diff options
author | ReinUsesLisp <reinuseslisp@airmail.cc> | 2021-05-10 06:37:30 +0200 |
---|---|---|
committer | ameerj <52414509+ameerj@users.noreply.github.com> | 2021-07-23 03:51:31 +0200 |
commit | 6237300e3605c0b12fb65e2a8818487ec2cb4580 (patch) | |
tree | 9551638b7a4807cc1449c50214cd55de29c31e98 | |
parent | glasm: Fix moving U64 immediates to registers in GLASM (diff) | |
download | yuzu-6237300e3605c0b12fb65e2a8818487ec2cb4580.tar yuzu-6237300e3605c0b12fb65e2a8818487ec2cb4580.tar.gz yuzu-6237300e3605c0b12fb65e2a8818487ec2cb4580.tar.bz2 yuzu-6237300e3605c0b12fb65e2a8818487ec2cb4580.tar.lz yuzu-6237300e3605c0b12fb65e2a8818487ec2cb4580.tar.xz yuzu-6237300e3605c0b12fb65e2a8818487ec2cb4580.tar.zst yuzu-6237300e3605c0b12fb65e2a8818487ec2cb4580.zip |
-rw-r--r-- | src/shader_recompiler/backend/glasm/emit_glasm_floating_point.cpp | 27 |
1 files changed, 15 insertions, 12 deletions
diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_floating_point.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_floating_point.cpp index 2aee5a56c..2e1c7d55f 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_floating_point.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_floating_point.cpp @@ -9,11 +9,10 @@ #include "shader_recompiler/frontend/ir/value.h" namespace Shader::Backend::GLASM { - +namespace { template <typename InputType> -static void Compare(EmitContext& ctx, IR::Inst& inst, InputType lhs, InputType rhs, - std::string_view op, std::string_view type, bool ordered, - bool inequality = false) { +void Compare(EmitContext& ctx, IR::Inst& inst, InputType lhs, InputType rhs, std::string_view op, + std::string_view type, bool ordered, bool inequality = false) { const Register ret{ctx.reg_alloc.Define(inst)}; ctx.Add("{}.{} RC.x,{},{};", op, type, lhs, rhs); if (ordered && inequality) { @@ -35,6 +34,16 @@ static void Compare(EmitContext& ctx, IR::Inst& inst, InputType lhs, InputType r } } +template <typename InputType> +void Clamp(EmitContext& ctx, Register ret, InputType value, InputType min_value, + InputType max_value) { + // Call MAX first to properly clamp nan to min_value instead + ctx.Add("MAX.F {}.x,{},{};" + "MIN.F {}.x,{},{};", + ret, min_value, value, ret, ret, max_value); +} +} // Anonymous namespace + void EmitFPAbs16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst, [[maybe_unused]] Register value) { throw NotImplementedException("GLASM instruction"); @@ -171,18 +180,12 @@ void EmitFPClamp16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] Register void EmitFPClamp32(EmitContext& ctx, IR::Inst& inst, ScalarF32 value, ScalarF32 min_value, ScalarF32 max_value) { - const Register ret{ctx.reg_alloc.Define(inst)}; - ctx.Add("MIN.F {}.x,{},{};" - "MAX.F {}.x,{},{};", - ret, max_value, value, ret, ret, min_value); + Clamp(ctx, ctx.reg_alloc.Define(inst), value, min_value, max_value); } void EmitFPClamp64(EmitContext& ctx, IR::Inst& inst, ScalarF64 value, ScalarF64 min_value, ScalarF64 max_value) { - const Register ret{ctx.reg_alloc.LongDefine(inst)}; - ctx.Add("MIN.F64 {}.x,{},{};" - "MAX.F64 {}.x,{},{};", - ret, max_value, value, ret, ret, min_value); + Clamp(ctx, ctx.reg_alloc.LongDefine(inst), value, min_value, max_value); } void EmitFPRoundEven16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] Register value) { |