steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-06 17:00:19 +02:00
Code Issues Packages Projects Releases 66 Wiki Activity

OpcodeDispatcher: reassociate and sink W in sha1

Browse Source 
We only need each part of W extracted in the corresponding round, so sink the
extract into the round to reduce pressure.

Further, W and E are added and then never used again. So, by reassociating we
can do the add upfront, killing W and E at the start and further reducing
pressure.

Eliminates spilling in sha1rnds4.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
This commit is contained in:
Alyssa Rosenzweig committed 2024-02-02 13:03:07 -04:00
1 parent 60f7b9bcc4
commit 8d3f0b6f02
1 file changed
+9 -8
Download Patch File Download Diff File Expand all files Collapse all files
FEXCore/Source/Interface/Core/OpcodeDispatcher/Crypto.cpp
+9 -8
View File
@@ -128,9 +128,6 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
auto W0E = _VExtractToGPR(16, 4, Src, 3);
auto W1 = _VExtractToGPR(16, 4, Src, 2);
auto W2 = _VExtractToGPR(16, 4, Src, 1);
auto W3 = _VExtractToGPR(16, 4, Src, 0);
using RoundResult = std::tuple<OrderedNode*, OrderedNode*, OrderedNode*, OrderedNode*, OrderedNode*>;
@@ -149,8 +146,12 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
return {A1, B1, C1, D1, E1};
};
const auto Round1To3 = [&](OrderedNode *A, OrderedNode *B, OrderedNode *C,
OrderedNode *D, OrderedNode *E, OrderedNode *W) -> RoundResult {
auto ANext = _Add(OpSize::i32Bit, _Add(OpSize::i32Bit, _Add(OpSize::i32Bit, _Add(OpSize::i32Bit, Fn(*this, B, C, D), _Ror(OpSize::i32Bit, A, _Constant(32, 27))), W), E), K);
OrderedNode *D, OrderedNode *E, OrderedNode *Src, unsigned W_idx) -> RoundResult {
// Kill W and E at the beginning
auto W = _VExtractToGPR(16, 4, Src, W_idx);
auto Q = _Add(OpSize::i32Bit, W, E);
auto ANext = _Add(OpSize::i32Bit, _Add(OpSize::i32Bit, _Add(OpSize::i32Bit, Fn(*this, B, C, D), _Ror(OpSize::i32Bit, A, _Constant(32, 27))), Q), K);
auto BNext = A;
auto CNext = _Ror(OpSize::i32Bit, B, _Constant(32, 2));
auto DNext = C;
@@ -160,9 +161,9 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
};
auto [A1, B1, C1, D1, E1] = Round0();
auto [A2, B2, C2, D2, E2] = Round1To3(A1, B1, C1, D1, E1, W1);
auto [A3, B3, C3, D3, E3] = Round1To3(A2, B2, C2, D2, E2, W2);
auto Final = Round1To3(A3, B3, C3, D3, E3, W3);
auto [A2, B2, C2, D2, E2] = Round1To3(A1, B1, C1, D1, E1, Src, 2);
auto [A3, B3, C3, D3, E3] = Round1To3(A2, B2, C2, D2, E2, Src, 1);
auto Final = Round1To3(A3, B3, C3, D3, E3, Src, 0);
auto Dest3 = _VInsGPR(16, 4, 3, Dest, std::get<0>(Final));
auto Dest2 = _VInsGPR(16, 4, 2, Dest3, std::get<1>(Final));