1
0
Fork 0
forked from suyu/suyu

shader/video: Partially implement VMNMX

Implements the common usages for VMNMX. Inputs with a different size
than 32 bits are not supported and sign mismatches aren't supported
either.

VMNMX works as follows:
It grabs Ra and Rb and applies a maximum/minimum on them (this is
defined by .MX), having in mind the input sign. This result can then be
saturated. After the intermediate result is calculated, it applies
another operation on it using Rc. These operations are merges,
accumulations or another min/max pass.

This instruction allows to implement with a more flexible approach GCN's
min3 and max3 instructions (for instance).
This commit is contained in:
ReinUsesLisp 2020-03-28 18:49:50 -03:00
parent 08470d261d
commit 76f178ba6e
3 changed files with 116 additions and 0 deletions

View file

@ -302,6 +302,23 @@ enum class VmadShr : u64 {
Shr15 = 2, Shr15 = 2,
}; };
enum class VmnmxType : u64 {
Bits8,
Bits16,
Bits32,
};
enum class VmnmxOperation : u64 {
Mrg_16H = 0,
Mrg_16L = 1,
Mrg_8B0 = 2,
Mrg_8B2 = 3,
Acc = 4,
Min = 5,
Max = 6,
Nop = 7,
};
enum class XmadMode : u64 { enum class XmadMode : u64 {
None = 0, None = 0,
CLo = 1, CLo = 1,
@ -1662,6 +1679,42 @@ union Instruction {
BitField<47, 1, u64> cc; BitField<47, 1, u64> cc;
} vmad; } vmad;
union {
BitField<54, 1, u64> is_dest_signed;
BitField<48, 1, u64> is_src_a_signed;
BitField<49, 1, u64> is_src_b_signed;
BitField<37, 2, u64> src_format_a;
BitField<29, 2, u64> src_format_b;
BitField<56, 1, u64> mx;
BitField<55, 1, u64> sat;
BitField<36, 2, u64> selector_a;
BitField<28, 2, u64> selector_b;
BitField<50, 1, u64> is_op_b_register;
BitField<51, 3, VmnmxOperation> operation;
VmnmxType SourceFormatA() const {
switch (src_format_a) {
case 0b11:
return VmnmxType::Bits32;
case 0b10:
return VmnmxType::Bits16;
default:
return VmnmxType::Bits8;
}
}
VmnmxType SourceFormatB() const {
switch (src_format_b) {
case 0b11:
return VmnmxType::Bits32;
case 0b10:
return VmnmxType::Bits16;
default:
return VmnmxType::Bits8;
}
}
} vmnmx;
union { union {
BitField<20, 16, u64> imm20_16; BitField<20, 16, u64> imm20_16;
BitField<35, 1, u64> high_b_rr; // used on RR BitField<35, 1, u64> high_b_rr; // used on RR
@ -1773,6 +1826,7 @@ public:
MEMBAR, MEMBAR,
VMAD, VMAD,
VSETP, VSETP,
VMNMX,
FFMA_IMM, // Fused Multiply and Add FFMA_IMM, // Fused Multiply and Add
FFMA_CR, FFMA_CR,
FFMA_RC, FFMA_RC,
@ -2078,6 +2132,7 @@ private:
INST("1110111110011---", Id::MEMBAR, Type::Trivial, "MEMBAR"), INST("1110111110011---", Id::MEMBAR, Type::Trivial, "MEMBAR"),
INST("01011111--------", Id::VMAD, Type::Video, "VMAD"), INST("01011111--------", Id::VMAD, Type::Video, "VMAD"),
INST("0101000011110---", Id::VSETP, Type::Video, "VSETP"), INST("0101000011110---", Id::VSETP, Type::Video, "VSETP"),
INST("0011101---------", Id::VMNMX, Type::Video, "VMNMX"),
INST("0011001-1-------", Id::FFMA_IMM, Type::Ffma, "FFMA_IMM"), INST("0011001-1-------", Id::FFMA_IMM, Type::Ffma, "FFMA_IMM"),
INST("010010011-------", Id::FFMA_CR, Type::Ffma, "FFMA_CR"), INST("010010011-------", Id::FFMA_CR, Type::Ffma, "FFMA_CR"),
INST("010100011-------", Id::FFMA_RC, Type::Ffma, "FFMA_RC"), INST("010100011-------", Id::FFMA_RC, Type::Ffma, "FFMA_RC"),

View file

@ -10,16 +10,24 @@
namespace VideoCommon::Shader { namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::Instruction; using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode; using Tegra::Shader::OpCode;
using Tegra::Shader::Pred; using Tegra::Shader::Pred;
using Tegra::Shader::VideoType; using Tegra::Shader::VideoType;
using Tegra::Shader::VmadShr; using Tegra::Shader::VmadShr;
using Tegra::Shader::VmnmxOperation;
using Tegra::Shader::VmnmxType;
u32 ShaderIR::DecodeVideo(NodeBlock& bb, u32 pc) { u32 ShaderIR::DecodeVideo(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]}; const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr); const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::VMNMX) {
DecodeVMNMX(bb, instr);
return pc;
}
const Node op_a = const Node op_a =
GetVideoOperand(GetRegister(instr.gpr8), instr.video.is_byte_chunk_a, instr.video.signed_a, GetVideoOperand(GetRegister(instr.gpr8), instr.video.is_byte_chunk_a, instr.video.signed_a,
instr.video.type_a, instr.video.byte_height_a); instr.video.type_a, instr.video.byte_height_a);
@ -109,4 +117,54 @@ Node ShaderIR::GetVideoOperand(Node op, bool is_chunk, bool is_signed,
} }
} }
void ShaderIR::DecodeVMNMX(NodeBlock& bb, Tegra::Shader::Instruction instr) {
UNIMPLEMENTED_IF(!instr.vmnmx.is_op_b_register);
UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatA() != VmnmxType::Bits32);
UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatB() != VmnmxType::Bits32);
UNIMPLEMENTED_IF(instr.vmnmx.is_src_a_signed != instr.vmnmx.is_src_b_signed);
UNIMPLEMENTED_IF(instr.vmnmx.sat);
UNIMPLEMENTED_IF(instr.generates_cc);
Node op_a = GetRegister(instr.gpr8);
Node op_b = GetRegister(instr.gpr20);
Node op_c = GetRegister(instr.gpr39);
const bool is_oper1_signed = instr.vmnmx.is_src_a_signed; // Stubbed
const bool is_oper2_signed = instr.vmnmx.is_dest_signed;
const auto operation_a = instr.vmnmx.mx ? OperationCode::IMax : OperationCode::IMin;
Node value = SignedOperation(operation_a, is_oper1_signed, move(op_a), move(op_b));
switch (instr.vmnmx.operation) {
case VmnmxOperation::Mrg_16H:
value = BitfieldInsert(move(op_c), move(value), 16, 16);
break;
case VmnmxOperation::Mrg_16L:
value = BitfieldInsert(move(op_c), move(value), 0, 16);
break;
case VmnmxOperation::Mrg_8B0:
value = BitfieldInsert(move(op_c), move(value), 0, 8);
break;
case VmnmxOperation::Mrg_8B2:
value = BitfieldInsert(move(op_c), move(value), 16, 8);
break;
case VmnmxOperation::Acc:
value = Operation(OperationCode::IAdd, move(value), move(op_c));
break;
case VmnmxOperation::Min:
value = SignedOperation(OperationCode::IMin, is_oper2_signed, move(value), move(op_c));
break;
case VmnmxOperation::Max:
value = SignedOperation(OperationCode::IMax, is_oper2_signed, move(value), move(op_c));
break;
case VmnmxOperation::Nop:
break;
default:
UNREACHABLE();
break;
}
SetRegister(bb, instr.gpr0, move(value));
}
} // namespace VideoCommon::Shader } // namespace VideoCommon::Shader

View file

@ -350,6 +350,9 @@ private:
/// Marks the usage of a input or output attribute. /// Marks the usage of a input or output attribute.
void MarkAttributeUsage(Tegra::Shader::Attribute::Index index, u64 element); void MarkAttributeUsage(Tegra::Shader::Attribute::Index index, u64 element);
/// Decodes VMNMX instruction and inserts its code into the passed basic block.
void DecodeVMNMX(NodeBlock& bb, Tegra::Shader::Instruction instr);
void WriteTexInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr, void WriteTexInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
const Node4& components); const Node4& components);