mrq/GLdc
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix perf hit

Browse Source
This commit is contained in:
Luke Benstead 2021-09-14 09:20:00 +01:00
parent 9c2907e41d
commit 7992c89758
1 changed files with 22 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

78
GL/draw.c
View File

@ -843,9 +843,6 @@ static void generateArraysFastPath(SubmissionTarget* target, const GLsizei first
const GLubyte* st = (ENABLED_VERTEX_ATTRIBUTES & ST_ENABLED_FLAG) ? ATTRIB_POINTERS.st.ptr + (first * ststride) : NULL; const GLubyte* st = (ENABLED_VERTEX_ATTRIBUTES & ST_ENABLED_FLAG) ? ATTRIB_POINTERS.st.ptr + (first * ststride) : NULL;
const GLubyte* n = (ENABLED_VERTEX_ATTRIBUTES & NORMAL_ENABLED_FLAG) ? ATTRIB_POINTERS.normal.ptr + (first * nstride) : NULL; const GLubyte* n = (ENABLED_VERTEX_ATTRIBUTES & NORMAL_ENABLED_FLAG) ? ATTRIB_POINTERS.normal.ptr + (first * nstride) : NULL;
VertexExtra* ve = ve_start;
Vertex* it = start;
const float w = 1.0f; const float w = 1.0f;
if(!pos) { if(!pos) {
@ -873,65 +870,34 @@ static void generateArraysFastPath(SubmissionTarget* target, const GLsizei first
nstride = 0; nstride = 0;
} }
#define BATCH_SIZE 64 int_fast32_t i = count;
VertexExtra* ve = ve_start;
Vertex* it = start;
int_fast32_t start_idx = 0; while(i--) {
int_fast32_t end_idx = BATCH_SIZE; it->flags = GPU_CMD_VERTEX;
TransformVertex((const float*) pos, &w, it->xyz, &it->w);
pos += vstride;
PREFETCH(pos);
while(1) { *((Float2*) it->uv) = *((Float2*) uv);
for(int_fast32_t i = start_idx; i < end_idx; ++i) { uv += uvstride;
PREFETCH(pos + vstride); PREFETCH(uv);
it = start + i; *((uint32_t*) it->bgra) = *((uint32_t*) col);
col += dstride;
PREFETCH(col);
it->flags = GPU_CMD_VERTEX; *((Float2*) ve->st) = *((Float2*) st);
st += ststride;
PREFETCH(st);
TransformVertex((const float*) pos, &w, it->xyz, &it->w); *((Float3*) ve->nxyz) = *((Float3*) n);
pos += vstride; n += nstride;
} PREFETCH(n);
for(int_fast32_t i = start_idx; i < end_idx; ++i) { ++it;
PREFETCH(uv + uvstride); ++ve;
it = start + i;
*((Float2*) it->uv) = *((Float2*) uv);
uv += uvstride;
}
for(int_fast32_t i = start_idx; i < end_idx; ++i) {
PREFETCH(col + dstride);
it = start + i;
*((uint32_t*) it->bgra) = *((uint32_t*) col);
col += dstride;
}
for(int_fast32_t i = start_idx; i < end_idx; ++i) {
PREFETCH(st + ststride);
ve = ve_start + i;
*((Float2*) ve->st) = *((Float2*) st);
st += ststride;
}
for(int_fast32_t i = start_idx; i < end_idx; ++i) {
PREFETCH(n + nstride);
ve = ve_start + i;
*((Float3*) ve->nxyz) = *((Float3*) n);
n += nstride;
}
if(end_idx == count) {
break;
}
start_idx = end_idx;
PREFETCH(it + start_idx);
end_idx += BATCH_SIZE;
end_idx = MIN(end_idx, count);
} }
} }