summaryrefslogtreecommitdiffstats
path: root/src/gallium
diff options
context:
space:
mode:
authorMarek Olšák <[email protected]>2015-08-23 18:57:44 +0200
committerMarek Olšák <[email protected]>2015-08-23 19:01:15 +0200
commita83c36b5c0c64c717ced76db89bab900006648aa (patch)
tree137aca2eece388c6142ad225f88c668f08bda6a8 /src/gallium
parente18c29b03105567cf20bc235ce23cf08986cc537 (diff)
Revert "radeon/winsys: increase the IB size for VM"
This reverts commit 567394112d904096abff1d994ab952f475dfb444. It regressed performance. It looks like smaller IBs are better, because the GPU goes idle quicker and there is less waiting for buffers and fences. Cc: 11.0 <[email protected]>
Diffstat (limited to 'src/gallium')
-rw-r--r--src/gallium/winsys/radeon/drm/radeon_drm_cs.c8
-rw-r--r--src/gallium/winsys/radeon/drm/radeon_drm_cs.h2
-rw-r--r--src/gallium/winsys/radeon/drm/radeon_drm_winsys.c12
-rw-r--r--src/gallium/winsys/radeon/drm/radeon_drm_winsys.h1
4 files changed, 6 insertions, 17 deletions
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index 7a267f9acbf..f04a696988a 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -97,22 +97,17 @@ static boolean radeon_init_cs_context(struct radeon_cs_context *csc,
{
int i;
- csc->buf = MALLOC(ws->ib_max_size);
- if (!csc->buf)
- return FALSE;
csc->fd = ws->fd;
csc->nrelocs = 512;
csc->relocs_bo = (struct radeon_bo**)
CALLOC(1, csc->nrelocs * sizeof(struct radeon_bo*));
if (!csc->relocs_bo) {
- FREE(csc->buf);
return FALSE;
}
csc->relocs = (struct drm_radeon_cs_reloc*)
CALLOC(1, csc->nrelocs * sizeof(struct drm_radeon_cs_reloc));
if (!csc->relocs) {
- FREE(csc->buf);
FREE(csc->relocs_bo);
return FALSE;
}
@@ -165,7 +160,6 @@ static void radeon_destroy_cs_context(struct radeon_cs_context *csc)
radeon_cs_context_cleanup(csc);
FREE(csc->relocs_bo);
FREE(csc->relocs);
- FREE(csc->buf);
}
@@ -206,7 +200,7 @@ radeon_drm_cs_create(struct radeon_winsys_ctx *ctx,
cs->cst = &cs->csc2;
cs->base.buf = cs->csc->buf;
cs->base.ring_type = ring_type;
- cs->base.max_dw = ws->ib_max_size / 4;
+ cs->base.max_dw = ARRAY_SIZE(cs->csc->buf);
p_atomic_inc(&ws->num_cs);
return &cs->base;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.h b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
index ab154945880..6ceb8e98ee7 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
@@ -30,7 +30,7 @@
#include "radeon_drm_bo.h"
struct radeon_cs_context {
- uint32_t *buf;
+ uint32_t buf[16 * 1024];
int fd;
struct drm_radeon_cs cs;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
index b70bbaa54a3..f7784fb795e 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
@@ -395,20 +395,16 @@ static boolean do_winsys_init(struct radeon_drm_winsys *ws)
}
ws->info.r600_virtual_address = FALSE;
- ws->ib_max_size = 64 * 1024;
-
if (ws->info.drm_minor >= 13) {
+ uint32_t ib_vm_max_size;
+
ws->info.r600_virtual_address = TRUE;
if (!radeon_get_drm_value(ws->fd, RADEON_INFO_VA_START, NULL,
&ws->va_start))
ws->info.r600_virtual_address = FALSE;
-
- if (radeon_get_drm_value(ws->fd, RADEON_INFO_IB_VM_MAX_SIZE, NULL,
- &ws->ib_max_size))
- ws->ib_max_size *= 4; /* the kernel returns the size in dwords */
- else
+ if (!radeon_get_drm_value(ws->fd, RADEON_INFO_IB_VM_MAX_SIZE, NULL,
+ &ib_vm_max_size))
ws->info.r600_virtual_address = FALSE;
-
radeon_get_drm_value(ws->fd, RADEON_INFO_VA_UNMAP_WORKING, NULL,
&ws->va_unmap_working);
}
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
index c1a8d6ae564..308b5bd976d 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
@@ -73,7 +73,6 @@ struct radeon_drm_winsys {
enum radeon_generation gen;
struct radeon_info info;
- uint32_t ib_max_size;
uint32_t va_start;
uint32_t va_unmap_working;
uint32_t accel_working2;