summaryrefslogtreecommitdiffstats
path: root/src/video_core/shader
diff options
context:
space:
mode:
authorbunnei <bunneidev@gmail.com>2019-09-20 23:10:42 +0200
committerGitHub <noreply@github.com>2019-09-20 23:10:42 +0200
commit88d857499b6168d7bcea9b91fa5bdd8b0144c07a (patch)
tree1efa1eddb43c0afd3003b449675997abc9a7637f /src/video_core/shader
parentMerge pull request #2784 from ReinUsesLisp/smem (diff)
parentshader_ir/warp: Implement SHFL (diff)
downloadyuzu-88d857499b6168d7bcea9b91fa5bdd8b0144c07a.tar
yuzu-88d857499b6168d7bcea9b91fa5bdd8b0144c07a.tar.gz
yuzu-88d857499b6168d7bcea9b91fa5bdd8b0144c07a.tar.bz2
yuzu-88d857499b6168d7bcea9b91fa5bdd8b0144c07a.tar.lz
yuzu-88d857499b6168d7bcea9b91fa5bdd8b0144c07a.tar.xz
yuzu-88d857499b6168d7bcea9b91fa5bdd8b0144c07a.tar.zst
yuzu-88d857499b6168d7bcea9b91fa5bdd8b0144c07a.zip
Diffstat (limited to 'src/video_core/shader')
-rw-r--r--src/video_core/shader/decode/warp.cpp47
-rw-r--r--src/video_core/shader/node.h10
2 files changed, 57 insertions, 0 deletions
diff --git a/src/video_core/shader/decode/warp.cpp b/src/video_core/shader/decode/warp.cpp
index 04ca74f46..a8e481b3c 100644
--- a/src/video_core/shader/decode/warp.cpp
+++ b/src/video_core/shader/decode/warp.cpp
@@ -13,6 +13,7 @@ namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
+using Tegra::Shader::ShuffleOperation;
using Tegra::Shader::VoteOperation;
namespace {
@@ -44,6 +45,52 @@ u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
SetPredicate(bb, instr.vote.dest_pred, vote);
break;
}
+ case OpCode::Id::SHFL: {
+ Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm))
+ : GetRegister(instr.gpr39);
+ Node width = [&] {
+ // Convert the obscure SHFL mask back into GL_NV_shader_thread_shuffle's width. This has
+ // been done reversing Nvidia's math. It won't work on all cases due to SHFL having
+ // different parameters that don't properly map to GLSL's interface, but it should work
+ // for cases emitted by Nvidia's compiler.
+ if (instr.shfl.operation == ShuffleOperation::Up) {
+ return Operation(
+ OperationCode::ILogicalShiftRight,
+ Operation(OperationCode::IAdd, std::move(mask), Immediate(-0x2000)),
+ Immediate(8));
+ } else {
+ return Operation(OperationCode::ILogicalShiftRight,
+ Operation(OperationCode::IAdd, Immediate(0x201F),
+ Operation(OperationCode::INegate, std::move(mask))),
+ Immediate(8));
+ }
+ }();
+
+ const auto [operation, in_range] = [instr]() -> std::pair<OperationCode, OperationCode> {
+ switch (instr.shfl.operation) {
+ case ShuffleOperation::Idx:
+ return {OperationCode::ShuffleIndexed, OperationCode::InRangeShuffleIndexed};
+ case ShuffleOperation::Up:
+ return {OperationCode::ShuffleUp, OperationCode::InRangeShuffleUp};
+ case ShuffleOperation::Down:
+ return {OperationCode::ShuffleDown, OperationCode::InRangeShuffleDown};
+ case ShuffleOperation::Bfly:
+ return {OperationCode::ShuffleButterfly, OperationCode::InRangeShuffleButterfly};
+ }
+ UNREACHABLE_MSG("Invalid SHFL operation: {}",
+ static_cast<u64>(instr.shfl.operation.Value()));
+ return {};
+ }();
+
+ // Setting the predicate before the register is intentional to avoid overwriting.
+ Node index = instr.shfl.is_index_imm ? Immediate(static_cast<u32>(instr.shfl.index_imm))
+ : GetRegister(instr.gpr20);
+ SetPredicate(bb, instr.shfl.pred48, Operation(in_range, index, width));
+ SetRegister(
+ bb, instr.gpr0,
+ Operation(operation, GetRegister(instr.gpr8), std::move(index), std::move(width)));
+ break;
+ }
default:
UNIMPLEMENTED_MSG("Unhandled warp instruction: {}", opcode->get().GetName());
break;
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index 425111cc4..abf2cb1ab 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -181,6 +181,16 @@ enum class OperationCode {
VoteAny, /// (bool) -> bool
VoteEqual, /// (bool) -> bool
+ ShuffleIndexed, /// (uint value, uint index, uint width) -> uint
+ ShuffleUp, /// (uint value, uint index, uint width) -> uint
+ ShuffleDown, /// (uint value, uint index, uint width) -> uint
+ ShuffleButterfly, /// (uint value, uint index, uint width) -> uint
+
+ InRangeShuffleIndexed, /// (uint index, uint width) -> bool
+ InRangeShuffleUp, /// (uint index, uint width) -> bool
+ InRangeShuffleDown, /// (uint index, uint width) -> bool
+ InRangeShuffleButterfly, /// (uint index, uint width) -> bool
+
Amount,
};