X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fpanfrost%2Fpan_cmdstream.c;h=bc9158b03af6b026b79709eac2de53cef317bc8e;hb=cfef6de429d40785a306ef2a2db7092c4ce0388a;hp=02e50cb3c78b8662fc3a98ebab12886f184b9bc6;hpb=17c617cdb7f9710651b10f5b00669fc31c372c50;p=mesa.git diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.c b/src/gallium/drivers/panfrost/pan_cmdstream.c index 02e50cb3c78..bc9158b03af 100644 --- a/src/gallium/drivers/panfrost/pan_cmdstream.c +++ b/src/gallium/drivers/panfrost/pan_cmdstream.c @@ -58,13 +58,21 @@ panfrost_vt_emit_shared_memory(struct panfrost_context *ctx, struct panfrost_device *dev = pan_device(ctx->base.screen); struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx); - unsigned shift = panfrost_get_stack_shift(batch->stack_size); struct mali_shared_memory shared = { - .stack_shift = shift, - .scratchpad = panfrost_batch_get_scratchpad(batch, shift, dev->thread_tls_alloc, dev->core_count)->gpu, .shared_workgroup_count = ~0, }; - postfix->shared_memory = panfrost_pool_upload(&batch->pool, &shared, sizeof(shared)); + + if (batch->stack_size) { + struct panfrost_bo *stack = + panfrost_batch_get_scratchpad(batch, batch->stack_size, + dev->thread_tls_alloc, + dev->core_count); + + shared.stack_shift = panfrost_get_stack_shift(batch->stack_size); + shared.scratchpad = stack->gpu; + } + + postfix->shared_memory = panfrost_pool_upload_aligned(&batch->pool, &shared, sizeof(shared), 64); } static void @@ -209,9 +217,13 @@ panfrost_get_index_buffer_bounded(struct panfrost_context *ctx, } else { /* Otherwise, we need to upload to transient memory */ const uint8_t *ibuf8 = (const uint8_t *) info->index.user; - out = panfrost_pool_upload(&batch->pool, ibuf8 + offset, - info->count * - info->index_size); + struct panfrost_transfer T = + panfrost_pool_alloc_aligned(&batch->pool, + info->count * info->index_size, + info->index_size); + + memcpy(T.cpu, ibuf8 + offset, info->count * info->index_size); + out = T.gpu; } if (needs_indices) { @@ -528,7 +540,8 @@ panfrost_fs_required( static void panfrost_frag_meta_blend_update(struct panfrost_context *ctx, struct mali_shader_meta *fragmeta, - void *rts) + void *rts, + struct panfrost_blend_final *blend) { struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx); const struct panfrost_device *dev = pan_device(ctx->base.screen); @@ -543,15 +556,7 @@ panfrost_frag_meta_blend_update(struct panfrost_context *ctx, ctx->blend->base.alpha_to_coverage); /* Get blending setup */ - unsigned rt_count = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1); - - struct panfrost_blend_final blend[PIPE_MAX_COLOR_BUFS]; - unsigned shader_offset = 0; - struct panfrost_bo *shader_bo = NULL; - - for (unsigned c = 0; c < rt_count; ++c) - blend[c] = panfrost_get_blend_for_context(ctx, c, &shader_bo, - &shader_offset); + unsigned rt_count = ctx->pipe_framebuffer.nr_cbufs; /* Disable shader execution if we can */ if (dev->quirks & MIDGARD_SHADERLESS @@ -587,7 +592,7 @@ panfrost_frag_meta_blend_update(struct panfrost_context *ctx, fragmeta->blend.shader = 0; - for (signed rt = (rt_count - 1); rt >= 0; --rt) { + for (signed rt = ((signed) rt_count - 1); rt >= 0; --rt) { if (!blend[rt].is_shader) continue; @@ -629,16 +634,22 @@ panfrost_frag_meta_blend_update(struct panfrost_context *ctx, /* Additional blend descriptor tacked on for jobs using MFBD */ + struct bifrost_blend_rt *brts = rts; + struct midgard_blend_rt *mrts = rts; + + /* Disable blending for depth-only on Bifrost */ + + if (rt_count == 0 && dev->quirks & IS_BIFROST) + brts[0].unk2 = 0x3; + for (unsigned i = 0; i < rt_count; ++i) { unsigned flags = 0; - if (ctx->pipe_framebuffer.nr_cbufs > i && !blend[i].no_colour) { + if (!blend[i].no_colour) { flags = 0x200; batch->draws |= (PIPE_CLEAR_COLOR0 << i); - bool is_srgb = (ctx->pipe_framebuffer.nr_cbufs > i) && - (ctx->pipe_framebuffer.cbufs[i]) && - util_format_is_srgb(ctx->pipe_framebuffer.cbufs[i]->format); + bool is_srgb = util_format_is_srgb(ctx->pipe_framebuffer.cbufs[i]->format); SET_BIT(flags, MALI_BLEND_MRT_SHADER, blend[i].is_shader); SET_BIT(flags, MALI_BLEND_LOAD_TIB, !blend[i].no_blending); @@ -647,8 +658,6 @@ panfrost_frag_meta_blend_update(struct panfrost_context *ctx, } if (dev->quirks & IS_BIFROST) { - struct bifrost_blend_rt *brts = rts; - brts[i].flags = flags; if (blend[i].is_shader) { @@ -660,7 +669,7 @@ panfrost_frag_meta_blend_update(struct panfrost_context *ctx, (fs->bo->gpu & (0xffffffffull << 32))); brts[i].shader = blend[i].shader.gpu; brts[i].unk2 = 0x0; - } else if (ctx->pipe_framebuffer.nr_cbufs > i) { + } else { enum pipe_format format = ctx->pipe_framebuffer.cbufs[i]->format; const struct util_format_description *format_desc; format_desc = util_format_description(format); @@ -678,14 +687,9 @@ panfrost_frag_meta_blend_update(struct panfrost_context *ctx, * blending. */ brts[i].unk2 = blend[i].no_blending ? 0x19 : 0x1a; - brts[i].shader_type = fs->blend_types[i]; - } else { - /* Dummy attachment for depth-only */ - brts[i].unk2 = 0x3; brts[i].shader_type = fs->blend_types[i]; } } else { - struct midgard_blend_rt *mrts = rts; mrts[i].flags = flags; if (blend[i].is_shader) { @@ -701,7 +705,8 @@ panfrost_frag_meta_blend_update(struct panfrost_context *ctx, static void panfrost_frag_shader_meta_init(struct panfrost_context *ctx, struct mali_shader_meta *fragmeta, - void *rts) + void *rts, + struct panfrost_blend_final *blend) { const struct panfrost_device *dev = pan_device(ctx->base.screen); struct panfrost_shader_state *fs; @@ -760,7 +765,7 @@ panfrost_frag_shader_meta_init(struct panfrost_context *ctx, panfrost_frag_meta_rasterizer_update(ctx, fragmeta); panfrost_frag_meta_zsa_update(ctx, fragmeta); - panfrost_frag_meta_blend_update(ctx, fragmeta, rts); + panfrost_frag_meta_blend_update(ctx, fragmeta, rts, blend); } void @@ -808,9 +813,14 @@ panfrost_emit_shader_meta(struct panfrost_batch *batch, if (rt_size) rts = rzalloc_size(ctx, rt_size * rt_count); - panfrost_frag_shader_meta_init(ctx, &meta, rts); + struct panfrost_blend_final blend[PIPE_MAX_COLOR_BUFS]; + + for (unsigned c = 0; c < ctx->pipe_framebuffer.nr_cbufs; ++c) + blend[c] = panfrost_get_blend_for_context(ctx, c); - xfer = panfrost_pool_alloc(&batch->pool, desc_size); + panfrost_frag_shader_meta_init(ctx, &meta, rts, blend); + + xfer = panfrost_pool_alloc_aligned(&batch->pool, desc_size, sizeof(meta)); memcpy(xfer.cpu, &meta, sizeof(meta)); memcpy(xfer.cpu + sizeof(meta), rts, rt_size * rt_count); @@ -897,10 +907,10 @@ panfrost_map_constant_buffer_gpu(struct panfrost_batch *batch, * PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT */ return rsrc->bo->gpu + cb->buffer_offset; } else if (cb->user_buffer) { - return panfrost_pool_upload(&batch->pool, + return panfrost_pool_upload_aligned(&batch->pool, cb->user_buffer + cb->buffer_offset, - cb->buffer_size); + cb->buffer_size, 16); } else { unreachable("No constant buffer"); } @@ -1102,8 +1112,8 @@ panfrost_emit_const_buf(struct panfrost_batch *batch, size_t sys_size = sizeof(float) * 4 * ss->sysval_count; size_t uniform_size = has_uniforms ? (buf->cb[0].buffer_size) : 0; size_t size = sys_size + uniform_size; - struct panfrost_transfer transfer = panfrost_pool_alloc(&batch->pool, - size); + struct panfrost_transfer transfer = + panfrost_pool_alloc_aligned(&batch->pool, size, 16); /* Upload sysvals requested by the shader */ panfrost_upload_sysvals(batch, transfer.cpu, ss, stage); @@ -1121,7 +1131,10 @@ panfrost_emit_const_buf(struct panfrost_batch *batch, assert(ubo_count >= 1); size_t sz = MALI_UNIFORM_BUFFER_LENGTH * ubo_count; - struct panfrost_transfer ubos = panfrost_pool_alloc(&batch->pool, sz); + struct panfrost_transfer ubos = + panfrost_pool_alloc_aligned(&batch->pool, sz, + MALI_UNIFORM_BUFFER_LENGTH); + uint64_t *ubo_ptr = (uint64_t *) ubos.cpu; /* Upload uniforms as a UBO */ @@ -1188,8 +1201,8 @@ panfrost_emit_shared_memory(struct panfrost_batch *batch, .shared_shift = util_logbase2(single_size) + 1 }; - vtp->postfix.shared_memory = panfrost_pool_upload(&batch->pool, &shared, - sizeof(shared)); + vtp->postfix.shared_memory = panfrost_pool_upload_aligned(&batch->pool, &shared, + sizeof(shared), 64); } static mali_ptr @@ -1240,9 +1253,10 @@ panfrost_emit_texture_descriptors(struct panfrost_batch *batch, return; if (device->quirks & IS_BIFROST) { - struct panfrost_transfer T = panfrost_pool_alloc(&batch->pool, + struct panfrost_transfer T = panfrost_pool_alloc_aligned(&batch->pool, MALI_BIFROST_TEXTURE_LENGTH * - ctx->sampler_view_count[stage]); + ctx->sampler_view_count[stage], + MALI_BIFROST_TEXTURE_LENGTH); struct mali_bifrost_texture_packed *out = (struct mali_bifrost_texture_packed *) T.cpu; @@ -1278,10 +1292,11 @@ panfrost_emit_texture_descriptors(struct panfrost_batch *batch, trampolines[i] = panfrost_get_tex_desc(batch, stage, view); } - postfix->textures = panfrost_pool_upload(&batch->pool, + postfix->textures = panfrost_pool_upload_aligned(&batch->pool, trampolines, sizeof(uint64_t) * - ctx->sampler_view_count[stage]); + ctx->sampler_view_count[stage], + sizeof(uint64_t)); } } @@ -1299,7 +1314,7 @@ panfrost_emit_sampler_descriptors(struct panfrost_batch *batch, assert(MALI_BIFROST_SAMPLER_LENGTH == MALI_MIDGARD_SAMPLER_LENGTH); size_t sz = desc_size * ctx->sampler_count[stage]; - struct panfrost_transfer T = panfrost_pool_alloc(&batch->pool, sz); + struct panfrost_transfer T = panfrost_pool_alloc_aligned(&batch->pool, sz, desc_size); struct mali_midgard_sampler_packed *out = (struct mali_midgard_sampler_packed *) T.cpu; for (unsigned i = 0; i < ctx->sampler_count[stage]; ++i) @@ -1314,17 +1329,23 @@ panfrost_emit_vertex_data(struct panfrost_batch *batch, { struct panfrost_context *ctx = batch->ctx; struct panfrost_vertex_state *so = ctx->vertex; + struct panfrost_shader_state *vs = panfrost_get_shader_state(ctx, PIPE_SHADER_VERTEX); unsigned instance_shift = vertex_postfix->instance_shift; unsigned instance_odd = vertex_postfix->instance_odd; - /* Worst case: everything is NPOT */ + /* Worst case: everything is NPOT, which is only possible if instancing + * is enabled. Otherwise single record is gauranteed */ + bool could_npot = instance_shift || instance_odd; - struct panfrost_transfer S = panfrost_pool_alloc(&batch->pool, - MALI_ATTRIBUTE_LENGTH * PIPE_MAX_ATTRIBS * 2); + struct panfrost_transfer S = panfrost_pool_alloc_aligned(&batch->pool, + MALI_ATTRIBUTE_BUFFER_LENGTH * vs->attribute_count * + (could_npot ? 2 : 1), + MALI_ATTRIBUTE_BUFFER_LENGTH * 2); - struct panfrost_transfer T = panfrost_pool_alloc(&batch->pool, - MALI_ATTRIBUTE_LENGTH * (PAN_INSTANCE_ID + 1)); + struct panfrost_transfer T = panfrost_pool_alloc_aligned(&batch->pool, + MALI_ATTRIBUTE_LENGTH * vs->attribute_count, + MALI_ATTRIBUTE_LENGTH); struct mali_attribute_buffer_packed *bufs = (struct mali_attribute_buffer_packed *) S.cpu; @@ -1431,18 +1452,20 @@ panfrost_emit_vertex_data(struct panfrost_batch *batch, /* Add special gl_VertexID/gl_InstanceID buffers */ - panfrost_vertex_id(ctx->padded_count, &bufs[k], ctx->instance_count > 1); + if (unlikely(vs->attribute_count >= PAN_VERTEX_ID)) { + panfrost_vertex_id(ctx->padded_count, &bufs[k], ctx->instance_count > 1); - pan_pack(out + PAN_VERTEX_ID, ATTRIBUTE, cfg) { - cfg.buffer_index = k++; - cfg.format = so->formats[PAN_VERTEX_ID]; - } + pan_pack(out + PAN_VERTEX_ID, ATTRIBUTE, cfg) { + cfg.buffer_index = k++; + cfg.format = so->formats[PAN_VERTEX_ID]; + } - panfrost_instance_id(ctx->padded_count, &bufs[k], ctx->instance_count > 1); + panfrost_instance_id(ctx->padded_count, &bufs[k], ctx->instance_count > 1); - pan_pack(out + PAN_INSTANCE_ID, ATTRIBUTE, cfg) { - cfg.buffer_index = k++; - cfg.format = so->formats[PAN_INSTANCE_ID]; + pan_pack(out + PAN_INSTANCE_ID, ATTRIBUTE, cfg) { + cfg.buffer_index = k++; + cfg.format = so->formats[PAN_INSTANCE_ID]; + } } /* Attribute addresses require 64-byte alignment, so let: @@ -1492,7 +1515,7 @@ panfrost_emit_varyings(struct panfrost_batch *batch, unsigned stride, unsigned count) { unsigned size = stride * count; - mali_ptr ptr = panfrost_pool_alloc(&batch->invisible_pool, size).gpu; + mali_ptr ptr = panfrost_pool_alloc_aligned(&batch->invisible_pool, size, 64).gpu; pan_pack(slot, ATTRIBUTE_BUFFER, cfg) { cfg.stride = stride; @@ -1927,9 +1950,8 @@ panfrost_emit_varying_descriptor(struct panfrost_batch *batch, vs_size = MALI_ATTRIBUTE_LENGTH * vs->varying_count; fs_size = MALI_ATTRIBUTE_LENGTH * fs->varying_count; - struct panfrost_transfer trans = panfrost_pool_alloc(&batch->pool, - vs_size + - fs_size); + struct panfrost_transfer trans = panfrost_pool_alloc_aligned( + &batch->pool, vs_size + fs_size, MALI_ATTRIBUTE_LENGTH); struct pipe_stream_output_info *so = &vs->stream_output; unsigned present = pan_varying_present(vs, fs, dev->quirks); @@ -1975,8 +1997,9 @@ panfrost_emit_varying_descriptor(struct panfrost_batch *batch, } unsigned xfb_base = pan_xfb_base(present); - struct panfrost_transfer T = panfrost_pool_alloc(&batch->pool, - MALI_ATTRIBUTE_BUFFER_LENGTH * (xfb_base + ctx->streamout.num_targets)); + struct panfrost_transfer T = panfrost_pool_alloc_aligned(&batch->pool, + MALI_ATTRIBUTE_BUFFER_LENGTH * (xfb_base + ctx->streamout.num_targets), + MALI_ATTRIBUTE_BUFFER_LENGTH * 2); struct mali_attribute_buffer_packed *varyings = (struct mali_attribute_buffer_packed *) T.cpu; @@ -2139,5 +2162,5 @@ panfrost_emit_sample_locations(struct panfrost_batch *batch) 0, 0, }; - return panfrost_pool_upload(&batch->pool, locations, 96 * sizeof(uint16_t)); + return panfrost_pool_upload_aligned(&batch->pool, locations, 96 * sizeof(uint16_t), 64); }