shader: Implement more of XMAD and FFMA32I and fix XMAD.CBCC

5 files changed, 76 insertions, 28 deletions

diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.h b/src/shader_recompiler/backend/spirv/emit_spirv.h
index 130c71996..4b74cf04d 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.h
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.h
@@ -224,8 +224,8 @@ void EmitShiftRightArithmetic32(EmitContext& ctx);
 Id EmitBitwiseAnd32(EmitContext& ctx, Id a, Id b);
 Id EmitBitwiseOr32(EmitContext& ctx, Id a, Id b);
 Id EmitBitwiseXor32(EmitContext& ctx, Id a, Id b);
-void EmitBitFieldInsert(EmitContext& ctx);
-void EmitBitFieldSExtract(EmitContext& ctx);
+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 EmitSLessThan(EmitContext& ctx, Id lhs, Id rhs);
 Id EmitULessThan(EmitContext& ctx, Id lhs, Id rhs);
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
index 329dcb351..8aaa0e381 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
@@ -90,12 +90,12 @@ Id EmitBitwiseXor32(EmitContext& ctx, Id a, Id b) {
     return ctx.OpBitwiseXor(ctx.U32[1], a, b);
 }
 
-void EmitBitFieldInsert(EmitContext&) {
-    throw NotImplementedException("SPIR-V Instruction");
+Id EmitBitFieldInsert(EmitContext& ctx, Id base, Id insert, Id offset, Id count) {
+    return ctx.OpBitFieldInsert(ctx.U32[1], base, insert, offset, count);
 }
 
-void EmitBitFieldSExtract(EmitContext&) {
-    throw NotImplementedException("SPIR-V Instruction");
+Id EmitBitFieldSExtract(EmitContext& ctx, Id base, Id offset, Id count) {
+    return ctx.OpBitFieldSExtract(ctx.U32[1], base, offset, count);
 }
 
 Id EmitBitFieldUExtract(EmitContext& ctx, Id base, Id offset, Id count) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
index c2ca0873b..18561bc9c 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
@@ -17,9 +17,6 @@ void FFMA(TranslatorVisitor& v, u64 insn, const IR::F32& src_b, const IR::F32& s
         BitField<8, 8, IR::Reg> src_a;
     } const ffma{insn};
 
-    if (sat) {
-        throw NotImplementedException("FFMA SAT");
-    }
     if (cc) {
         throw NotImplementedException("FFMA CC");
     }
@@ -31,7 +28,20 @@ void FFMA(TranslatorVisitor& v, u64 insn, const IR::F32& src_b, const IR::F32& s
         .rounding{CastFpRounding(fp_rounding)},
         .fmz_mode{CastFmzMode(fmz_mode)},
     };
-    v.F(ffma.dest_reg, v.ir.FPFma(op_a, op_b, op_c, fp_control));
+    IR::F32 value{v.ir.FPFma(op_a, op_b, op_c, fp_control)};
+    if (fmz_mode == FmzMode::FMZ && !sat) {
+        // Do not implement FMZ if SAT is enabled, as it does the logic for us.
+        // On D3D9 mode, anything * 0 is zero, even NAN and infinity
+        const IR::F32 zero{v.ir.Imm32(0.0f)};
+        const IR::U1 zero_a{v.ir.FPEqual(op_a, zero)};
+        const IR::U1 zero_b{v.ir.FPEqual(op_b, zero)};
+        const IR::U1 any_zero{v.ir.LogicalOr(zero_a, zero_b)};
+        value = IR::F32{v.ir.Select(any_zero, op_c, value)};
+    }
+    if (sat) {
+        value = v.ir.FPSaturate(value);
+    }
+    v.F(ffma.dest_reg, value);
 }
 
 void FFMA(TranslatorVisitor& v, u64 insn, const IR::F32& src_b, const IR::F32& src_c) {
@@ -54,20 +64,31 @@ void TranslatorVisitor::FFMA_reg(u64 insn) {
     FFMA(*this, insn, GetFloatReg20(insn), GetFloatReg39(insn));
 }
 
-void TranslatorVisitor::FFMA_rc(u64) {
-    throw NotImplementedException("FFMA (rc)");
+void TranslatorVisitor::FFMA_rc(u64 insn) {
+    FFMA(*this, insn, GetFloatReg39(insn), GetFloatCbuf(insn));
 }
 
 void TranslatorVisitor::FFMA_cr(u64 insn) {
     FFMA(*this, insn, GetFloatCbuf(insn), GetFloatReg39(insn));
 }
 
-void TranslatorVisitor::FFMA_imm(u64) {
-    throw NotImplementedException("FFMA (imm)");
+void TranslatorVisitor::FFMA_imm(u64 insn) {
+    FFMA(*this, insn, GetFloatImm20(insn), GetFloatReg39(insn));
 }
 
-void TranslatorVisitor::FFMA32I(u64) {
-    throw NotImplementedException("FFMA32I");
+void TranslatorVisitor::FFMA32I(u64 insn) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> src_c; // FFMA32I mirrors the destination and addition register
+        BitField<52, 1, u64> cc;
+        BitField<53, 2, FmzMode> fmz_mode;
+        BitField<55, 1, u64> sat;
+        BitField<56, 1, u64> neg_a;
+        BitField<57, 1, u64> neg_c;
+    } const ffma32i{insn};
+
+    FFMA(*this, insn, GetFloatImm32(insn), F(ffma32i.src_c), ffma32i.neg_a != 0, false,
+         ffma32i.neg_c != 0, ffma32i.sat != 0, ffma32i.cc != 0, ffma32i.fmz_mode, FpRounding::RN);
 }
 
 } // namespace Shader::Maxwell
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
index edf2cadae..72f0a18ae 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
@@ -94,6 +94,7 @@ void FMUL(TranslatorVisitor& v, u64 insn, const IR::F32& src_b) {
         BitField<48, 1, u64> neg_b;
         BitField<50, 1, u64> sat;
     } const fmul{insn};
+
     FMUL(v, insn, src_b, fmul.fmz, fmul.fp_rounding, fmul.scale, fmul.sat != 0, fmul.cc != 0,
          fmul.neg_b != 0);
 }
@@ -118,6 +119,7 @@ void TranslatorVisitor::FMUL32I(u64 insn) {
         BitField<53, 2, FmzMode> fmz;
         BitField<55, 1, u64> sat;
     } const fmul32i{insn};
+
     FMUL(*this, insn, GetFloatImm32(insn), fmul32i.fmz, FpRounding::RN, Scale::None,
          fmul32i.sat != 0, fmul32i.cc != 0, false);
 }
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_short_multiply_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_short_multiply_add.cpp
index 70a7c76c5..2932cdc42 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_short_multiply_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_short_multiply_add.cpp
@@ -58,7 +58,7 @@ void XMAD(TranslatorVisitor& v, u64 insn, const IR::U32& src_b, const IR::U32& s
         case SelectMode::CHI:
             return ExtractHalf(v, src_c, Half::H1, false);
         case SelectMode::CBCC:
-            return v.ir.IAdd(v.ir.ShiftLeftLogical(src_b, v.ir.Imm32(16)), src_b);
+            return v.ir.IAdd(v.ir.ShiftLeftLogical(src_b, v.ir.Imm32(16)), src_c);
         case SelectMode::CSFU:
             throw NotImplementedException("XMAD CSFU");
         }
@@ -78,16 +78,44 @@ void XMAD(TranslatorVisitor& v, u64 insn, const IR::U32& src_b, const IR::U32& s
 }
 } // Anonymous namespace
 
-void TranslatorVisitor::XMAD_reg(u64) {
-    throw NotImplementedException("XMAD (reg)");
+void TranslatorVisitor::XMAD_reg(u64 insn) {
+    union {
+        u64 raw;
+        BitField<35, 1, Half> half_b;
+        BitField<36, 1, u64> psl;
+        BitField<37, 1, u64> mrg;
+        BitField<38, 1, u64> x;
+        BitField<50, 3, SelectMode> select_mode;
+    } const xmad{insn};
+
+    XMAD(*this, insn, GetReg20(insn), GetReg39(insn), xmad.select_mode, xmad.half_b, xmad.psl != 0,
+         xmad.mrg != 0, xmad.x != 0);
 }
 
-void TranslatorVisitor::XMAD_rc(u64) {
-    throw NotImplementedException("XMAD (rc)");
+void TranslatorVisitor::XMAD_rc(u64 insn) {
+    union {
+        u64 raw;
+        BitField<50, 2, SelectMode> select_mode;
+        BitField<52, 1, Half> half_b;
+        BitField<54, 1, u64> x;
+    } const xmad{insn};
+
+    XMAD(*this, insn, GetReg39(insn), GetCbuf(insn), xmad.select_mode, xmad.half_b, false, false,
+         xmad.x != 0);
 }
 
-void TranslatorVisitor::XMAD_cr(u64) {
-    throw NotImplementedException("XMAD (cr)");
+void TranslatorVisitor::XMAD_cr(u64 insn) {
+    union {
+        u64 raw;
+        BitField<50, 2, SelectMode> select_mode;
+        BitField<52, 1, Half> half_b;
+        BitField<54, 1, u64> x;
+        BitField<55, 1, u64> psl;
+        BitField<56, 1, u64> mrg;
+    } const xmad{insn};
+
+    XMAD(*this, insn, GetCbuf(insn), GetReg39(insn), xmad.select_mode, xmad.half_b, xmad.psl != 0,
+         xmad.mrg != 0, xmad.x != 0);
 }
 
 void TranslatorVisitor::XMAD_imm(u64 insn) {
@@ -97,14 +125,11 @@ void TranslatorVisitor::XMAD_imm(u64 insn) {
         BitField<36, 1, u64> psl;
         BitField<37, 1, u64> mrg;
         BitField<38, 1, u64> x;
-        BitField<39, 8, IR::Reg> src_c;
         BitField<50, 3, SelectMode> select_mode;
     } const xmad{insn};
 
-    const IR::U32 src_b{ir.Imm32(static_cast<u32>(xmad.src_b))};
-    const IR::U32 src_c{X(xmad.src_c)};
-    XMAD(*this, insn, src_b, src_c, xmad.select_mode, Half::H0, xmad.psl != 0, xmad.mrg != 0,
-         xmad.x != 0);
+    XMAD(*this, insn, ir.Imm32(static_cast<u32>(xmad.src_b)), GetReg39(insn), xmad.select_mode,
+         Half::H0, xmad.psl != 0, xmad.mrg != 0, xmad.x != 0);
 }
 
 } // namespace Shader::Maxwell