r600g: emit CS using radeon_winsys
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 0562b6d..84cfa2a 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -236,16 +236,16 @@
struct list_head resource_dirty;
struct list_head enable_list;
unsigned pm4_ndwords;
- unsigned pm4_cdwords;
unsigned pm4_dirty_cdwords;
unsigned ctx_pm4_ndwords;
unsigned init_dwords;
unsigned creloc;
- unsigned *reloc;
struct radeon_bo **bo;
u32 *pm4;
+ unsigned pm4_cdwords;
+
struct list_head query_list;
unsigned num_query_running;
unsigned backend_mask;
diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index cd63c8d..29da7be 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -995,11 +995,7 @@
goto out_err;
}
ctx->pm4_ndwords = RADEON_CTX_MAX_PM4;
- ctx->pm4 = calloc(ctx->pm4_ndwords, 4);
- if (ctx->pm4 == NULL) {
- r = -ENOMEM;
- goto out_err;
- }
+ ctx->pm4 = ctx->cs->buf;
r600_init_cs(ctx);
/* save 16dwords space for fence mecanism */
diff --git a/src/gallium/winsys/r600/drm/r600_drm.c b/src/gallium/winsys/r600/drm/r600_drm.c
index f0ef55e..270a07a 100644
--- a/src/gallium/winsys/r600/drm/r600_drm.c
+++ b/src/gallium/winsys/r600/drm/r600_drm.c
@@ -292,6 +292,9 @@
radeon_get_backend_map(radeon);
}
+ /* XXX disable ioctl thread offloading until the porting is done. */
+ setenv("RADEON_THREAD", "0", 0);
+
return radeon;
}
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 1950e58..59450b5 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -776,7 +776,6 @@
free(ctx->range);
free(ctx->blocks);
free(ctx->bo);
- free(ctx->pm4);
ctx->radeon->ws->cs_destroy(ctx->cs);
memset(ctx, 0, sizeof(struct r600_context));
@@ -920,11 +919,7 @@
goto out_err;
}
ctx->pm4_ndwords = RADEON_CTX_MAX_PM4;
- ctx->pm4 = calloc(ctx->pm4_ndwords, 4);
- if (ctx->pm4 == NULL) {
- r = -ENOMEM;
- goto out_err;
- }
+ ctx->pm4 = ctx->cs->buf;
r600_init_cs(ctx);
/* save 16dwords space for fence mecanism */
@@ -1492,10 +1487,6 @@
void r600_context_flush(struct r600_context *ctx)
{
- struct drm_radeon_cs drmib = {};
- struct drm_radeon_cs_chunk chunks[2];
- uint64_t chunk_array[2];
- int r;
struct r600_block *enable_block = NULL;
if (ctx->pm4_cdwords == ctx->init_dwords)
@@ -1513,27 +1504,12 @@
ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
-#if 1
- /* emit cs */
- drmib.num_chunks = 2;
- drmib.chunks = (uint64_t)(uintptr_t)chunk_array;
- chunks[0].chunk_id = RADEON_CHUNK_ID_IB;
- chunks[0].length_dw = ctx->pm4_cdwords;
- chunks[0].chunk_data = (uint64_t)(uintptr_t)ctx->pm4;
- chunks[1].chunk_id = RADEON_CHUNK_ID_RELOCS;
- chunks[1].length_dw = ctx->creloc * 4;
- chunks[1].chunk_data = (uint64_t)(uintptr_t)ctx->reloc;
- chunk_array[0] = (uint64_t)(uintptr_t)&chunks[0];
- chunk_array[1] = (uint64_t)(uintptr_t)&chunks[1];
- r = drmCommandWriteRead(ctx->radeon->info.fd, DRM_RADEON_CS, &drmib,
- sizeof(struct drm_radeon_cs));
- if (r) {
- fprintf(stderr, "radeon: The kernel rejected CS, "
- "see dmesg for more information.\n");
- }
-#else
- *ctx->radeon->cfence = ctx->radeon->fence;
-#endif
+ /* Flush the CS. */
+ ctx->cs->cdw = ctx->pm4_cdwords;
+ ctx->radeon->ws->cs_flush(ctx->cs, 0);
+ /* We need to get the pointer to the other CS,
+ * the command streams are double-buffered. */
+ ctx->pm4 = ctx->cs->buf;
/* restart */
for (int i = 0; i < ctx->creloc; i++) {
@@ -1544,7 +1520,6 @@
ctx->pm4_dirty_cdwords = 0;
ctx->pm4_cdwords = 0;
ctx->flags = 0;
- ctx->radeon->ws->cs_flush(ctx->cs, 0);
r600_init_cs(ctx);
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h
index 5bb515d..83e964b 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -134,10 +134,11 @@
assert(bo != NULL);
- reloc_index = ctx->radeon->ws->trans_add_reloc(
- ctx->cs, bo->cs_buf,
- rbo->domains, rbo->domains,
- (void**)&ctx->reloc, &ctx->creloc);
+ reloc_index =
+ ctx->radeon->ws->trans_add_reloc(ctx->cs, bo->cs_buf, rbo->domains, rbo->domains);
+
+ if (reloc_index >= ctx->creloc)
+ ctx->creloc = reloc_index+1;
radeon_bo_reference(ctx->radeon, &ctx->bo[reloc_index], bo);
return reloc_index * 4;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index 9a1e169..09befb3 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -473,8 +473,7 @@
static unsigned trans_add_reloc(struct radeon_winsys_cs *rcs,
struct radeon_winsys_cs_handle *buf,
enum radeon_bo_domain rd,
- enum radeon_bo_domain wd,
- void **reloc_list, unsigned *reloc_count)
+ enum radeon_bo_domain wd)
{
struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
struct radeon_bo *bo = (struct radeon_bo*)buf;
@@ -487,8 +486,6 @@
if (added_domains & RADEON_DOMAIN_VRAM)
cs->csc->used_vram += bo->size;
- *reloc_list = cs->csc->relocs;
- *reloc_count = cs->csc->crelocs;
return index;
}
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index 8e81fa1..073b6aa 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -332,8 +332,7 @@
unsigned (*trans_add_reloc)(struct radeon_winsys_cs *cs,
struct radeon_winsys_cs_handle *buf,
enum radeon_bo_domain rd,
- enum radeon_bo_domain wd,
- void **reloc_list, unsigned *reloc_count);
+ enum radeon_bo_domain wd);
};
#endif