diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp index 152659618e5..1857edeb27d 100644 --- a/src/amd/compiler/aco_instruction_selection.cpp +++ b/src/amd/compiler/aco_instruction_selection.cpp @@ -9124,6 +9124,28 @@ tex_fetch_ptrs(isel_context* ctx, nir_tex_instr* instr, Temp* res_ptr, Temp* sam } if (samp_ptr) { *samp_ptr = get_sampler_desc(ctx, sampler_deref_instr, ACO_DESC_SAMPLER, instr, false); + + if (ctx->options->disable_aniso_single_level && + instr->sampler_dim < GLSL_SAMPLER_DIM_RECT && ctx->options->chip_class < GFX8) { + /* fix sampler aniso on SI/CI: samp[0] = samp[0] & img[7] */ + Builder bld(ctx->program, ctx->block); + + /* to avoid unnecessary moves, we split and recombine sampler and image */ + Temp img[8] = {bld.tmp(s1), bld.tmp(s1), bld.tmp(s1), bld.tmp(s1), + bld.tmp(s1), bld.tmp(s1), bld.tmp(s1), bld.tmp(s1)}; + Temp samp[4] = {bld.tmp(s1), bld.tmp(s1), bld.tmp(s1), bld.tmp(s1)}; + bld.pseudo(aco_opcode::p_split_vector, Definition(img[0]), Definition(img[1]), + Definition(img[2]), Definition(img[3]), Definition(img[4]), Definition(img[5]), + Definition(img[6]), Definition(img[7]), *res_ptr); + bld.pseudo(aco_opcode::p_split_vector, Definition(samp[0]), Definition(samp[1]), + Definition(samp[2]), Definition(samp[3]), *samp_ptr); + + samp[0] = bld.sop2(aco_opcode::s_and_b32, bld.def(s1), bld.def(s1, scc), samp[0], img[7]); + *res_ptr = bld.pseudo(aco_opcode::p_create_vector, bld.def(s8), img[0], img[1], img[2], + img[3], img[4], img[5], img[6], img[7]); + *samp_ptr = bld.pseudo(aco_opcode::p_create_vector, bld.def(s4), samp[0], samp[1], samp[2], + samp[3]); + } } } diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c index d9cdd61236c..72f6481f3ca 100644 --- a/src/amd/vulkan/radv_device.c +++ b/src/amd/vulkan/radv_device.c @@ -916,6 +916,7 @@ static const driOptionDescription radv_dri_options[] = { DRI_CONF_RADV_REPORT_APU_AS_DGPU(false) DRI_CONF_RADV_REQUIRE_ETC2(false) DRI_CONF_RADV_DISABLE_HTILE_LAYERS(false) + DRI_CONF_RADV_DISABLE_ANISO_SINGLE_LEVEL(false) DRI_CONF_SECTION_END }; // clang-format on @@ -964,6 +965,9 @@ radv_init_dri_options(struct radv_instance *instance) instance->disable_htile_layers = driQueryOptionb(&instance->dri_options, "radv_disable_htile_layers"); + + instance->disable_aniso_single_level = + driQueryOptionb(&instance->dri_options, "radv_disable_aniso_single_level"); } VKAPI_ATTR VkResult VKAPI_CALL @@ -6096,10 +6100,15 @@ radv_init_sampler(struct radv_device *device, struct radv_sampler *sampler, sampler->state[3] = (S_008F3C_BORDER_COLOR_PTR(border_color_ptr) | S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(border_color))); - if (device->physical_device->rad_info.chip_class < GFX10) { + if (device->physical_device->rad_info.chip_class >= GFX10) { + sampler->state[2] |= + S_008F38_ANISO_OVERRIDE_GFX10(device->instance->disable_aniso_single_level); + } else { sampler->state[2] |= S_008F38_DISABLE_LSB_CEIL(device->physical_device->rad_info.chip_class <= GFX8) | - S_008F38_FILTER_PREC_FIX(1); + S_008F38_FILTER_PREC_FIX(1) | + S_008F38_ANISO_OVERRIDE_GFX8(device->instance->disable_aniso_single_level && + device->physical_device->rad_info.chip_class >= GFX8); } } diff --git a/src/amd/vulkan/radv_image.c b/src/amd/vulkan/radv_image.c index cfbf272f660..712dfe3c81b 100644 --- a/src/amd/vulkan/radv_image.c +++ b/src/amd/vulkan/radv_image.c @@ -1183,6 +1183,18 @@ si_make_texture_descriptor(struct radv_device *device, struct radv_image *image, if (!(image->planes[0].surface.flags & RADEON_SURF_Z_OR_SBUFFER) && image->planes[0].surface.meta_offset) { state[6] = S_008F28_ALPHA_IS_ON_MSB(vi_alpha_is_on_msb(device, vk_format)); + } else { + if (device->instance->disable_aniso_single_level) { + /* The last dword is unused by hw. The shader uses it to clear + * bits in the first dword of sampler state. + */ + if (device->physical_device->rad_info.chip_class <= GFX7 && image->info.samples <= 1) { + if (first_level == last_level) + state[7] = C_008F30_MAX_ANISO_RATIO; + else + state[7] = 0xffffffff; + } + } } /* Initialize the sampler view for FMASK. */ diff --git a/src/amd/vulkan/radv_nir_to_llvm.c b/src/amd/vulkan/radv_nir_to_llvm.c index 2dbd0def8df..2687fc7d116 100644 --- a/src/amd/vulkan/radv_nir_to_llvm.c +++ b/src/amd/vulkan/radv_nir_to_llvm.c @@ -2394,7 +2394,7 @@ ac_translate_nir_to_llvm(struct ac_llvm_compiler *ac_llvm, ctx.abi.clamp_shadow_reference = false; ctx.abi.adjust_frag_coord_z = options->adjust_frag_coord_z; ctx.abi.robust_buffer_access = options->robust_buffer_access; - ctx.abi.disable_aniso_single_level = false; + ctx.abi.disable_aniso_single_level = options->disable_aniso_single_level; bool is_ngg = is_pre_gs_stage(shaders[0]->info.stage) && info->is_ngg; if (shader_count >= 2 || is_ngg) diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h index c72567cc14c..c9484a4f0d9 100644 --- a/src/amd/vulkan/radv_private.h +++ b/src/amd/vulkan/radv_private.h @@ -338,6 +338,7 @@ struct radv_instance { bool absolute_depth_bias; bool report_apu_as_dgpu; bool disable_htile_layers; + bool disable_aniso_single_level; }; VkResult radv_init_wsi(struct radv_physical_device *physical_device); diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c index afc94245725..b43dea2cb23 100644 --- a/src/amd/vulkan/radv_shader.c +++ b/src/amd/vulkan/radv_shader.c @@ -1853,6 +1853,7 @@ shader_compile(struct radv_device *device, struct vk_shader_module *module, options->enable_mrt_output_nan_fixup = module && !is_meta_shader(module->nir) && options->key.ps.enable_mrt_output_nan_fixup; options->adjust_frag_coord_z = device->adjust_frag_coord_z; + options->disable_aniso_single_level = device->instance->disable_aniso_single_level; options->has_image_load_dcc_bug = device->physical_device->rad_info.has_image_load_dcc_bug; options->debug.func = radv_compiler_debug; options->debug.private_data = &debug_data; diff --git a/src/amd/vulkan/radv_shader.h b/src/amd/vulkan/radv_shader.h index 978b086c540..7403b06fc16 100644 --- a/src/amd/vulkan/radv_shader.h +++ b/src/amd/vulkan/radv_shader.h @@ -122,6 +122,7 @@ struct radv_nir_compiler_options { bool enable_mrt_output_nan_fixup; bool wgp_mode; bool remap_spi_ps_input; + bool disable_aniso_single_level; enum radeon_family family; enum chip_class chip_class; const struct radeon_info *info; diff --git a/src/util/driconf.h b/src/util/driconf.h index 43a9f109450..5e1c9591853 100644 --- a/src/util/driconf.h +++ b/src/util/driconf.h @@ -568,4 +568,8 @@ DRI_CONF_OPT_B(radv_disable_htile_layers, def, \ "Disable HTILE for layered depth/stencil formats") +#define DRI_CONF_RADV_DISABLE_ANISO_SINGLE_LEVEL(def) \ + DRI_CONF_OPT_B(radv_disable_aniso_single_level, def, \ + "Disable anisotropic filtering for single level images") + #endif