igt tests/rendercopy: Add option to test resource streamer functionality
This patchseries tests a resource streamer feature called hw-generated binding tables, which can be enabled from userspace. It is essentially a newer format that uses the hw to generate and submit binding tables to the GPU. It uses the basic rendercopy test to toggle the feature on and off and to verify if the resulting data is still correct.
Abdiel Janulgue (2):
rendercopy/bdw: Enable hw-generated binding tables tests/gem_render_copy: Add option to test resource streamer
lib/gen8_render.h | 13 +++++++ lib/intel_batchbuffer.c | 12 ++++++ lib/intel_batchbuffer.h | 3 ++ lib/intel_reg.h | 3 ++ lib/rendercopy_gen8.c | 97 ++++++++++++++++++++++++++++++++++++++++++++--- tests/gem_render_copy.c | 8 +++- 6 files changed, 130 insertions(+), 6 deletions(-)
PATCH 1/2 rendercopy/bdw: Enable hw-generated binding tables
Use on-chip hw-binding table generator to generate binding tables when the test emits SURFACE_STATES packets. The hw generates these binding table packets internally so we don't have to allocate space on the batchbuffer.
--- lib/gen8_render.h | 13 +++++++ lib/intel_batchbuffer.c | 12 ++++++ lib/intel_batchbuffer.h | 3 ++ lib/intel_reg.h | 3 ++ lib/rendercopy_gen8.c | 97 ++++++++++++++++++++++++++++++++++++++++++++--- 5 files changed, 123 insertions(+), 5 deletions(-)
lib/gen8_render.h +13 -0
@@ -83,6 +83,19 @@/* Random shifts */ #define GEN8_3DSTATE_PS_MAX_THREADS_SHIFT 23 +/* GEN8+ resource streamer */ +#define GEN8_3DSTATE_BINDING_TABLE_POOL_ALLOC (0x7919 << 16)/* GEN8 */ +# define BINDING_TABLE_POOL_ENABLE 0x0800 +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_VS (0x7843 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_GS (0x7844 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_HS (0x7845 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_DS (0x7846 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_PS (0x7847 << 16)/* GEN8 */ + +/* Toggling RS needs PIPE_CONTROL SC invalidate */ +#define _3DSTATE_PIPE_CONTROL ((0x3 << 29) | (3 << 27) | (2 << 24)) +#define PIPE_CONTROL_STATE_CACHE_INVALIDATE (1 << 2) +/* Shamelessly ripped from mesa */ struct gen8_surface_state {
lib/intel_batchbuffer.c +12 -0
@@ -81,6 +81,14 @@ intel_batchbuffer_reset(struct intel_batchbuffer *batch)memset (batch->buffer,0 ,sizeof (batch->buffer)); batch->ptr = batch->buffer; + +if (batch->hw_bt_pool_bo !=NULL ) { +drm_intel_bo_unreference (batch->hw_bt_pool_bo); + batch->hw_bt_pool_bo =NULL ; + } + + batch->hw_bt_pool_bo =drm_intel_bo_alloc (batch->bufmgr,"hw_bt" , +131072 ,4096 ); }/**
@@ -116,6 +124,10 @@ intel_batchbuffer_free(struct intel_batchbuffer *batch) {drm_intel_bo_unreference (batch->bo); batch->bo =NULL ; + +drm_intel_bo_unreference (batch->hw_bt_pool_bo); + batch->hw_bt_pool_bo =NULL ; +free (batch); }
lib/intel_batchbuffer.h +3 -0
@@ -18,6 +18,9 @@ struct intel_batchbuffer {uint8_t buffer[BATCH_SZ ];uint8_t *ptr;uint8_t *state; + +drm_intel_bo *hw_bt_pool_bo; +int use_resource_streamer; }; struct intel_batchbuffer *intel_batchbuffer_alloc(drm_intel_bufmgr *bufmgr,
lib/intel_reg.h +3 -0
@@ -2560,6 +2560,9 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.#define MI_BATCH_NON_SECURE (1)#define MI_BATCH_NON_SECURE_I965 (1 << 8) +/* Resource Streamer */ +#define MI_RS_CONTROL (0x6 << 23) +#define MAX_DISPLAY_PIPES 2 typedef enum {
lib/rendercopy_gen8.c +92 -5
165 lines not on this page · raw message
Signature · 1 line
1.7.9.5
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 124 lines · Abdiel Janulgue
Use on-chip hw-binding table generator to generate binding tables when the test emits
SURFACE_STATESpackets. The hw generates these binding table packets internally so we don't have to allocate space on the batchbuffer.Signed-off-by Abdiel Janulgue <[email protected]>--- lib/gen8_render.h | 13 +++++++ lib/intel_batchbuffer.c | 12 ++++++ lib/intel_batchbuffer.h | 3 ++ lib/intel_reg.h | 3 ++ lib/rendercopy_gen8.c | 97 ++++++++++++++++++++++++++++++++++++++++++++--- 5 files changed, 123 insertions(+), 5 deletions(-)lib/gen8_render.h +13
-0@@ -83,6 +83,19 @@/* Random shifts */ #define GEN8_3DSTATE_PS_MAX_THREADS_SHIFT 23 +/* GEN8+ resource streamer */ +#define GEN8_3DSTATE_BINDING_TABLE_POOL_ALLOC (0x7919 << 16)/* GEN8 */ +# define BINDING_TABLE_POOL_ENABLE 0x0800 +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_VS (0x7843 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_GS (0x7844 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_HS (0x7845 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_DS (0x7846 << 16)/* GEN8 */ +#define GEN8_3DSTATE_BINDING_TABLE_EDIT_PS (0x7847 << 16)/* GEN8 */ + +/* Toggling RS needs PIPE_CONTROL SC invalidate */ +#define _3DSTATE_PIPE_CONTROL ((0x3 << 29) | (3 << 27) | (2 << 24)) +#define PIPE_CONTROL_STATE_CACHE_INVALIDATE (1 << 2) +/* Shamelessly ripped from mesa */ struct gen8_surface_state {lib/intel_batchbuffer.c +12
-0@@ -81,6 +81,14 @@ intel_batchbuffer_reset(struct intel_batchbuffer *batch)memset (batch->buffer,0 ,sizeof (batch->buffer)); batch->ptr = batch->buffer; + +if (batch->hw_bt_pool_bo !=NULL ) { +drm_intel_bo_unreference (batch->hw_bt_pool_bo); + batch->hw_bt_pool_bo =NULL ; + } + + batch->hw_bt_pool_bo =drm_intel_bo_alloc (batch->bufmgr,"hw_bt" , +131072 ,4096 ); }/** @@ -116,6 +124,10 @@ intel_batchbuffer_free(struct intel_batchbuffer *batch) {drm_intel_bo_unreference (batch->bo); batch->bo =NULL ; + +drm_intel_bo_unreference (batch->hw_bt_pool_bo); + batch->hw_bt_pool_bo =NULL ; +free (batch); }lib/intel_batchbuffer.h +3
-0@@ -18,6 +18,9 @@ struct intel_batchbuffer {uint8_t buffer[BATCH_SZ ];uint8_t *ptr;uint8_t *state; + +drm_intel_bo *hw_bt_pool_bo; +int use_resource_streamer; }; struct intel_batchbuffer *intel_batchbuffer_alloc(drm_intel_bufmgr *bufmgr,lib/intel_reg.h +3
-0@@ -2560,6 +2560,9 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.#define MI_BATCH_NON_SECURE (1)#define MI_BATCH_NON_SECURE_I965 (1 << 8) +/* Resource Streamer */ +#define MI_RS_CONTROL (0x6 << 23) +#define MAX_DISPLAY_PIPES 2 typedef enum {lib/rendercopy_gen8.c +7
-2@@ -169,12 +169,14 @@ static void gen6_render_flush(struct intel_batchbuffer *batch, drm_intel_context *context, uint32_t batch_end) { - int ret; + int ret = 0; +uint32_t flags =0 ; ret =drm_intel_bo_subdata (batch->bo,0 ,4096 , batch->buffer); + flags = batch->use_resource_streamer ?I915_EXEC_RESOURCE_STREAMER :0 ;if (ret ==0 ) ret =drm_intel_gem_bo_context_exec (batch->bo, context, - batch_end,0 ); + batch_end, flags |I915_EXEC_RENDER );assert (ret ==0 ); }@@ -228,18 +230,83 @@ gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf,return offset; } +static void +gen8_hw_binding_table (struct intel_batchbuffer *batch,bool enable) +{
lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 245
Quoted 10 lines
lib/rendercopy_gen8.c +10
-0+if (!enable) { +OUT_BATCH (MI_RS_CONTROL |0x0 ); + +OUT_BATCH (GEN8_3DSTATE_BINDING_TABLE_POOL_ALLOC | (4 -2 )); +/* binding table pool base address */ +OUT_BATCH (0 ); +OUT_BATCH (0 ); +/* Upper bound */ +OUT_BATCH (0 ); +
WaCsStallBeforeStateCacheInvalidate says you need to emit another PIPE_CONTROL here with CS stall bit set. And thanks to the other restrictions that PIPE_CONTORL needs one of the other magic bits also set. Stall at scoreboard perhaps? I think that's what we set in the kernel usually.
Quoted 5 lines
lib/rendercopy_gen8.c +5
-0+OUT_BATCH (_3DSTATE_PIPE_CONTROL | (6 -2 )); +OUT_BATCH (PIPE_CONTROL_STATE_CACHE_INVALIDATE ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +OUT_BATCH (0 );
lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 260
Quoted 10 lines
lib/rendercopy_gen8.c +10
-0+OUT_BATCH (0 ); + +return ; + } +OUT_BATCH (GEN8_3DSTATE_BINDING_TABLE_POOL_ALLOC | (4 -2 )); + +/* binding table pool base address */ +OUT_RELOC (batch->hw_bt_pool_bo,I915_GEM_DOMAIN_SAMPLER ,0 , +BINDING_TABLE_POOL_ENABLE ); +OUT_BATCH (0 );
I suppose it doesn't matter much which domain we use here. But the hardware writes to the BT pool so maybe you should set write domain as well?
lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 264
Quoted 4 lines
lib/rendercopy_gen8.c +4
-0+ +/* Upper bound */ +OUT_RELOC (batch->hw_bt_pool_bo,I915_GEM_DOMAIN_SAMPLER ,0 , + batch->hw_bt_pool_bo->size);
AFAICS on bdw this should be just the bt pool size and not a reloc.
lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 265
Quoted 1 line
lib/rendercopy_gen8.c +1
-0+
Again need another CS stall PIPE_CONTROL.
Quoted 16 lines
lib/rendercopy_gen8.c +16
-0+OUT_BATCH (_3DSTATE_PIPE_CONTROL | (6 -2 )); +OUT_BATCH (PIPE_CONTROL_STATE_CACHE_INVALIDATE ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +} + +static uint32_t +gen8_rs_bind_surfaces (struct intel_batchbuffer *batch, +struct igt_buf *src, +struct igt_buf *dst, +uint32_t *surf0,uint32_t *surf1) +{ + *surf0 =gen8_bind_buf (batch, dst,GEN6_SURFACEFORMAT_B8G8R8A8_UNORM ,1 ); + *surf1 =gen8_bind_buf (batch, src,GEN6_SURFACEFORMAT_B8G8R8A8_UNORM ,0 );
lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 291
Quoted 10 lines
lib/rendercopy_gen8.c +10
-0+ +return 0 ; +} + +static void +gen8_rs_edit_surfaces (struct intel_batchbuffer *batch, +uint32_t surf0,uint32_t surf1) +{ +OUT_BATCH (GEN8_3DSTATE_BINDING_TABLE_EDIT_PS | (4 -2 )); +OUT_BATCH (0x3 );
So I take it there's no need to clear the entries first?
Quoted 101 lines
lib/rendercopy_gen8.c +29
-3+ { +OUT_BATCH (0 <<16 | surf0 >>6 ); +OUT_BATCH (1 <<16 | surf1 >>6 ); + } +} +static uint32_t gen8_bind_surfaces (struct intel_batchbuffer *batch,struct igt_buf *src,struct igt_buf *dst) { -uint32_t *binding_table, offset; +uint32_t offset =0 ; +uint32_t *binding_table; binding_table =batch_alloc (batch,8 ,32 ); offset =batch_offset (batch, binding_table);annotation_add_state (&aub_annotations,AUB_TRACE_BINDING_TABLE , offset,8 ); - binding_table[0 ] =gen8_bind_buf (batch, dst,GEN6_SURFACEFORMAT_B8G8R8A8_UNORM ,1 ); binding_table[1] =@@ -512,6 +579,9 @@ gen7_emit_push_constants(struct intel_batchbuffer *batch) {static void gen8_emit_state_base_address (struct intel_batchbuffer *batch) { +if (batch->use_resource_streamer) +OUT_BATCH (MI_RS_CONTROL |0x0 ); +OUT_BATCH (GEN6_STATE_BASE_ADDRESS | (16 -2 ));/* general */ @@ -546,6 +616,9 @@ gen8_emit_state_base_address(struct intel_batchbuffer *batch) {OUT_BATCH (0xfffff000 |1 );/* intruction buffer size */ OUT_BATCH (1 <<12 |1 ); + +if (batch->use_resource_streamer) +OUT_BATCH (MI_RS_CONTROL |0x1 ); } static void@@ -918,6 +991,7 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,uint32_t scissor_state;uint32_t vertex_buffer;uint32_t batch_end; +uint32_t surf0 =0 , surf1 =1 ;intel_batchbuffer_flush_with_context (batch, context);@@ -927,7 +1001,11 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,annotation_init (&aub_annotations); - ps_binding_table =gen8_bind_surfaces (batch, src, dst); +if (batch->use_resource_streamer) + ps_binding_table =gen8_rs_bind_surfaces (batch, src, dst, + &surf0, &surf1); +else + ps_binding_table =gen8_bind_surfaces (batch, src, dst); ps_sampler_state =gen8_create_sampler (batch); ps_kernel_off =gen8_fill_ps (batch, ps_kernel,sizeof (ps_kernel)); vertex_buffer = gen7_fill_vertex_buffer_data(batch, src,@@ -955,6 +1033,9 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,gen8_emit_state_base_address (batch); +if (batch->use_resource_streamer) +gen8_hw_binding_table (batch, true); +OUT_BATCH (GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC );OUT_BATCH (viewport.cc_state);OUT_BATCH (GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP );@@ -978,6 +1059,9 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,gen8_emit_sf (batch); +if (batch->use_resource_streamer) +gen8_rs_edit_surfaces (batch, surf0, surf1); +OUT_BATCH (GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS );OUT_BATCH (ps_binding_table);@@ -1001,6 +1085,9 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,gen8_emit_vf_topology (batch);gen8_emit_primitive (batch, vertex_buffer); +if (batch->use_resource_streamer) +gen8_hw_binding_table (batch, false); +OUT_BATCH (MI_BATCH_BUFFER_END ); batch_end =batch_align (batch,8 );Signature · 6 lines
1.7.9.5 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 2 lines
Ville Syrjälä Intel OTC
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 168 lines · Ville Syrjälä
Quoted 167 lines · Abdiel Janulgue
Use on-chip hw-binding table generator to generate binding tables when the test emits
SURFACE_STATESpackets. The hw generates these binding table packets internally so we don't have to allocate space on the batchbuffer.Signed-off-by Abdiel Janulgue <[email protected]>---
lib/gen8_render.h | 13 +++++++ lib/intel_batchbuffer.c | 12 ++++++ lib/intel_batchbuffer.h | 3 ++ lib/intel_reg.h | 3 ++ lib/rendercopy_gen8.c | 97++++++++++++++++++++++++++++++++++++++++++++--- 5 files changed, 123 insertions(+), 5 deletions(-)
lib/gen8_render.h +2
-0@@ -83,6 +83,19 @@/* Random shifts */ #define GEN8_3DSTATE_PS_MAX_THREADS_SHIFT 23 +/* GEN8+ resource streamer */ +#define GEN8_3DSTATE_BINDING_TABLE_POOL_ALLOC (0x7919 << 16)/* GEN8 */ +# define
BINDING_TABLE_POOL_ENABLE0x0800
+#defineGEN8_3DSTATE_BINDING_TABLE_EDIT_VS(0x7843<< 16) /*
GEN8 */ +#defineGEN8_3DSTATE_BINDING_TABLE_EDIT_GS(0x7844<<
16) /* GEN8 */ +#defineGEN8_3DSTATE_BINDING_TABLE_EDIT_HS(0x7845<< 16) /* GEN8 */ +#defineGEN8_3DSTATE_BINDING_TABLE_EDIT_DS(0x7846 << 16) /* GEN8 */ +#define
GEN8_3DSTATE_BINDING_TABLE_EDIT_PS(0x7847<< 16) /* GEN8 */ +lib/gen8_render.h +4
-0+/* Toggling RS needs PIPE_CONTROL SC invalidate */ +#define _3DSTATE_PIPE_CONTROL ((0x3 << 29) | (3 << 27) | (2 << 24)) +#define PIPE_CONTROL_STATE_CACHE_INVALIDATE (1 << 2) +/* Shamelessly ripped from mesa */ struct gen8_surface_state {lib/intel_batchbuffer.c +8
-0@@ -81,6 +81,14 @@ intel_batchbuffer_reset(struct intel_batchbuffer *batch)memset (batch->buffer,0 ,sizeof (batch->buffer)); batch->ptr = batch->buffer; + +if (batch->hw_bt_pool_bo !=NULL ) { +drm_intel_bo_unreference (batch->hw_bt_pool_bo); + batch->hw_bt_pool_bo =NULL ; + } + + batch->hw_bt_pool_bo =drm_intel_bo_alloc (batch->bufmgr,"hw_bt" , +131072 ,4096 );}
/**
lib/intel_batchbuffer.c +4
-0@@ -116,6 +124,10 @@ intel_batchbuffer_free(struct intel_batchbuffer *batch)> {drm_intel_bo_unreference (batch->bo); batch->bo =NULL ; + +drm_intel_bo_unreference (batch->hw_bt_pool_bo); + batch->hw_bt_pool_bo =NULL ; +lib/intel_batchbuffer.c +0
-0free (batch); }lib/intel_batchbuffer.h +3
-0@@ -18,6 +18,9 @@ struct intel_batchbuffer {uint8_t buffer[BATCH_SZ ];uint8_t *ptr;uint8_t *state; + +drm_intel_bo *hw_bt_pool_bo; +int use_resource_streamer;};
struct
intel_batchbuffer*intel_batchbuffer_alloc(drm_intel_bufmgr *bufmgr,lib/intel_reg.h +0
-0@@ -2560,6 +2560,9 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THESOFTWARE.>
#defineMI_BATCH_NON_SECURE(1)
#defineMI_BATCH_NON_SECURE_I965(1 << 8)lib/intel_reg.h +3
-0+/* Resource Streamer */ +#define MI_RS_CONTROL (0x6 << 23) +#define MAX_DISPLAY_PIPES 2 typedef enum {lib/rendercopy_gen8.c +3
-1@@ -169,12 +169,14 @@ static void gen6_render_flush(struct intel_batchbuffer *batch, drm_intel_context *context, uint32_t batch_end) { - int ret; + int ret = 0; +uint32_t flags =0 ; ret =drm_intel_bo_subdata (batch->bo,0 ,4096 , batch->buffer); + flags = batch->use_resource_streamer ?I915_EXEC_RESOURCE_STREAMER :0 ;lib/rendercopy_gen8.c +1
-1if (ret ==0 ) ret =drm_intel_gem_bo_context_exec (batch->bo, context, - batch_end,0 ); + batch_end, flags |I915_EXEC_RENDER );assert (ret ==0 ); }@@ -228,18 +230,83 @@ gen8_bind_buf(struct intel_batchbuffer *batch,struct
igt_buf*buf,>lib/rendercopy_gen8.c +11
-0return offset; } +static void +gen8_hw_binding_table (struct intel_batchbuffer *batch,bool enable) +{ +if (!enable) { +OUT_BATCH (MI_RS_CONTROL |0x0 ); + +OUT_BATCH (GEN8_3DSTATE_BINDING_TABLE_POOL_ALLOC | (4 -2 )); +/* binding table pool base address */ +OUT_BATCH (0 ); +OUT_BATCH (0 ); +/* Upper bound */
Quoted 8 lines
lib/rendercopy_gen8.c
Quoted 2 lines
lib/rendercopy_gen8.c +2
-0+OUT_BATCH (0 ); +WaCsStallBeforeStateCacheInvalidate says you need to emit another
PIPE_CONTROLhere with CS stall bit set. And thanks to the other restrictions thatPIPE_CONTORLneeds one of the other magic bits also set. Stall at scoreboard perhaps? I think that's what we set in the kernel usually.
It used to have the CS stall bit. But it seems to work without it. In any case, you might be right. It's a good idea to put in back in.
Quoted 54 lines
Quoted 6 lines
lib/rendercopy_gen8.c +5
-0+OUT_BATCH (_3DSTATE_PIPE_CONTROL | (6 -2 )); +OUT_BATCH (PIPE_CONTROL_STATE_CACHE_INVALIDATE ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +OUT_BATCH (0 );lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 260
Quoted 10 lines
lib/rendercopy_gen8.c +10
-0+OUT_BATCH (0 ); + +return ; + } +OUT_BATCH (GEN8_3DSTATE_BINDING_TABLE_POOL_ALLOC | (4 -2 )); + +/* binding table pool base address */ +OUT_RELOC (batch->hw_bt_pool_bo,I915_GEM_DOMAIN_SAMPLER ,0 , +BINDING_TABLE_POOL_ENABLE ); +OUT_BATCH (0 );I suppose it doesn't matter much which domain we use here. But the hardware writes to the BT pool so maybe you should set write domain as well?
lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 264
Quoted 4 lines
lib/rendercopy_gen8.c +4
-0+ +/* Upper bound */ +OUT_RELOC (batch->hw_bt_pool_bo,I915_GEM_DOMAIN_SAMPLER ,0 , + batch->hw_bt_pool_bo->size);AFAICS on bdw this should be just the bt pool size and not a reloc.
Quoted 1 line
+
Again need another CS stall
PIPE_CONTROL.Quoted 22 lines
lib/rendercopy_gen8.c +15
-0+OUT_BATCH (_3DSTATE_PIPE_CONTROL | (6 -2 )); +OUT_BATCH (PIPE_CONTROL_STATE_CACHE_INVALIDATE ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); +} + +static uint32_t +gen8_rs_bind_surfaces (struct intel_batchbuffer *batch, +struct igt_buf *src, +struct igt_buf *dst, +uint32_t *surf0,uint32_t *surf1) +{ + *surf0 = gen8_bind_buf(batch, dst, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM,1);
lib/rendercopy_gen8.c +1
-0+ *surf1 = gen8_bind_buf(batch, src, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM,0);
lib/rendercopy_gen8.c +4
-0+ +return 0 ; +} +
Quoted 8 lines
lib/rendercopy_gen8.c gen8_bind_buf(struct intel_batchbuffer *batch, struct igt_buf *buf, · line 291
Quoted 6 lines
lib/rendercopy_gen8.c +6
-0+static void +gen8_rs_edit_surfaces (struct intel_batchbuffer *batch, +uint32_t surf0,uint32_t surf1) +{ +OUT_BATCH (GEN8_3DSTATE_BINDING_TABLE_EDIT_PS | (4 -2 )); +OUT_BATCH (0x3 );So I take it there's no need to clear the entries first?
I don't think it's necessary on this case. The clear is only neded if the binding table pool is re-used from the previous batch - which might contain stale entries. At least it causes hungs in Mesa. In this test, we are allocating the pool same time as the batch is allocated.
Quoted 114 lines
Quoted 114 lines
lib/rendercopy_gen8.c +11
-2+ { +OUT_BATCH (0 <<16 | surf0 >>6 ); +OUT_BATCH (1 <<16 | surf1 >>6 ); + } +} +static uint32_t gen8_bind_surfaces (struct intel_batchbuffer *batch,struct igt_buf *src,struct igt_buf *dst) { -uint32_t *binding_table, offset; +uint32_t offset =0 ; +uint32_t *binding_table; binding_table =batch_alloc (batch,8 ,32 ); offset =batch_offset (batch, binding_table);annotation_add_state (&aub_annotations,AUB_TRACE_BINDING_TABLE , offset,8 ); - binding_table[0 ] =gen8_bind_buf (batch, dst,GEN6_SURFACEFORMAT_B8G8R8A8_UNORM ,1 ); binding_table[1] =@@ -512,6 +579,9 @@ gen7_emit_push_constants(struct intel_batchbuffer *batch) {> static voidgen8_emit_state_base_address (struct intel_batchbuffer *batch) { +if (batch->use_resource_streamer) +OUT_BATCH (MI_RS_CONTROL |0x0 ); +OUT_BATCH (GEN6_STATE_BASE_ADDRESS | (16 -2 ));lib/rendercopy_gen8.c +3
-0/* general */ @@ -546,6 +616,9 @@ gen8_emit_state_base_address(struct intel_batchbuffer *batch) {>OUT_BATCH (0xfffff000 |1 );/* intruction buffer size */ OUT_BATCH (1 <<12 |1 ); + +if (batch->use_resource_streamer) +OUT_BATCH (MI_RS_CONTROL |0x1 );}
static void
lib/rendercopy_gen8.c +1
-0@@ -918,6 +991,7 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,>uint32_t scissor_state;uint32_t vertex_buffer;uint32_t batch_end; +uint32_t surf0 =0 , surf1 =1 ;lib/rendercopy_gen8.c +5
-1intel_batchbuffer_flush_with_context (batch, context);@@ -927,7 +1001,11 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,>annotation_init (&aub_annotations); - ps_binding_table =gen8_bind_surfaces (batch, src, dst); +if (batch->use_resource_streamer) + ps_binding_table =gen8_rs_bind_surfaces (batch, src, dst, + &surf0, &surf1); +else + ps_binding_table =gen8_bind_surfaces (batch, src, dst); ps_sampler_state =gen8_create_sampler (batch); ps_kernel_off =gen8_fill_ps (batch, ps_kernel,sizeof (ps_kernel));lib/rendercopy_gen8.c +3
-0vertex_buffer = gen7_fill_vertex_buffer_data(batch, src,@@ -955,6 +1033,9 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,>gen8_emit_state_base_address (batch); +if (batch->use_resource_streamer) +gen8_hw_binding_table (batch, true); +OUT_BATCH (GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC );OUT_BATCH (viewport.cc_state);lib/rendercopy_gen8.c +3
-0OUT_BATCH (GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP );@@ -978,6 +1059,9 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,>gen8_emit_sf (batch); +if (batch->use_resource_streamer) +gen8_rs_edit_surfaces (batch, surf0, surf1); +OUT_BATCH (GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS );OUT_BATCH (ps_binding_table);@@ -1001,6 +1085,9 @@ void gen8_render_copyfunc(struct intel_batchbuffer *batch,>gen8_emit_vf_topology (batch);gen8_emit_primitive (batch, vertex_buffer);
Quoted 7 lines
Quoted 7 lines
lib/rendercopy_gen8.c +3
-0+if (batch->use_resource_streamer) +gen8_hw_binding_table (batch, false); +OUT_BATCH (MI_BATCH_BUFFER_END );
batch_end=batch_align(batch, 8);
Thanks for the review. I'll update v2 soon to address the comments
-abdiel
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
In reply to Abdiel JanulgueI quickly cobbled together a hsw version of this and gave it a whirl on one machine. Seems to work just fine here, and no lockups when switching between hw and sw binding tables. Did you get the lockups on hsw even with rendercopy?
Here's my hsw version:
Quoted 1 line
From
17eeb8021815e2c18d6ba9b2185a37904296c2d9Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ville=20Syrj=C3=A4l=C3=A4?= <[email protected]>
Date: Wed, 7 May 2014 12:33:01 +0300
Subject: [PATCH] rendercopy: use resource streamer on hsw
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
--- lib/gen7_render.h | 16 +++++++- lib/rendercopy_gen7.c | 103 +++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 115 insertions(+), 4 deletions(-)
lib/gen7_render.h +14 -2
@@ -155,8 +155,11 @@#define GEN7_PIPE_CONTROL_IS_FLUSH (1 << 11)#define GEN7_PIPE_CONTROL_TC_FLUSH (1 << 10)#define GEN7_PIPE_CONTROL_NOTIFY_ENABLE (1 << 8) -#define GEN7_PIPE_CONTROL_GLOBAL_GTT (1 << 2) -#define GEN7_PIPE_CONTROL_LOCAL_PGTT (0 << 2) +#define GEN7_PIPE_CONTROL_FLUSH (1 << 7) +#define GEN7_PIPE_CONTROL_DC_FLUSH (1 << 5) +#define GEN7_PIPE_CONTROL_VF_INVALIDATE (1 << 4) +#define GEN7_PIPE_CONTROL_CC_INVALIDATE (1 << 2) +#define GEN7_PIPE_CONTROL_SC_INVALIDATE (1 << 2)#define GEN7_PIPE_CONTROL_STALL_AT_SCOREBOARD (1 << 1)#define GEN7_PIPE_CONTROL_DEPTH_CACHE_FLUSH (1 << 0)
@@ -1361,4 +1364,13 @@ typedef enum {EXTEND_COUNT } sampler_extend_t; +/* HSW+ resource streamer */ +#define HSW_3DSTATE_BINDING_TABLE_POOL_ALLOC GEN7_3D(3, 1, 0x19) +# define BINDING_TABLE_POOL_ENABLE (1 << 11) +#define HSW_3DSTATE_BINDING_TABLE_EDIT_VS GEN7_3D(3, 0, 0x43) +#define HSW_3DSTATE_BINDING_TABLE_EDIT_GS GEN7_3D(3, 0, 0x44) +#define HSW_3DSTATE_BINDING_TABLE_EDIT_HS GEN7_3D(3, 0, 0x45) +#define HSW_3DSTATE_BINDING_TABLE_EDIT_DS GEN7_3D(3, 0, 0x46) +#define HSW_3DSTATE_BINDING_TABLE_EDIT_PS GEN7_3D(3, 0, 0x47) +#endif
lib/rendercopy_gen7.c +101 -2
156 lines not on this page · raw message
Signature · 5 lines
1.8.3.2 -- Ville Syrjälä Intel OTC
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 4 lines · Ville Syrjälä
I quickly cobbled together a hsw version of this and gave it a whirl on one machine. Seems to work just fine here, and no lockups when switching between hw and sw binding tables. Did you get the lockups on hsw even with rendercopy?
This intrigues me that we should do a context continuation rendercpy (with the state setup split across multiple batches) and lots of context switching.
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
In reply to Ville SyrjäläQuoted 8 lines · Ville Syrjälä
I quickly cobbled together a hsw version of this and gave it a whirl on one machine. Seems to work just fine here, and no lockups when switching between hw and sw binding tables. Did you get the lockups on hsw even with rendercopy?
Here's my hsw version:
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 235
Quoted 8 lines
lib/rendercopy_gen7.c +8
-0+static void +gen7_hw_binding_table (struct intel_batchbuffer *batch,bool enable) +{ +if (!enable) { +OUT_BATCH (MI_RS_CONTROL |0x0 ); + +OUT_BATCH (HSW_3DSTATE_BINDING_TABLE_POOL_ALLOC | (3 -2 )); +/* binding table pool base address */
This bit I missed and the source of my troubles for the past few months.
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 236
Quoted 1 line
lib/rendercopy_gen7.c +1
-0+OUT_BATCH (3 <<5 );
Yep, I confirm toggling on HSW does work quite well now. I'll now update the patches to include HSW path on the kernel. I also take back my previous statement that RS is broken on HSW! :)
Thanks a lot Ville!
Quoted 15 lines
lib/rendercopy_gen7.c +5
-0+/* Upper bound */ +OUT_BATCH (0 ); + +OUT_BATCH (GEN7_PIPE_CONTROL | (4 -2 )); + OUT_BATCH(GEN7_PIPE_CONTROL_CS_STALL |
GEN7_PIPE_CONTROL_STALL_AT_SCOREBOARD); +OUT_BATCH(0);lib/rendercopy_gen7.c +9
-0+OUT_BATCH (0 ); + +OUT_BATCH (GEN7_PIPE_CONTROL | (4 -2 )); +OUT_BATCH (GEN7_PIPE_CONTROL_SC_INVALIDATE ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); + +return ; + }
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 20 lines · Abdiel Janulgue
Quoted 8 lines · Ville Syrjälä
I quickly cobbled together a hsw version of this and gave it a whirl on one machine. Seems to work just fine here, and no lockups when switching between hw and sw binding tables. Did you get the lockups on hsw even with rendercopy?
Here's my hsw version:
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 235
Quoted 8 lines
lib/rendercopy_gen7.c +8
-0+static void +gen7_hw_binding_table (struct intel_batchbuffer *batch,bool enable) +{ +if (!enable) { +OUT_BATCH (MI_RS_CONTROL |0x0 ); + +OUT_BATCH (HSW_3DSTATE_BINDING_TABLE_POOL_ALLOC | (3 -2 )); +/* binding table pool base address */ This bit I missed and the source of my troubles for the past few months.
Quoted 5 lines
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 236
Quoted 1 line
lib/rendercopy_gen7.c +1
-0+OUT_BATCH (3 <<5 );Yep, I confirm toggling on HSW does work quite well now. I'll now update the patches to include HSW path on the kernel. I also take back my previous statement that RS is broken on HSW! :)
Excellent.
I was wondering a bit if we need to make the kernel turn off the hw binding tables between batches, but since we now have per fd default contexts and 3DSTATE_BINDING_TABLE_POOL_ALLOC should be saved in the context, maybe we don't actually need to do that. Although it seems like that would cause problems when we switch to the global default context since we use the restore_inhibit flag there. So maybe we need to special case the default context here and force the hw binding tables off when switching to it.
Would be nice to have some igts for this to test how things work when you combine rs vs. no-rs in different ways using single context, multiple contexts, and multiple fds.
Quoted 2 lines
Thanks a lot Ville!
I'm glad I could help.
Quoted 17 lines
Quoted 17 lines
lib/rendercopy_gen7.c +5
-0+/* Upper bound */ +OUT_BATCH (0 ); + +OUT_BATCH (GEN7_PIPE_CONTROL | (4 -2 )); + OUT_BATCH(GEN7_PIPE_CONTROL_CS_STALL |
GEN7_PIPE_CONTROL_STALL_AT_SCOREBOARD); +OUT_BATCH(0);lib/rendercopy_gen7.c +9
-0+OUT_BATCH (0 ); + +OUT_BATCH (GEN7_PIPE_CONTROL | (4 -2 )); +OUT_BATCH (GEN7_PIPE_CONTROL_SC_INVALIDATE ); +OUT_BATCH (0 ); +OUT_BATCH (0 ); + +return ; + }
Signature · 2 lines
Ville Syrjälä Intel OTC
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 29 lines · Ville Syrjälä
Quoted 20 lines · Abdiel Janulgue
Quoted 8 lines · Ville Syrjälä
I quickly cobbled together a hsw version of this and gave it a whirl on one machine. Seems to work just fine here, and no lockups when switching between hw and sw binding tables. Did you get the lockups on hsw even with rendercopy?
Here's my hsw version:
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 235
Quoted 8 lines
lib/rendercopy_gen7.c +8
-0+static void +gen7_hw_binding_table (struct intel_batchbuffer *batch,bool enable) +{ +if (!enable) { +OUT_BATCH (MI_RS_CONTROL |0x0 ); + +OUT_BATCH (HSW_3DSTATE_BINDING_TABLE_POOL_ALLOC | (3 -2 )); +/* binding table pool base address */ This bit I missed and the source of my troubles for the past few months.
Quoted 5 lines
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 236
Quoted 1 line
lib/rendercopy_gen7.c +1
-0+OUT_BATCH (3 <<5 );Yep, I confirm toggling on HSW does work quite well now. I'll now update the patches to include HSW path on the kernel. I also take back my previous statement that RS is broken on HSW! :)
Excellent.
Quoted 8 lines
I was wondering a bit if we need to make the kernel turn off the hw binding tables between batches, but since we now have per fd default contexts and 3DSTATE_BINDING_TABLE_POOL_ALLOC should be saved in the context, maybe we don't actually need to do that. Although it seems like that would cause problems when we switch to the global default context since we use the
restore_inhibitflag there. So maybe we need to special case the default context here and force the hw binding tables off when switching to it.
Ah actually we still inhibit restore even with the per-file default contexts. So I guess it's going to be a problemn whenever we switch to any default context and the hw binding tables were left enabled.
So I guess there are two options: either drop the restore_inhibit flag for default contexts (I kind of like this idea but maybe other people hate it), or have the kernel turn off the hw binding tables when switching to any default context. The latter would also imply that if user space wants to use the resource streamer with the default context, it has to turn the binding tables on at the start of every batch since it can't know whether the kernel turned them off.
Signature · 2 lines
Ville Syrjälä Intel OTC
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 43 lines · Ville Syrjälä
Quoted 29 lines · Ville Syrjälä
Quoted 20 lines · Abdiel Janulgue
Quoted 8 lines · Ville Syrjälä
I quickly cobbled together a hsw version of this and gave it a whirl on one machine. Seems to work just fine here, and no lockups when switching between hw and sw binding tables. Did you get the lockups on hsw even with rendercopy?
Here's my hsw version:
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 235
Quoted 8 lines
lib/rendercopy_gen7.c +8
-0+static void +gen7_hw_binding_table (struct intel_batchbuffer *batch,bool enable) +{ +if (!enable) { +OUT_BATCH (MI_RS_CONTROL |0x0 ); + +OUT_BATCH (HSW_3DSTATE_BINDING_TABLE_POOL_ALLOC | (3 -2 )); +/* binding table pool base address */ This bit I missed and the source of my troubles for the past few months.
Quoted 5 lines
lib/rendercopy_gen7.c static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, · line 236
Quoted 1 line
lib/rendercopy_gen7.c +1
-0+OUT_BATCH (3 <<5 );Yep, I confirm toggling on HSW does work quite well now. I'll now update the patches to include HSW path on the kernel. I also take back my previous statement that RS is broken on HSW! :)
Excellent.
Quoted 8 lines
I was wondering a bit if we need to make the kernel turn off the hw binding tables between batches, but since we now have per fd default contexts and 3DSTATE_BINDING_TABLE_POOL_ALLOC should be saved in the context, maybe we don't actually need to do that. Although it seems like that would cause problems when we switch to the global default context since we use the
restore_inhibitflag there. So maybe we need to special case the default context here and force the hw binding tables off when switching to it.Ah actually we still inhibit restore even with the per-file default contexts. So I guess it's going to be a problemn whenever we switch to any default context and the hw binding tables were left enabled.
Quoted 7 lines
So I guess there are two options: either drop the
restore_inhibitflag for default contexts (I kind of like this idea but maybe other people hate it), or have the kernel turn off the hw binding tables when switching to any default context. The latter would also imply that if user space wants to use the resource streamer with the default context, it has to turn the binding tables on at the start of every batch since it can't know whether the kernel turned them off.
I think as soon as we have the golden context stuff from Mika we could drop our usage of restore_inhibit. We only need that to avoid the hw getting angry if the context state is illegal afaik. -Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 3 lines · Daniel Vetter
I think as soon as we have the golden context stuff from Mika we could drop our usage of
restore_inhibit. We only need that to avoid the hw getting angry if the context state is illegal afaik.
Apart from the contexts being over 100x larger than the state required to switch between clients, and that the current code regressed to always update the context between every batch.
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 8 lines · Chris Wilson
Quoted 3 lines · Daniel Vetter
I think as soon as we have the golden context stuff from Mika we could drop our usage of
restore_inhibit. We only need that to avoid the hw getting angry if the context state is illegal afaik.Apart from the contexts being over 100x larger than the state required to switch between clients, and that the current code regressed to always update the context between every batch.
We still have the from == to early return in do_switch. That doesn't do what it should any more?
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 6 lines · Daniel Vetter
Quoted 5 lines · Chris Wilson
Quoted 3 lines · Daniel Vetter
I think as soon as we have the golden context stuff from Mika we could drop our usage of
restore_inhibit. We only need that to avoid the hw getting angry if the context state is illegal afaik.
Quoted 6 lines
Quoted 3 lines
Apart from the contexts being over 100x larger than the state required to switch between clients, and that the current code regressed to always update the context between every batch.
We still have the from == to early return in
do_switch. That doesn't do what it should any more?
No.
commit 0009e46cd54324c4af20b0b52b89973b1b914167
Author: Ben Widawsky <[email protected]> Date: Fri Dec 6 14:11:02 2013 -0800
drm/i915: Track which ring a context ran on
Previously we dropped the association of a context to a ring. It is however very important to know which ring a context ran on (we could have reused the other member, but I was nitpicky).
This is very important when we switch address spaces, which unlike context objects, do change per ring.
As an example, if we have:
RCS BCS
ctx A
ctx A
ctx B
ctx B
Without tracking the last ring B ran on, we wouldn't know to switch the address space on BCS in the last row.
As a result, we no longer need to track which ring a context "belongs" to, as it never really made much sense anyway.
was just baloney.
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 1/2] rendercopy/bdw: Enable hw-generated binding tables
Quoted 40 lines · Chris Wilson
Quoted 6 lines · Daniel Vetter
Quoted 5 lines · Chris Wilson
Quoted 3 lines · Daniel Vetter
I think as soon as we have the golden context stuff from Mika we could drop our usage of
restore_inhibit. We only need that to avoid the hw getting angry if the context state is illegal afaik.Quoted 6 lines
Quoted 3 lines
Apart from the contexts being over 100x larger than the state required to switch between clients, and that the current code regressed to always update the context between every batch.
We still have the from == to early return in
do_switch. That doesn't do what it should any more?No.
commit
0009e46cd54324c4af20b0b52b89973b1b914167Author: Ben Widawsky <[email protected]> Date: Fri Dec 6 14:11:02 2013 -0800drm/i915: Track which ring a context ran on
Previously we dropped the association of a context to a ring. It is however very important to know which ring a context ran on (we could have reused the other member, but I was nitpicky).
This is very important when we switch address spaces, which unlike context objects, do change per ring.
As an example, if we have:
RCS BCS
ctx A
ctx A
ctx B
ctx BWithout tracking the last ring B ran on, we wouldn't know to switch the address space on BCS in the last row.
Quoted 7 lines
As a result, we no longer need to track which ring a context "belongs" to, as it never really made much sense anyway.
Signed-off-by Ben Widawsky <[email protected]>Signed-off-by Daniel Vetter <[email protected]>was just baloney.
Oh, totally forgotten about that fun. I'll add it as a new subtask to the big full ppgtt jira :(
Thanks digging this out again.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 2/2 tests/gem_render_copy: Add option to test resource streamer
In reply to Abdiel JanulgueAdd option in basic test for the render_copy to test and toggle hw-generated binding tables feature.
--- tests/gem_render_copy.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
tests/gem_render_copy.c +7 -1
@@ -131,14 +131,19 @@ int main(int argc, char **argv)int opt;int opt_dump_png = false;int opt_dump_aub =igt_aub_dump_enabled (); +int opt_use_rs = false;igt_simple_init (); -while ((opt =getopt (argc, argv,"d" )) !=-1 ) { +while ((opt =getopt (argc, argv,"d:r" )) !=-1 ) {switch (opt) {case 'd' : opt_dump_png = true;break ; +case 'r' : +printf ("Use resource streamer\n" ); + opt_use_rs = true; +break ;default :break ; }
@@ -156,6 +161,7 @@ int main(int argc, char **argv) "no render-copy function\n"); batch =intel_batchbuffer_alloc (data.bufmgr, data.devid); + batch->use_resource_streamer = opt_use_rs;igt_assert (batch); }
Signature · 1 line
1.7.9.5
Quoted 8 lines · Abdiel Janulgue
This patchseries tests a resource streamer feature called hw-generated binding tables, which can be enabled from userspace. It is essentially a newer format that uses the hw to generate and submit binding tables to the GPU. It uses the basic rendercopy test to toggle the feature on and off and to verify if the resulting data is still correct.
Abdiel Janulgue (2):
rendercopy/bdw: Enable hw-generated binding tables tests/gem_render_copy: Add option to test resource streamer
Since you add a new execbuf flag you also need to update the gem_exec_params test. I think we should have tests for both !bdw and !rcs to make sure the checking is complete.
-Daniel
Quoted 15 lines
lib/gen8_render.h | 13 +++++++ lib/intel_batchbuffer.c | 12 ++++++ lib/intel_batchbuffer.h | 3 ++ lib/intel_reg.h | 3 ++ lib/rendercopy_gen8.c | 97 ++++++++++++++++++++++++++++++++++++++++++++--- tests/gem_render_copy.c | 8 +++- 6 files changed, 130 insertions(+), 6 deletions(-)_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch