shader: Implement SHR

5 years ago · b3d166ad4c
8 changed files with 80 additions and 18 deletions
--- a/src/shader_recompiler/CMakeLists.txt
+++ b/src/shader_recompiler/CMakeLists.txt
@ -72,6 +72,7 @@ add_library(shader_recompiler STATIC
    frontend/maxwell/translate/impl/integer_scaled_add.cpp
    frontend/maxwell/translate/impl/integer_set_predicate.cpp
    frontend/maxwell/translate/impl/integer_shift_left.cpp
+    frontend/maxwell/translate/impl/integer_shift_right.cpp
    frontend/maxwell/translate/impl/integer_short_multiply_add.cpp
    frontend/maxwell/translate/impl/load_store_attribute.cpp
    frontend/maxwell/translate/impl/load_store_memory.cpp
--- a/src/shader_recompiler/backend/spirv/emit_spirv.h
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.h
@ -219,14 +219,15 @@ Id EmitIMul32(EmitContext& ctx, Id a, Id b);
 Id EmitINeg32(EmitContext& ctx, Id value);
 Id EmitIAbs32(EmitContext& ctx, Id value);
 Id EmitShiftLeftLogical32(EmitContext& ctx, Id base, Id shift);
-void EmitShiftRightLogical32(EmitContext& ctx);
-void EmitShiftRightArithmetic32(EmitContext& ctx);
+Id EmitShiftRightLogical32(EmitContext& ctx, Id a, Id b);
+Id EmitShiftRightArithmetic32(EmitContext& ctx, Id a, Id b);
 Id EmitBitwiseAnd32(EmitContext& ctx, Id a, Id b);
 Id EmitBitwiseOr32(EmitContext& ctx, Id a, Id b);
 Id EmitBitwiseXor32(EmitContext& ctx, Id a, Id b);
 Id EmitBitFieldInsert(EmitContext& ctx, Id base, Id insert, Id offset, Id count);
 Id EmitBitFieldSExtract(EmitContext& ctx, Id base, Id offset, Id count);
 Id EmitBitFieldUExtract(EmitContext& ctx, Id base, Id offset, Id count);
+Id EmitBitReverse32(EmitContext& ctx, Id value);
 Id EmitSLessThan(EmitContext& ctx, Id lhs, Id rhs);
 Id EmitULessThan(EmitContext& ctx, Id lhs, Id rhs);
 Id EmitIEqual(EmitContext& ctx, Id lhs, Id rhs);
--- a/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
@ -70,12 +70,12 @@ Id EmitShiftLeftLogical32(EmitContext& ctx, Id base, Id shift) {
    return ctx.OpShiftLeftLogical(ctx.U32[1], base, shift);
 }

-void EmitShiftRightLogical32(EmitContext&) {
-    throw NotImplementedException("SPIR-V Instruction");
+Id EmitShiftRightLogical32(EmitContext& ctx, Id a, Id b) {
+    return ctx.OpShiftRightLogical(ctx.U32[1], a, b);
 }

-void EmitShiftRightArithmetic32(EmitContext&) {
-    throw NotImplementedException("SPIR-V Instruction");
+Id EmitShiftRightArithmetic32(EmitContext& ctx, Id a, Id b) {
+    return ctx.OpShiftRightArithmetic(ctx.U32[1], a, b);
 }

 Id EmitBitwiseAnd32(EmitContext& ctx, Id a, Id b) {
@ -102,6 +102,10 @@ Id EmitBitFieldUExtract(EmitContext& ctx, Id base, Id offset, Id count) {
    return ctx.OpBitFieldUExtract(ctx.U32[1], base, offset, count);
 }

+Id EmitBitReverse32(EmitContext& ctx, Id value) {
+    return ctx.OpBitReverse(ctx.U32[1], value);
+}
+
 Id EmitSLessThan(EmitContext& ctx, Id lhs, Id rhs) {
    return ctx.OpSLessThan(ctx.U1, lhs, rhs);
 }
--- a/src/shader_recompiler/frontend/ir/ir_emitter.cpp
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
@ -804,6 +804,10 @@ U32 IREmitter::BitFieldExtract(const U32& base, const U32& offset, const U32& co
                     count);
 }

+U32 IREmitter::BitReverse(const U32& value) {
+    return Inst<U32>(Opcode::BitReverse32, value);
+}
+
 U1 IREmitter::ILessThan(const U32& lhs, const U32& rhs, bool is_signed) {
    return Inst<U1>(is_signed ? Opcode::SLessThan : Opcode::ULessThan, lhs, rhs);
 }
--- a/src/shader_recompiler/frontend/ir/ir_emitter.h
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.h
@ -159,6 +159,7 @@ public:
                                     const U32& count);
    [[nodiscard]] U32 BitFieldExtract(const U32& base, const U32& offset, const U32& count,
                                      bool is_signed);
+    [[nodiscard]] U32 BitReverse(const U32& value);

    [[nodiscard]] U1 ILessThan(const U32& lhs, const U32& rhs, bool is_signed);
    [[nodiscard]] U1 IEqual(const U32& lhs, const U32& rhs);
--- a/src/shader_recompiler/frontend/ir/opcodes.inc
+++ b/src/shader_recompiler/frontend/ir/opcodes.inc
@ -231,6 +231,7 @@ OPCODE(BitwiseXor32,                                        U32,            U32,
 OPCODE(BitFieldInsert,                                      U32,            U32,            U32,            U32,            U32,            )
 OPCODE(BitFieldSExtract,                                    U32,            U32,            U32,            U32,                            )
 OPCODE(BitFieldUExtract,                                    U32,            U32,            U32,            U32,                            )
+OPCODE(BitReverse32,                                        U32,            U32,                                                            )

 OPCODE(SLessThan,                                           U1,             U32,            U32,                                            )
 OPCODE(ULessThan,                                           U1,             U32,            U32,                                            )
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_shift_right.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_shift_right.cpp
@ -0,0 +1,62 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+void SHR(TranslatorVisitor& v, u64 insn, const IR::U32& shift) {
+    union {
+        u64 insn;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_reg_a;
+        BitField<39, 1, u64> is_wrapped;
+        BitField<40, 1, u64> brev;
+        BitField<43, 1, u64> xmode;
+        BitField<48, 1, u64> is_arithmetic;
+    } const shr{insn};
+
+    if (shr.xmode != 0) {
+        throw NotImplementedException("SHR.XMODE");
+    }
+
+    IR::U32 base{v.X(shr.src_reg_a)};
+    if (shr.brev == 1) {
+        base = v.ir.BitReverse(base);
+    }
+    IR::U32 result;
+    const IR::U32 safe_shift = shr.is_wrapped == 0 ? shift : v.ir.BitwiseAnd(shift, v.ir.Imm32(31));
+    if (shr.is_arithmetic == 1) {
+        result = IR::U32{v.ir.ShiftRightArithmetic(base, safe_shift)};
+    } else {
+        result = IR::U32{v.ir.ShiftRightLogical(base, safe_shift)};
+    }
+
+    if (shr.is_wrapped == 0) {
+        const IR::U32 zero{v.ir.Imm32(0)};
+        const IR::U32 safe_bits{v.ir.Imm32(32)};
+
+        const IR::U1 is_negative{v.ir.ILessThan(result, zero, true)};
+        const IR::U1 is_safe{v.ir.ILessThan(shift, safe_bits, false)};
+        const IR::U32 clamped_value{v.ir.Select(is_negative, v.ir.Imm32(-1), zero)};
+        result = IR::U32{v.ir.Select(is_safe, result, clamped_value)};
+    }
+    v.X(shr.dest_reg, result);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::SHR_reg(u64 insn) {
+    SHR(*this, insn, GetReg20(insn));
+}
+
+void TranslatorVisitor::SHR_cbuf(u64 insn) {
+    SHR(*this, insn, GetCbuf(insn));
+}
+
+void TranslatorVisitor::SHR_imm(u64 insn) {
+    SHR(*this, insn, GetImm20(insn));
+}
+} // namespace Shader::Maxwell
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
@ -757,18 +757,6 @@ void TranslatorVisitor::SHFL(u64) {
    ThrowNotImplemented(Opcode::SHFL);
 }

-void TranslatorVisitor::SHR_reg(u64) {
-    ThrowNotImplemented(Opcode::SHR_reg);
-}
-
-void TranslatorVisitor::SHR_cbuf(u64) {
-    ThrowNotImplemented(Opcode::SHR_cbuf);
-}
-
-void TranslatorVisitor::SHR_imm(u64) {
-    ThrowNotImplemented(Opcode::SHR_imm);
-}
-
 void TranslatorVisitor::SSY() {
    // SSY is a no-op
 }