diff options
-rw-r--r-- | src/freedreno/registers/adreno_pm4.xml | 10 | ||||
-rw-r--r-- | src/freedreno/vulkan/tu_cmd_buffer.c | 33 | ||||
-rw-r--r-- | src/gallium/drivers/freedreno/a6xx/fd6_emit.c | 53 | ||||
-rw-r--r-- | src/gallium/drivers/freedreno/a6xx/fd6_emit.h | 5 |
4 files changed, 53 insertions, 48 deletions
diff --git a/src/freedreno/registers/adreno_pm4.xml b/src/freedreno/registers/adreno_pm4.xml index f94e8e4e10b..3a7865b489d 100644 --- a/src/freedreno/registers/adreno_pm4.xml +++ b/src/freedreno/registers/adreno_pm4.xml @@ -762,13 +762,9 @@ opcode: CP_LOAD_STATE4 (30) (4 dwords) <bitfield name="DISABLE" pos="17" type="boolean"/> <bitfield name="DISABLE_ALL_GROUPS" pos="18" type="boolean"/> <bitfield name="LOAD_IMMED" pos="19" type="boolean"/> - <!-- - I think this is a bitmask of states that this group applies to - (ie. binning/bypass/gmem)? At least starting w/ a6xx blob - emits different VS state at the same time, with ENABLE_MASK=0x1 - for binning pass VS state, and ENABLE_MASK=0x6 for full VS. - --> - <bitfield name="ENABLE_MASK" low="20" high="23" variants="A6XX-"/> + <bitfield name="BINNING" pos="20" variants="A6XX-" type="boolean"/> + <bitfield name="GMEM" pos="21" variants="A6XX-" type="boolean"/> + <bitfield name="SYSMEM" pos="22" variants="A6XX-" type="boolean"/> <bitfield name="GROUP_ID" low="24" high="28" type="uint"/> </reg32> <reg32 offset="1" name="1"> diff --git a/src/freedreno/vulkan/tu_cmd_buffer.c b/src/freedreno/vulkan/tu_cmd_buffer.c index 3d5a9835970..49ea11acfaa 100644 --- a/src/freedreno/vulkan/tu_cmd_buffer.c +++ b/src/freedreno/vulkan/tu_cmd_buffer.c @@ -2462,6 +2462,9 @@ struct tu_draw_info uint64_t count_buffer_offset; }; +#define ENABLE_ALL (CP_SET_DRAW_STATE__0_BINNING | CP_SET_DRAW_STATE__0_GMEM | CP_SET_DRAW_STATE__0_SYSMEM) +#define ENABLE_DRAW (CP_SET_DRAW_STATE__0_GMEM | CP_SET_DRAW_STATE__0_SYSMEM) + enum tu_draw_state_group_id { TU_DRAW_STATE_PROGRAM, @@ -3155,49 +3158,49 @@ tu6_bind_draw_states(struct tu_cmd_buffer *cmd, draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_PROGRAM, - .enable_mask = 0x6, + .enable_mask = ENABLE_DRAW, .ib = pipeline->program.state_ib, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_PROGRAM_BINNING, - .enable_mask = 0x1, + .enable_mask = CP_SET_DRAW_STATE__0_BINNING, .ib = pipeline->program.binning_state_ib, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_VI, - .enable_mask = 0x6, + .enable_mask = ENABLE_DRAW, .ib = pipeline->vi.state_ib, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_VI_BINNING, - .enable_mask = 0x1, + .enable_mask = CP_SET_DRAW_STATE__0_BINNING, .ib = pipeline->vi.binning_state_ib, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_VP, - .enable_mask = 0x7, + .enable_mask = ENABLE_ALL, .ib = pipeline->vp.state_ib, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_RAST, - .enable_mask = 0x7, + .enable_mask = ENABLE_ALL, .ib = pipeline->rast.state_ib, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_DS, - .enable_mask = 0x7, + .enable_mask = ENABLE_ALL, .ib = pipeline->ds.state_ib, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_BLEND, - .enable_mask = 0x7, + .enable_mask = ENABLE_ALL, .ib = pipeline->blend.state_ib, }; } @@ -3207,13 +3210,13 @@ tu6_bind_draw_states(struct tu_cmd_buffer *cmd, draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_VS_CONST, - .enable_mask = 0x7, + .enable_mask = ENABLE_ALL, .ib = tu6_emit_consts(cmd, pipeline, descriptors_state, MESA_SHADER_VERTEX) }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_FS_CONST, - .enable_mask = 0x6, + .enable_mask = ENABLE_DRAW, .ib = tu6_emit_consts(cmd, pipeline, descriptors_state, MESA_SHADER_FRAGMENT) }; } @@ -3241,19 +3244,19 @@ tu6_bind_draw_states(struct tu_cmd_buffer *cmd, draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_VS_TEX, - .enable_mask = 0x7, + .enable_mask = ENABLE_ALL, .ib = vs_tex, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_FS_TEX, - .enable_mask = 0x6, + .enable_mask = ENABLE_DRAW, .ib = fs_tex, }; draw_state_groups[draw_state_group_count++] = (struct tu_draw_state_group) { .id = TU_DRAW_STATE_FS_IBO, - .enable_mask = 0x6, + .enable_mask = ENABLE_DRAW, .ib = fs_ibo, }; @@ -3267,10 +3270,10 @@ tu6_bind_draw_states(struct tu_cmd_buffer *cmd, tu_cs_emit_pkt7(cs, CP_SET_DRAW_STATE, 3 * draw_state_group_count); for (uint32_t i = 0; i < draw_state_group_count; i++) { const struct tu_draw_state_group *group = &draw_state_groups[i]; - + debug_assert((group->enable_mask & ~ENABLE_ALL) == 0); uint32_t cp_set_draw_state = CP_SET_DRAW_STATE__0_COUNT(group->ib.size / 4) | - CP_SET_DRAW_STATE__0_ENABLE_MASK(group->enable_mask) | + group->enable_mask | CP_SET_DRAW_STATE__0_GROUP_ID(group->id); uint64_t iova; if (group->ib.size) { diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c index e878c0c1370..0f9b68b1c4b 100644 --- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c +++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c @@ -574,11 +574,11 @@ fd6_emit_combined_textures(struct fd_ringbuffer *ring, struct fd6_emit *emit, enum fd6_state_id state_id; unsigned enable_mask; } s[PIPE_SHADER_TYPES] = { - [PIPE_SHADER_VERTEX] = { FD6_GROUP_VS_TEX, 0x7 }, - [PIPE_SHADER_TESS_CTRL] = { FD6_GROUP_HS_TEX, 0x7 }, - [PIPE_SHADER_TESS_EVAL] = { FD6_GROUP_DS_TEX, 0x7 }, - [PIPE_SHADER_GEOMETRY] = { FD6_GROUP_GS_TEX, 0x7 }, - [PIPE_SHADER_FRAGMENT] = { FD6_GROUP_FS_TEX, 0x6 }, + [PIPE_SHADER_VERTEX] = { FD6_GROUP_VS_TEX, ENABLE_ALL }, + [PIPE_SHADER_TESS_CTRL] = { FD6_GROUP_HS_TEX, ENABLE_ALL }, + [PIPE_SHADER_TESS_EVAL] = { FD6_GROUP_DS_TEX, ENABLE_ALL }, + [PIPE_SHADER_GEOMETRY] = { FD6_GROUP_GS_TEX, ENABLE_ALL }, + [PIPE_SHADER_FRAGMENT] = { FD6_GROUP_FS_TEX, ENABLE_DRAW }, }; debug_assert(s[type].state_id); @@ -889,7 +889,7 @@ fd6_emit_tess_const(struct fd6_emit *emit) emit_stage_tess_consts(constobj, emit->gs, gs_params, ARRAY_SIZE(gs_params)); } - fd6_emit_take_group(emit, constobj, FD6_GROUP_PRIMITIVE_PARAMS, 0x7); + fd6_emit_take_group(emit, constobj, FD6_GROUP_PRIMITIVE_PARAMS, ENABLE_ALL); } static void @@ -936,26 +936,27 @@ fd6_emit_state(struct fd_ringbuffer *ring, struct fd6_emit *emit) struct fd_ringbuffer *state; state = build_vbo_state(emit, emit->vs); - fd6_emit_take_group(emit, state, FD6_GROUP_VBO, 0x7); + fd6_emit_take_group(emit, state, FD6_GROUP_VBO, ENABLE_ALL); } if (dirty & FD_DIRTY_ZSA) { struct fd6_zsa_stateobj *zsa = fd6_zsa_stateobj(ctx->zsa); if (util_format_is_pure_integer(pipe_surface_format(pfb->cbufs[0]))) - fd6_emit_add_group(emit, zsa->stateobj_no_alpha, FD6_GROUP_ZSA, 0x7); + fd6_emit_add_group(emit, zsa->stateobj_no_alpha, FD6_GROUP_ZSA, ENABLE_ALL); else - fd6_emit_add_group(emit, zsa->stateobj, FD6_GROUP_ZSA, 0x7); + fd6_emit_add_group(emit, zsa->stateobj, FD6_GROUP_ZSA, ENABLE_ALL); } if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_BLEND | FD_DIRTY_PROG)) && pfb->zsbuf) { struct fd_ringbuffer *state; state = build_lrz(emit, false); - fd6_emit_take_group(emit, state, FD6_GROUP_LRZ, 0x6); + fd6_emit_take_group(emit, state, FD6_GROUP_LRZ, ENABLE_DRAW); state = build_lrz(emit, true); - fd6_emit_take_group(emit, state, FD6_GROUP_LRZ_BINNING, 0x1); + fd6_emit_take_group(emit, state, + FD6_GROUP_LRZ_BINNING, CP_SET_DRAW_STATE__0_BINNING); } if (dirty & FD_DIRTY_STENCIL_REF) { @@ -1008,10 +1009,10 @@ fd6_emit_state(struct fd_ringbuffer *ring, struct fd6_emit *emit) } if (dirty & FD_DIRTY_PROG) { - fd6_emit_add_group(emit, prog->config_stateobj, FD6_GROUP_PROG_CONFIG, 0x7); - fd6_emit_add_group(emit, prog->stateobj, FD6_GROUP_PROG, 0x6); + fd6_emit_add_group(emit, prog->config_stateobj, FD6_GROUP_PROG_CONFIG, ENABLE_ALL); + fd6_emit_add_group(emit, prog->stateobj, FD6_GROUP_PROG, ENABLE_DRAW); fd6_emit_add_group(emit, prog->binning_stateobj, - FD6_GROUP_PROG_BINNING, 0x1); + FD6_GROUP_PROG_BINNING, CP_SET_DRAW_STATE__0_BINNING); /* emit remaining non-stateobj program state, ie. what depends * on other emit state, so cannot be pre-baked. This could @@ -1025,7 +1026,7 @@ fd6_emit_state(struct fd_ringbuffer *ring, struct fd6_emit *emit) struct fd6_rasterizer_stateobj *rasterizer = fd6_rasterizer_stateobj(ctx->rasterizer); fd6_emit_add_group(emit, rasterizer->stateobj, - FD6_GROUP_RASTERIZER, 0x7); + FD6_GROUP_RASTERIZER, ENABLE_ALL); } /* Since the primitive restart state is not part of a tracked object, we @@ -1063,11 +1064,11 @@ fd6_emit_state(struct fd_ringbuffer *ring, struct fd6_emit *emit) OUT_RING(ring, A6XX_SP_FS_OUTPUT_CNTL1_MRT(nr)); } - fd6_emit_consts(emit, vs, PIPE_SHADER_VERTEX, FD6_GROUP_VS_CONST, 0x7); - fd6_emit_consts(emit, hs, PIPE_SHADER_TESS_CTRL, FD6_GROUP_HS_CONST, 0x7); - fd6_emit_consts(emit, ds, PIPE_SHADER_TESS_EVAL, FD6_GROUP_DS_CONST, 0x7); - fd6_emit_consts(emit, gs, PIPE_SHADER_GEOMETRY, FD6_GROUP_GS_CONST, 0x7); - fd6_emit_consts(emit, fs, PIPE_SHADER_FRAGMENT, FD6_GROUP_FS_CONST, 0x6); + fd6_emit_consts(emit, vs, PIPE_SHADER_VERTEX, FD6_GROUP_VS_CONST, ENABLE_ALL); + fd6_emit_consts(emit, hs, PIPE_SHADER_TESS_CTRL, FD6_GROUP_HS_CONST, ENABLE_ALL); + fd6_emit_consts(emit, ds, PIPE_SHADER_TESS_EVAL, FD6_GROUP_DS_CONST, ENABLE_ALL); + fd6_emit_consts(emit, gs, PIPE_SHADER_GEOMETRY, FD6_GROUP_GS_CONST, ENABLE_ALL); + fd6_emit_consts(emit, fs, PIPE_SHADER_FRAGMENT, FD6_GROUP_FS_CONST, ENABLE_DRAW); if (emit->key.key.has_gs || emit->key.key.tessellation) fd6_emit_tess_const(emit); @@ -1077,9 +1078,9 @@ fd6_emit_state(struct fd_ringbuffer *ring, struct fd6_emit *emit) struct fd_ringbuffer *dpconstobj = fd_submit_new_ringbuffer( ctx->batch->submit, IR3_DP_VS_COUNT * 4, FD_RINGBUFFER_STREAMING); ir3_emit_vs_driver_params(vs, dpconstobj, ctx, emit->info); - fd6_emit_take_group(emit, dpconstobj, FD6_GROUP_VS_DRIVER_PARAMS, 0x7); + fd6_emit_take_group(emit, dpconstobj, FD6_GROUP_VS_DRIVER_PARAMS, ENABLE_ALL); } else { - fd6_emit_take_group(emit, NULL, FD6_GROUP_VS_DRIVER_PARAMS, 0x7); + fd6_emit_take_group(emit, NULL, FD6_GROUP_VS_DRIVER_PARAMS, ENABLE_ALL); } struct ir3_stream_output_info *info = &fd6_last_shader(prog)->shader->stream_output; @@ -1193,7 +1194,7 @@ fd6_emit_state(struct fd_ringbuffer *ring, struct fd6_emit *emit) ir3_emit_image_dims(ctx->screen, fs, obj, &ctx->shaderimg[PIPE_SHADER_FRAGMENT]); - fd6_emit_take_group(emit, obj, FD6_GROUP_IBO, 0x6); + fd6_emit_take_group(emit, obj, FD6_GROUP_IBO, ENABLE_DRAW); fd_ringbuffer_del(state); } @@ -1204,16 +1205,18 @@ fd6_emit_state(struct fd_ringbuffer *ring, struct fd6_emit *emit) unsigned n = g->stateobj ? fd_ringbuffer_size(g->stateobj) / 4 : 0; + debug_assert((g->enable_mask & ~ENABLE_ALL) == 0); + if (n == 0) { OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(0) | CP_SET_DRAW_STATE__0_DISABLE | - CP_SET_DRAW_STATE__0_ENABLE_MASK(g->enable_mask) | + g->enable_mask | CP_SET_DRAW_STATE__0_GROUP_ID(g->group_id)); OUT_RING(ring, 0x00000000); OUT_RING(ring, 0x00000000); } else { OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(n) | - CP_SET_DRAW_STATE__0_ENABLE_MASK(g->enable_mask) | + g->enable_mask | CP_SET_DRAW_STATE__0_GROUP_ID(g->group_id)); OUT_RB(ring, g->stateobj); } diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.h b/src/gallium/drivers/freedreno/a6xx/fd6_emit.h index 81cd4d97f7f..7ca42adc372 100644 --- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.h +++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.h @@ -66,13 +66,16 @@ enum fd6_state_id { FD6_GROUP_ZSA, }; +#define ENABLE_ALL (CP_SET_DRAW_STATE__0_BINNING | CP_SET_DRAW_STATE__0_GMEM | CP_SET_DRAW_STATE__0_SYSMEM) +#define ENABLE_DRAW (CP_SET_DRAW_STATE__0_GMEM | CP_SET_DRAW_STATE__0_SYSMEM) + struct fd6_state_group { struct fd_ringbuffer *stateobj; enum fd6_state_id group_id; /* enable_mask controls which states the stateobj is evaluated in, * b0 is binning pass b1 and/or b2 is draw pass */ - uint8_t enable_mask; + uint32_t enable_mask; }; /* grouped together emit-state for prog/vertex/state emit: */ |