From 9a2df304edbe7976c5cd7c522e06dc66b8b08b6e Mon Sep 17 00:00:00 2001 From: Alyssa Rosenzweig Date: Wed, 19 Aug 2020 10:25:32 -0400 Subject: [PATCH] panfrost: Specialize compute vs frag shader init In exchange for a bit of code duplication, we can streamline both routines. A huge amount of the descriptor is unused for non-fragment shaders, and even some of the parts that are used appear to have varying meanings. Given we want to emit the descriptors atomically, this seems like a reasonable tradeoff. Signed-off-by: Alyssa Rosenzweig Reviewed-by: Tomeu Vizoso Part-of: --- src/gallium/drivers/panfrost/pan_cmdstream.c | 56 +++++++++++++------- 1 file changed, 37 insertions(+), 19 deletions(-) diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.c b/src/gallium/drivers/panfrost/pan_cmdstream.c index 45465a5413e..c75989c435b 100644 --- a/src/gallium/drivers/panfrost/pan_cmdstream.c +++ b/src/gallium/drivers/panfrost/pan_cmdstream.c @@ -307,7 +307,7 @@ panfrost_vt_set_draw_info(struct panfrost_context *ctx, } static void -panfrost_shader_meta_init(struct panfrost_context *ctx, +panfrost_compute_shader_meta_init(struct panfrost_context *ctx, enum pipe_shader_type st, struct mali_shader_meta *meta) { @@ -322,23 +322,10 @@ panfrost_shader_meta_init(struct panfrost_context *ctx, meta->sampler_count = ctx->sampler_count[st]; if (dev->quirks & IS_BIFROST) { - if (st == PIPE_SHADER_VERTEX) - meta->bifrost1.unk1 = 0x800000; - else { - /* First clause ATEST |= 0x4000000. - * Less than 32 regs |= 0x200 */ - meta->bifrost1.unk1 = 0x950020; - } - - meta->bifrost1.uniform_buffer_count = panfrost_ubo_count(ctx, st); - if (st == PIPE_SHADER_VERTEX) - meta->bifrost2.preload_regs = 0xC0; - else { - meta->bifrost2.preload_regs = 0x1; - SET_BIT(meta->bifrost2.preload_regs, 0x10, ss->reads_frag_coord); - } - + meta->bifrost1.unk1 = 0x800000; + meta->bifrost2.preload_regs = 0xC0; meta->bifrost2.uniform_count = ss->uniform_count; + meta->bifrost1.uniform_buffer_count = panfrost_ubo_count(ctx, st); } else { meta->midgard1.uniform_count = ss->uniform_count; meta->midgard1.work_count = ss->work_reg_count; @@ -566,6 +553,37 @@ panfrost_frag_shader_meta_init(struct panfrost_context *ctx, struct pipe_rasterizer_state *rast = &ctx->rasterizer->base; const struct panfrost_zsa_state *zsa = ctx->depth_stencil; + memset(fragmeta, 0, sizeof(*fragmeta)); + + fragmeta->shader = fs->shader; + fragmeta->attribute_count = fs->attribute_count; + fragmeta->varying_count = fs->varying_count; + fragmeta->texture_count = ctx->sampler_view_count[PIPE_SHADER_FRAGMENT]; + fragmeta->sampler_count = ctx->sampler_count[PIPE_SHADER_FRAGMENT]; + + if (dev->quirks & IS_BIFROST) { + /* First clause ATEST |= 0x4000000. + * Lefs than 32 regs |= 0x200 */ + fragmeta->bifrost1.unk1 = 0x950020; + + fragmeta->bifrost1.uniform_buffer_count = panfrost_ubo_count(ctx, PIPE_SHADER_FRAGMENT); + fragmeta->bifrost2.preload_regs = 0x1; + SET_BIT(fragmeta->bifrost2.preload_regs, 0x10, fs->reads_frag_coord); + + fragmeta->bifrost2.uniform_count = fs->uniform_count; + } else { + fragmeta->midgard1.uniform_count = fs->uniform_count; + fragmeta->midgard1.work_count = fs->work_reg_count; + + /* TODO: This is not conformant on ES3 */ + fragmeta->midgard1.flags_hi = MALI_SUPPRESS_INF_NAN; + + fragmeta->midgard1.flags_lo = 0x20; + fragmeta->midgard1.uniform_buffer_count = panfrost_ubo_count(ctx, PIPE_SHADER_FRAGMENT); + + SET_BIT(fragmeta->midgard1.flags_lo, MALI_WRITES_GLOBAL, fs->writes_global); + } + bool msaa = rast->multisample; fragmeta->coverage_mask = msaa ? ctx->sample_mask : ~0; @@ -755,8 +773,6 @@ panfrost_emit_shader_meta(struct panfrost_batch *batch, struct mali_shader_meta meta; - panfrost_shader_meta_init(ctx, st, &meta); - /* Add the shader BO to the batch. */ panfrost_batch_add_bo(batch, ss->bo, PAN_BO_ACCESS_PRIVATE | @@ -807,6 +823,8 @@ panfrost_emit_shader_meta(struct panfrost_batch *batch, shader_ptr = xfer.gpu; } else { + panfrost_compute_shader_meta_init(ctx, st, &meta); + shader_ptr = panfrost_pool_upload(&batch->pool, &meta, sizeof(meta)); } -- 2.30.2