diff options
Diffstat (limited to '')
5 files changed, 68 insertions, 8 deletions
diff --git a/src/shader_recompiler/backend/glsl/emit_glsl.cpp b/src/shader_recompiler/backend/glsl/emit_glsl.cpp index d91e04446..66ecfc9f7 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl.cpp @@ -242,6 +242,7 @@ std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info, IR } if (program.info.uses_subgroup_shuffles) { ctx.header += "bool shfl_in_bounds;"; + ctx.header += "uint shfl_result;"; } ctx.code.insert(0, ctx.header); ctx.code += '}'; diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp index 1245c9429..f9be5de1c 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp @@ -141,7 +141,8 @@ void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value, const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)}; ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index, @@ -158,7 +159,8 @@ void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std const auto src_thread_id{fmt::format("({}-{})", THREAD_ID, index)}; ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value, @@ -175,7 +177,8 @@ void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value, const auto src_thread_id{fmt::format("({}+{})", THREAD_ID, index)}; ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value, @@ -193,7 +196,8 @@ void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view val const auto src_thread_id{fmt::format("({}^{})", THREAD_ID, index)}; ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id); SetInBoundsFlag(ctx, inst); - ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value); + ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id); + ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value); } void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b, diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp index 2868fc57d..1d77426e0 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp @@ -111,16 +111,33 @@ Id GetCbuf(EmitContext& ctx, Id result_type, Id UniformDefinitions::*member_ptr, } else if (element_size > 1) { const u32 log2_element_size{static_cast<u32>(std::countr_zero(element_size))}; const Id shift{ctx.Const(log2_element_size)}; - buffer_offset = ctx.OpShiftRightArithmetic(ctx.U32[1], ctx.Def(offset), shift); + buffer_offset = ctx.OpShiftRightLogical(ctx.U32[1], ctx.Def(offset), shift); } else { buffer_offset = ctx.Def(offset); } if (!binding.IsImmediate()) { return ctx.OpFunctionCall(result_type, indirect_func, ctx.Def(binding), buffer_offset); } + const Id cbuf{ctx.cbufs[binding.U32()].*member_ptr}; const Id access_chain{ctx.OpAccessChain(uniform_type, cbuf, ctx.u32_zero_value, buffer_offset)}; - return ctx.OpLoad(result_type, access_chain); + const Id val = ctx.OpLoad(result_type, access_chain); + + if (offset.IsImmediate() || !ctx.profile.has_broken_robust) { + return val; + } + + const auto is_float = UniformDefinitions::IsFloat(member_ptr); + const auto num_elements = UniformDefinitions::NumElements(member_ptr); + const std::array zero_vec{ + is_float ? ctx.Const(0.0f) : ctx.Const(0u), + is_float ? ctx.Const(0.0f) : ctx.Const(0u), + is_float ? ctx.Const(0.0f) : ctx.Const(0u), + is_float ? ctx.Const(0.0f) : ctx.Const(0u), + }; + const Id cond = ctx.OpULessThanEqual(ctx.TypeBool(), buffer_offset, ctx.Const(0xFFFFu)); + const Id zero = ctx.OpCompositeConstruct(result_type, std::span(zero_vec.data(), num_elements)); + return ctx.OpSelect(result_type, cond, val, zero); } Id GetCbufU32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) { @@ -138,7 +155,7 @@ Id GetCbufElement(EmitContext& ctx, Id vector, const IR::Value& offset, u32 inde const u32 element{(offset.U32() / 4) % 4 + index_offset}; return ctx.OpCompositeExtract(ctx.U32[1], vector, element); } - const Id shift{ctx.OpShiftRightArithmetic(ctx.U32[1], ctx.Def(offset), ctx.Const(2u))}; + const Id shift{ctx.OpShiftRightLogical(ctx.U32[1], ctx.Def(offset), ctx.Const(2u))}; Id element{ctx.OpBitwiseAnd(ctx.U32[1], shift, ctx.Const(3u))}; if (index_offset > 0) { element = ctx.OpIAdd(ctx.U32[1], element, ctx.Const(index_offset)); diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h index 7c49fd504..1aa79863d 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h @@ -64,6 +64,42 @@ struct UniformDefinitions { Id F32{}; Id U32x2{}; Id U32x4{}; + + constexpr static size_t NumElements(Id UniformDefinitions::*member_ptr) { + if (member_ptr == &UniformDefinitions::U8) { + return 1; + } + if (member_ptr == &UniformDefinitions::S8) { + return 1; + } + if (member_ptr == &UniformDefinitions::U16) { + return 1; + } + if (member_ptr == &UniformDefinitions::S16) { + return 1; + } + if (member_ptr == &UniformDefinitions::U32) { + return 1; + } + if (member_ptr == &UniformDefinitions::F32) { + return 1; + } + if (member_ptr == &UniformDefinitions::U32x2) { + return 2; + } + if (member_ptr == &UniformDefinitions::U32x4) { + return 4; + } + ASSERT(false); + return 1; + } + + constexpr static bool IsFloat(Id UniformDefinitions::*member_ptr) { + if (member_ptr == &UniformDefinitions::F32) { + return true; + } + return false; + } }; struct StorageTypeDefinition { diff --git a/src/shader_recompiler/profile.h b/src/shader_recompiler/profile.h index 9ca97f6a4..38d820db2 100644 --- a/src/shader_recompiler/profile.h +++ b/src/shader_recompiler/profile.h @@ -9,7 +9,6 @@ namespace Shader { struct Profile { u32 supported_spirv{0x00010000}; - bool unified_descriptor_binding{}; bool support_descriptor_aliasing{}; bool support_int8{}; @@ -82,6 +81,9 @@ struct Profile { bool has_broken_spirv_subgroup_mask_vector_extract_dynamic{}; u32 gl_max_compute_smem_size{}; + + /// Maxwell and earlier nVidia architectures have broken robust support + bool has_broken_robust{}; }; } // namespace Shader |