CCF_100/yuzu-mirror
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

shader: Split SSY and PBK stack

Browse Source 
Hardware testing revealed that SSY and PBK push to a different stack,
allowing code like this:

        SSY label1;
        PBK label2;
        SYNC;
label1: PBK;
label2: EXIT;
This commit is contained in:
ReinUsesLisp
2019-06-02 18:52:07 -03:00
parent cd2d9628c9
commit fe8e6618f2
4 changed files with 78 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
src/video_core/shader/decode/other.cpp
View File

@@ -109,22 +109,20 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
"Constant buffer flow is not supported");
// The SSY opcode tells the GPU where to re-converge divergent execution paths, it sets the
// target of the jump that the SYNC instruction will make. The SSY opcode has a similar
// structure to the BRA opcode.
// The SSY opcode tells the GPU where to re-converge divergent execution paths with SYNC.
const u32 target = pc + instr.bra.GetBranchTarget();
bb.push_back(Operation(OperationCode::PushFlowStack, Immediate(target)));
bb.push_back(
Operation(OperationCode::PushFlowStack, MetaStackClass::Ssy, Immediate(target)));
break;
}
case OpCode::Id::PBK: {
UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
"Constant buffer PBK is not supported");
// PBK pushes to a stack the address where BRK will jump to. This shares stack with SSY but
// using SYNC on a PBK address will kill the shader execution. We don't emulate this because
// it's very unlikely a driver will emit such invalid shader.
// PBK pushes to a stack the address where BRK will jump to.
const u32 target = pc + instr.bra.GetBranchTarget();
bb.push_back(Operation(OperationCode::PushFlowStack, Immediate(target)));
bb.push_back(
Operation(OperationCode::PushFlowStack, MetaStackClass::Pbk, Immediate(target)));
break;
}
case OpCode::Id::SYNC: {
@@ -133,7 +131,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
static_cast<u32>(cc));
// The SYNC opcode jumps to the address previously set by the SSY opcode
bb.push_back(Operation(OperationCode::PopFlowStack));
bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Ssy));
break;
}
case OpCode::Id::BRK: {
@@ -142,7 +140,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
static_cast<u32>(cc));
// The BRK opcode jumps to the address previously set by the PBK opcode
bb.push_back(Operation(OperationCode::PopFlowStack));
bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Pbk));
break;
}
case OpCode::Id::IPA: {

7
src/video_core/shader/node.h
View File

@@ -174,6 +174,11 @@ enum class InternalFlag {
Amount = 4,
};
enum class MetaStackClass {
Ssy,
Pbk,
};
class OperationNode;
class ConditionalNode;
class GprNode;
@@ -285,7 +290,7 @@ struct MetaTexture {
};
/// Parameters that modify an operation but are not part of any particular operand
using Meta = std::variant<MetaArithmetic, MetaTexture, Tegra::Shader::HalfType>;
using Meta = std::variant<MetaArithmetic, MetaTexture, MetaStackClass, Tegra::Shader::HalfType>;
/// Holds any kind of operation that can be done in the IR
class OperationNode final {