned[SIZE + 1]
#define ALIGN16_ASSIGN(NAME) align16(NAME##___unaligned)
#define ALIGN16_ATTRIB
#endif

5
src/mesa/pipe/softpipe/sp_quad_fs.c
View file

@ -173,12 +173,11 @@ shade_quad_llvm(struct quad_stage *qs,
{
struct quad_shade_stage *qss = quad_shade_stage(qs);
struct softpipe_context *softpipe = qs->softpipe;
float dests[4][16][4];
float dests[4][16][4] ALIGN16_ATTRIB;
float inputs[4][16][4] ALIGN16_ATTRIB;
const float fx = (float) quad->x0;
const float fy = (float) quad->y0;
struct gallivm_prog *llvm = qss->llvm_prog;
float inputs[4][16][4];
memset(inputs, 0, sizeof(inputs));
inputs[0][0][0] = fx;
inputs[1][0][0] = fx + 1.0f;