H3cJP/citra
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

shader: implement LOP3 fast replace for old function

Browse Source 
ref: https://devtalk.nvidia.com/default/topic/1070081/cuda-programming-and-performance/reverse-lut-for-lop3-lut/
This commit is contained in:
Nguyen Dac Nam 2020-02-21 19:08:07 +07:00 committed by GitHub
parent fe8e5d8ae4
commit 1956a34ee5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 58 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

94
src/video_core/shader/decode/arithmetic_integer.cpp
View File

@ -293,44 +293,66 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
void ShaderIR::WriteLop3Instruction(NodeBlock& bb, Register dest, Node op_a, Node op_b, Node op_c, void ShaderIR::WriteLop3Instruction(NodeBlock& bb, Register dest, Node op_a, Node op_b, Node op_c,
Node imm_lut, bool sets_cc) { Node imm_lut, bool sets_cc) {
constexpr u32 lop_iterations = 32; const Node lop3_fast = [&](Node na, Node nb, Node nc, Node ttbl) {
const Node one = Immediate(1); Node value = Immediate(0);
const Node two = Immediate(2); ImmediateNode imm = std::get<ImmediateNode>(*ttbl);
if (imm.GetValue() & 0x01) {
Node value; Node a = Operation(OperationCode::IBitwiseNot, na);
for (u32 i = 0; i < lop_iterations; ++i) { Node b = Operation(OperationCode::IBitwiseNot, nb);
const Node shift_amount = Immediate(i); Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, b);
const Node a = Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, op_c, shift_amount); r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
const Node pack_0 = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, one); value = Operation(OperationCode::IBitwiseOr, value, r);
const Node b = Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, op_b, shift_amount);
const Node c = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, b, one);
const Node pack_1 = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, c, one);
const Node d = Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, op_a, shift_amount);
const Node e = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, d, one);
const Node pack_2 = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, e, two);
const Node pack_01 = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, pack_0, pack_1);
const Node pack_012 = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, pack_01, pack_2);
const Node shifted_bit =
Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, imm_lut, pack_012);
const Node bit = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, shifted_bit, one);
const Node right =
Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, bit, shift_amount);
if (i > 0) {
value = Operation(OperationCode::IBitwiseOr, NO_PRECISE, value, right);
} else {
value = right;
} }
} if (imm.GetValue() & 0x02) {
Node a = Operation(OperationCode::IBitwiseNot, na);
Node b = Operation(OperationCode::IBitwiseNot, nb);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, b);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x04) {
Node a = Operation(OperationCode::IBitwiseNot, na);
Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x08) {
Node a = Operation(OperationCode::IBitwiseNot, na);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x10) {
Node b = Operation(OperationCode::IBitwiseNot, nb);
Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, b);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x20) {
Node b = Operation(OperationCode::IBitwiseNot, nb);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, b);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x40) {
Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x80) {
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
return value;
}(op_a, op_b, op_c, imm_lut);
SetInternalFlagsFromInteger(bb, value, sets_cc); SetInternalFlagsFromInteger(bb, lop3_fast, sets_cc);
SetRegister(bb, dest, value); SetRegister(bb, dest, lop3_fast);
} }
} // namespace VideoCommon::Shader } // namespace VideoCommon::Shader