KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

winsys/radeon: consolidate hash table lookup 

I should have done this long ago.

Reviewed-by: Christian König <christian.koenig@amd.com>
This commit is contained in:
Marek Olšák 2014-04-11 13:20:02 +02:00
parent d3c0e236f2
commit 927213f33d
2 changed files with 47 additions and 66 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

106
src/gallium/winsys/radeon/drm/radeon_drm_cs.c
View File

@ -188,41 +188,40 @@ static INLINE void update_reloc(struct drm_radeon_cs_reloc *reloc,
reloc->flags = MAX2(reloc->flags, priority); reloc->flags = MAX2(reloc->flags, priority);
} }
int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo) int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo,
struct drm_radeon_cs_reloc **out_reloc)
{ {
struct drm_radeon_cs_reloc *reloc; struct drm_radeon_cs_reloc *reloc = NULL;
unsigned i;
unsigned hash = bo->handle & (sizeof(csc->is_handle_added)-1); unsigned hash = bo->handle & (sizeof(csc->is_handle_added)-1);
int i = -1;
if (csc->is_handle_added[hash]) { if (csc->is_handle_added[hash]) {
i = csc->reloc_indices_hashlist[hash]; i = csc->reloc_indices_hashlist[hash];
reloc = &csc->relocs[i]; reloc = &csc->relocs[i];
if (reloc->handle == bo->handle) {
return i;
}
/* Hash collision, look for the BO in the list of relocs linearly. */ if (reloc->handle != bo->handle) {
for (i = csc->crelocs; i != 0;) { /* Hash collision, look for the BO in the list of relocs linearly. */
--i; for (i = csc->crelocs - 1; i >= 0; i--) {
reloc = &csc->relocs[i]; reloc = &csc->relocs[i];
if (reloc->handle == bo->handle) { if (reloc->handle == bo->handle) {
/* Put this reloc in the hash list. /* Put this reloc in the hash list.
* This will prevent additional hash collisions if there are * This will prevent additional hash collisions if there are
* several consecutive get_reloc calls for the same buffer. * several consecutive get_reloc calls for the same buffer.
* *
* Example: Assuming buffers A,B,C collide in the hash list, * Example: Assuming buffers A,B,C collide in the hash list,
* the following sequence of relocs: * the following sequence of relocs:
* AAAAAAAAAAABBBBBBBBBBBBBBCCCCCCCC * AAAAAAAAAAABBBBBBBBBBBBBBCCCCCCCC
* will collide here: ^ and here: ^, * will collide here: ^ and here: ^,
* meaning that we should get very few collisions in the end. */ * meaning that we should get very few collisions in the end. */
csc->reloc_indices_hashlist[hash] = i; csc->reloc_indices_hashlist[hash] = i;
/*printf("write_reloc collision, hash: %i, handle: %i\n", hash, bo->handle);*/ break;
return i; }
} }
} }
} }
if (out_reloc)
return -1; *out_reloc = reloc;
return i;
} }
static unsigned radeon_add_reloc(struct radeon_drm_cs *cs, static unsigned radeon_add_reloc(struct radeon_drm_cs *cs,
@ -237,45 +236,28 @@ static unsigned radeon_add_reloc(struct radeon_drm_cs *cs,
unsigned hash = bo->handle & (sizeof(csc->is_handle_added)-1); unsigned hash = bo->handle & (sizeof(csc->is_handle_added)-1);
enum radeon_bo_domain rd = usage & RADEON_USAGE_READ ? domains : 0; enum radeon_bo_domain rd = usage & RADEON_USAGE_READ ? domains : 0;
enum radeon_bo_domain wd = usage & RADEON_USAGE_WRITE ? domains : 0; enum radeon_bo_domain wd = usage & RADEON_USAGE_WRITE ? domains : 0;
bool update_hash = TRUE; int i = -1;
int i;
priority = MIN2(priority, 15); priority = MIN2(priority, 15);
*added_domains = 0; *added_domains = 0;
if (csc->is_handle_added[hash]) { i = radeon_get_reloc(csc, bo, &reloc);
i = csc->reloc_indices_hashlist[hash];
reloc = &csc->relocs[i];
if (reloc->handle != bo->handle) { if (i >= 0) {
/* Hash collision, look for the BO in the list of relocs linearly. */ update_reloc(reloc, rd, wd, priority, added_domains);
for (i = csc->crelocs - 1; i >= 0; i--) {
reloc = &csc->relocs[i];
if (reloc->handle == bo->handle) {
/*printf("write_reloc collision, hash: %i, handle: %i\n", hash, bo->handle);*/
break;
}
}
}
if (i >= 0) { /* For async DMA, every add_reloc call must add a buffer to the list
update_reloc(reloc, rd, wd, priority, added_domains); * no matter how many duplicates there are. This is due to the fact
* the DMA CS checker doesn't use NOP packets for offset patching,
/* For async DMA, every add_reloc call must add a buffer to the list * but always uses the i-th buffer from the list to patch the i-th
* no matter how many duplicates there are. This is due to the fact * offset. If there are N offsets in a DMA CS, there must also be N
* the DMA CS checker doesn't use NOP packets for offset patching, * buffers in the relocation list.
* but always uses the i-th buffer from the list to patch the i-th *
* offset. If there are N offsets in a DMA CS, there must also be N * This doesn't have to be done if virtual memory is enabled,
* buffers in the relocation list. * because there is no offset patching with virtual memory.
* */
* This doesn't have to be done if virtual memory is enabled, if (cs->base.ring_type != RING_DMA || cs->ws->info.r600_virtual_address) {
* because there is no offset patching with virtual memory. return i;
*/
if (cs->base.ring_type != RING_DMA || cs->ws->info.r600_virtual_address) {
csc->reloc_indices_hashlist[hash] = i;
return i;
}
update_hash = FALSE;
} }
} }
@ -304,9 +286,7 @@ static unsigned radeon_add_reloc(struct radeon_drm_cs *cs,
reloc->flags = priority; reloc->flags = priority;
csc->is_handle_added[hash] = TRUE; csc->is_handle_added[hash] = TRUE;
if (update_hash) { csc->reloc_indices_hashlist[hash] = csc->crelocs;
csc->reloc_indices_hashlist[hash] = csc->crelocs;
}
csc->chunks[1].length_dw += RELOC_DWORDS; csc->chunks[1].length_dw += RELOC_DWORDS;
@ -384,7 +364,7 @@ static void radeon_drm_cs_write_reloc(struct radeon_winsys_cs *rcs,
{ {
struct radeon_drm_cs *cs = radeon_drm_cs(rcs); struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
struct radeon_bo *bo = (struct radeon_bo*)buf; struct radeon_bo *bo = (struct radeon_bo*)buf;
unsigned index = radeon_get_reloc(cs->csc, bo); unsigned index = radeon_get_reloc(cs->csc, bo, NULL);
if (index == -1) { if (index == -1) {
fprintf(stderr, "radeon: Cannot get a relocation in %s.\n", __func__); fprintf(stderr, "radeon: Cannot get a relocation in %s.\n", __func__);
@ -600,7 +580,7 @@ static boolean radeon_bo_is_referenced(struct radeon_winsys_cs *rcs,
if (!bo->num_cs_references) if (!bo->num_cs_references)
return FALSE; return FALSE;
index = radeon_get_reloc(cs->csc, bo); index = radeon_get_reloc(cs->csc, bo, NULL);
if (index == -1) if (index == -1)
return FALSE; return FALSE;

7
src/gallium/winsys/radeon/drm/radeon_drm_cs.h
View File

@ -80,7 +80,8 @@ struct radeon_drm_cs {
struct radeon_bo *trace_buf; struct radeon_bo *trace_buf;
}; };
int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo); int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo,
struct drm_radeon_cs_reloc **out_reloc);
static INLINE struct radeon_drm_cs * static INLINE struct radeon_drm_cs *
radeon_drm_cs(struct radeon_winsys_cs *base) radeon_drm_cs(struct radeon_winsys_cs *base)
@ -94,7 +95,7 @@ radeon_bo_is_referenced_by_cs(struct radeon_drm_cs *cs,
{ {
int num_refs = bo->num_cs_references; int num_refs = bo->num_cs_references;
return num_refs == bo->rws->num_cs || return num_refs == bo->rws->num_cs ||
(num_refs && radeon_get_reloc(cs->csc, bo) != -1); (num_refs && radeon_get_reloc(cs->csc, bo, NULL) != -1);
} }
static INLINE boolean static INLINE boolean
@ -106,7 +107,7 @@ radeon_bo_is_referenced_by_cs_for_write(struct radeon_drm_cs *cs,
if (!bo->num_cs_references) if (!bo->num_cs_references)
return FALSE; return FALSE;
index = radeon_get_reloc(cs->csc, bo); index = radeon_get_reloc(cs->csc, bo, NULL);
if (index == -1) if (index == -1)
return FALSE; return FALSE;