From f13de57edb1b0bea0090421e5f812f1041f36e12 Mon Sep 17 00:00:00 2001 From: Tomeu Vizoso Date: Tue, 17 Jul 2018 13:13:21 +0200 Subject: [PATCH] virgl: use hw-atomics instead of in-ssbo ones Emulating atomics on top of ssbos can lead to too small max SSBO count, so let's use the hw-atomics mechanism to expose atomic buffers instead. Signed-off-by: Erik Faye-Lund Reviewed-by: Gurchetan Singh --- src/gallium/drivers/virgl/virgl_context.c | 37 ++++++++++++++++++++++++++++++ src/gallium/drivers/virgl/virgl_context.h | 2 ++ src/gallium/drivers/virgl/virgl_encode.c | 23 +++++++++++++++++++ src/gallium/drivers/virgl/virgl_encode.h | 3 +++ src/gallium/drivers/virgl/virgl_hw.h | 5 ++++ src/gallium/drivers/virgl/virgl_protocol.h | 9 ++++++++ src/gallium/drivers/virgl/virgl_screen.c | 12 ++++++++-- 7 files changed, 89 insertions(+), 2 deletions(-) diff --git a/src/gallium/drivers/virgl/virgl_context.c b/src/gallium/drivers/virgl/virgl_context.c index edc03f5..4511bf3 100644 --- a/src/gallium/drivers/virgl/virgl_context.c +++ b/src/gallium/drivers/virgl/virgl_context.c @@ -196,6 +196,19 @@ static void virgl_attach_res_shader_images(struct virgl_context *vctx, } } +static void virgl_attach_res_atomic_buffers(struct virgl_context *vctx) +{ + struct virgl_winsys *vws = virgl_screen(vctx->base.screen)->vws; + struct virgl_resource *res; + unsigned i; + for (i = 0; i < PIPE_MAX_HW_ATOMIC_BUFFERS; i++) { + res = virgl_resource(vctx->atomic_buffers[i]); + if (res) { + vws->emit_res(vws, vctx->cbuf, res->hw_res, FALSE); + } + } +} + /* * after flushing, the hw context still has a bunch of * resources bound, so we need to rebind those here. @@ -214,6 +227,7 @@ static void virgl_reemit_res(struct virgl_context *vctx) virgl_attach_res_shader_buffers(vctx, shader_type); virgl_attach_res_shader_images(vctx, shader_type); } + virgl_attach_res_atomic_buffers(vctx); virgl_attach_res_vertex_buffers(vctx); virgl_attach_res_so_targets(vctx); } @@ -952,6 +966,28 @@ static void virgl_blit(struct pipe_context *ctx, blit); } +static void virgl_set_hw_atomic_buffers(struct pipe_context *ctx, + unsigned start_slot, + unsigned count, + const struct pipe_shader_buffer *buffers) +{ + struct virgl_context *vctx = virgl_context(ctx); + + for (unsigned i = 0; i < count; i++) { + unsigned idx = start_slot + i; + + if (buffers) { + if (buffers[i].buffer) { + pipe_resource_reference(&vctx->atomic_buffers[idx], + buffers[i].buffer); + continue; + } + } + pipe_resource_reference(&vctx->atomic_buffers[idx], NULL); + } + virgl_encode_set_hw_atomic_buffers(vctx, start_slot, count, buffers); +} + static void virgl_set_shader_buffers(struct pipe_context *ctx, enum pipe_shader_type shader, unsigned start_slot, unsigned count, @@ -1209,6 +1245,7 @@ struct pipe_context *virgl_context_create(struct pipe_screen *pscreen, vctx->base.blit = virgl_blit; vctx->base.set_shader_buffers = virgl_set_shader_buffers; + vctx->base.set_hw_atomic_buffers = virgl_set_hw_atomic_buffers; vctx->base.set_shader_images = virgl_set_shader_images; vctx->base.memory_barrier = virgl_memory_barrier; diff --git a/src/gallium/drivers/virgl/virgl_context.h b/src/gallium/drivers/virgl/virgl_context.h index 38d1f45..20988ba 100644 --- a/src/gallium/drivers/virgl/virgl_context.h +++ b/src/gallium/drivers/virgl/virgl_context.h @@ -75,6 +75,8 @@ struct virgl_context { int num_draws; struct list_head to_flush_bufs; + struct pipe_resource *atomic_buffers[PIPE_MAX_HW_ATOMIC_BUFFERS]; + struct primconvert_context *primconvert; uint32_t hw_sub_ctx_id; }; diff --git a/src/gallium/drivers/virgl/virgl_encode.c b/src/gallium/drivers/virgl/virgl_encode.c index bcb14d8..29920b2 100644 --- a/src/gallium/drivers/virgl/virgl_encode.c +++ b/src/gallium/drivers/virgl/virgl_encode.c @@ -958,6 +958,29 @@ int virgl_encode_set_shader_buffers(struct virgl_context *ctx, return 0; } +int virgl_encode_set_hw_atomic_buffers(struct virgl_context *ctx, + unsigned start_slot, unsigned count, + const struct pipe_shader_buffer *buffers) +{ + int i; + virgl_encoder_write_cmd_dword(ctx, VIRGL_CMD0(VIRGL_CCMD_SET_ATOMIC_BUFFERS, 0, VIRGL_SET_ATOMIC_BUFFER_SIZE(count))); + + virgl_encoder_write_dword(ctx->cbuf, start_slot); + for (i = 0; i < count; i++) { + if (buffers) { + struct virgl_resource *res = virgl_resource(buffers[i].buffer); + virgl_encoder_write_dword(ctx->cbuf, buffers[i].buffer_offset); + virgl_encoder_write_dword(ctx->cbuf, buffers[i].buffer_size); + virgl_encoder_write_res(ctx, res); + } else { + virgl_encoder_write_dword(ctx->cbuf, 0); + virgl_encoder_write_dword(ctx->cbuf, 0); + virgl_encoder_write_dword(ctx->cbuf, 0); + } + } + return 0; +} + int virgl_encode_set_shader_images(struct virgl_context *ctx, enum pipe_shader_type shader, unsigned start_slot, unsigned count, diff --git a/src/gallium/drivers/virgl/virgl_encode.h b/src/gallium/drivers/virgl/virgl_encode.h index 999123f..40e62d4 100644 --- a/src/gallium/drivers/virgl/virgl_encode.h +++ b/src/gallium/drivers/virgl/virgl_encode.h @@ -267,6 +267,9 @@ int virgl_encode_set_shader_images(struct virgl_context *ctx, enum pipe_shader_type shader, unsigned start_slot, unsigned count, const struct pipe_image_view *images); +int virgl_encode_set_hw_atomic_buffers(struct virgl_context *ctx, + unsigned start_slot, unsigned count, + const struct pipe_shader_buffer *buffers); int virgl_encode_memory_barrier(struct virgl_context *ctx, unsigned flags); int virgl_encode_launch_grid(struct virgl_context *ctx, diff --git a/src/gallium/drivers/virgl/virgl_hw.h b/src/gallium/drivers/virgl/virgl_hw.h index 9a358dd..7736ceb 100644 --- a/src/gallium/drivers/virgl/virgl_hw.h +++ b/src/gallium/drivers/virgl/virgl_hw.h @@ -351,6 +351,11 @@ struct virgl_caps_v2 { uint32_t max_texture_2d_size; uint32_t max_texture_3d_size; uint32_t max_texture_cube_size; + uint32_t max_combined_shader_buffers; + uint32_t max_atomic_counters[6]; + uint32_t max_atomic_counter_buffers[6]; + uint32_t max_combined_atomic_counters; + uint32_t max_combined_atomic_counter_buffers; }; union virgl_caps { diff --git a/src/gallium/drivers/virgl/virgl_protocol.h b/src/gallium/drivers/virgl/virgl_protocol.h index dd1a4ee..8d99c5e 100644 --- a/src/gallium/drivers/virgl/virgl_protocol.h +++ b/src/gallium/drivers/virgl/virgl_protocol.h @@ -91,6 +91,7 @@ enum virgl_context_cmd { VIRGL_CCMD_LAUNCH_GRID, VIRGL_CCMD_SET_FRAMEBUFFER_STATE_NO_ATTACH, VIRGL_CCMD_TEXTURE_BARRIER, + VIRGL_CCMD_SET_ATOMIC_BUFFERS, }; /* @@ -544,4 +545,12 @@ enum virgl_context_cmd { #define VIRGL_TEXTURE_BARRIER_SIZE 1 #define VIRGL_TEXTURE_BARRIER_FLAGS 1 +/* hw atomics */ +#define VIRGL_SET_ATOMIC_BUFFER_ELEMENT_SIZE 3 +#define VIRGL_SET_ATOMIC_BUFFER_SIZE(x) (VIRGL_SET_ATOMIC_BUFFER_ELEMENT_SIZE * (x)) + 1 +#define VIRGL_SET_ATOMIC_BUFFER_START_SLOT 1 +#define VIRGL_SET_ATOMIC_BUFFER_OFFSET(x) ((x) * VIRGL_SET_ATOMIC_BUFFER_ELEMENT_SIZE + 2) +#define VIRGL_SET_ATOMIC_BUFFER_LENGTH(x) ((x) * VIRGL_SET_ATOMIC_BUFFER_ELEMENT_SIZE + 3) +#define VIRGL_SET_ATOMIC_BUFFER_RES_HANDLE(x) ((x) * VIRGL_SET_ATOMIC_BUFFER_ELEMENT_SIZE + 4) + #endif diff --git a/src/gallium/drivers/virgl/virgl_screen.c b/src/gallium/drivers/virgl/virgl_screen.c index bdc8a78..8a78091 100644 --- a/src/gallium/drivers/virgl/virgl_screen.c +++ b/src/gallium/drivers/virgl/virgl_screen.c @@ -248,6 +248,12 @@ virgl_get_param(struct pipe_screen *screen, enum pipe_cap param) return vscreen->caps.caps.v2.capability_bits & VIRGL_CAP_SHADER_CLOCK; case PIPE_CAP_TGSI_ARRAY_COMPONENTS: return vscreen->caps.caps.v2.capability_bits & VIRGL_CAP_TGSI_COMPONENTS; + case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS: + return vscreen->caps.caps.v2.max_combined_shader_buffers; + case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS: + return vscreen->caps.caps.v2.max_combined_atomic_counters; + case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS: + return vscreen->caps.caps.v2.max_combined_atomic_counter_buffers; case PIPE_CAP_TEXTURE_GATHER_SM5: case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT: case PIPE_CAP_FAKE_SW_MSAA: @@ -410,12 +416,14 @@ virgl_get_shader_param(struct pipe_screen *screen, return vscreen->caps.caps.v2.max_shader_image_other_stages; case PIPE_SHADER_CAP_SUPPORTED_IRS: return (1 << PIPE_SHADER_IR_TGSI); + case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS: + return vscreen->caps.caps.v2.max_atomic_counters[shader]; + case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS: + return vscreen->caps.caps.v2.max_atomic_counter_buffers[shader]; case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD: case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS: case PIPE_SHADER_CAP_INT64_ATOMICS: case PIPE_SHADER_CAP_FP16: - case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS: - case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS: return 0; case PIPE_SHADER_CAP_SCALAR_ISA: return 1; -- 2.7.4