summaryrefslogtreecommitdiffstats
path: root/src/video_core/shader
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/shader')
-rw-r--r--src/video_core/shader/control_flow.cpp47
-rw-r--r--src/video_core/shader/control_flow.h30
-rw-r--r--src/video_core/shader/decode.cpp4
-rw-r--r--src/video_core/shader/decode/arithmetic.cpp13
-rw-r--r--src/video_core/shader/decode/arithmetic_half_immediate.cpp4
-rw-r--r--src/video_core/shader/decode/ffma.cpp10
-rw-r--r--src/video_core/shader/decode/half_set_predicate.cpp8
-rw-r--r--src/video_core/shader/decode/hfma2.cpp4
-rw-r--r--src/video_core/shader/decode/other.cpp6
-rw-r--r--src/video_core/shader/track.cpp4
10 files changed, 82 insertions, 48 deletions
diff --git a/src/video_core/shader/control_flow.cpp b/src/video_core/shader/control_flow.cpp
index fdcc970ff..ec3a76690 100644
--- a/src/video_core/shader/control_flow.cpp
+++ b/src/video_core/shader/control_flow.cpp
@@ -15,7 +15,7 @@
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
-
+namespace {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
@@ -29,8 +29,7 @@ struct Query {
struct BlockStack {
BlockStack() = default;
- BlockStack(const BlockStack& b) = default;
- BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
+ explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
std::stack<u32> ssy_stack{};
std::stack<u32> pbk_stack{};
};
@@ -58,7 +57,7 @@ struct BlockInfo {
struct CFGRebuildState {
explicit CFGRebuildState(const ProgramCode& program_code, const std::size_t program_size,
const u32 start)
- : program_code{program_code}, program_size{program_size}, start{start} {}
+ : start{start}, program_code{program_code}, program_size{program_size} {}
u32 start{};
std::vector<BlockInfo> block_info{};
@@ -85,7 +84,7 @@ std::pair<BlockCollision, u32> TryGetBlock(CFGRebuildState& state, u32 address)
return {BlockCollision::Inside, index};
}
}
- return {BlockCollision::None, -1};
+ return {BlockCollision::None, 0xFFFFFFFF};
}
struct ParseInfo {
@@ -365,27 +364,29 @@ bool TryQuery(CFGRebuildState& state) {
const auto gather_end = labels.upper_bound(block.end);
while (gather_start != gather_end) {
cc.push(gather_start->second);
- gather_start++;
+ ++gather_start;
}
};
if (state.queries.empty()) {
return false;
}
+
Query& q = state.queries.front();
const u32 block_index = state.registered[q.address];
BlockInfo& block = state.block_info[block_index];
- // If the block is visted, check if the stacks match, else gather the ssy/pbk
+ // If the block is visited, check if the stacks match, else gather the ssy/pbk
// labels into the current stack and look if the branch at the end of the block
// consumes a label. Schedule new queries accordingly
if (block.visited) {
BlockStack& stack = state.stacks[q.address];
- const bool all_okay = (stack.ssy_stack.size() == 0 || q.ssy_stack == stack.ssy_stack) &&
- (stack.pbk_stack.size() == 0 || q.pbk_stack == stack.pbk_stack);
+ const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) &&
+ (stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack);
state.queries.pop_front();
return all_okay;
}
block.visited = true;
- state.stacks[q.address] = BlockStack{q};
+ state.stacks.insert_or_assign(q.address, BlockStack{q});
+
Query q2(q);
state.queries.pop_front();
gather_labels(q2.ssy_stack, state.ssy_labels, block);
@@ -394,6 +395,7 @@ bool TryQuery(CFGRebuildState& state) {
q2.address = block.end + 1;
state.queries.push_back(q2);
}
+
Query conditional_query{q2};
if (block.branch.is_sync) {
if (block.branch.address == unassigned_branch) {
@@ -408,13 +410,15 @@ bool TryQuery(CFGRebuildState& state) {
conditional_query.pbk_stack.pop();
}
conditional_query.address = block.branch.address;
- state.queries.push_back(conditional_query);
+ state.queries.push_back(std::move(conditional_query));
return true;
}
+} // Anonymous namespace
-std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size,
- u32 start_address) {
+std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
+ std::size_t program_size, u32 start_address) {
CFGRebuildState state{program_code, program_size, start_address};
+
// Inspect Code and generate blocks
state.labels.clear();
state.labels.emplace(start_address);
@@ -424,10 +428,9 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
return {};
}
}
+
// Decompile Stacks
- Query start_query{};
- start_query.address = state.start;
- state.queries.push_back(start_query);
+ state.queries.push_back(Query{state.start, {}, {}});
bool decompiled = true;
while (!state.queries.empty()) {
if (!TryQuery(state)) {
@@ -435,14 +438,15 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
break;
}
}
+
// Sort and organize results
std::sort(state.block_info.begin(), state.block_info.end(),
- [](const BlockInfo& a, const BlockInfo& b) -> bool { return a.start < b.start; });
+ [](const BlockInfo& a, const BlockInfo& b) { return a.start < b.start; });
ShaderCharacteristics result_out{};
result_out.decompilable = decompiled;
result_out.start = start_address;
result_out.end = start_address;
- for (auto& block : state.block_info) {
+ for (const auto& block : state.block_info) {
ShaderBlock new_block{};
new_block.start = block.start;
new_block.end = block.end;
@@ -457,8 +461,9 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
}
if (result_out.decompilable) {
result_out.labels = std::move(state.labels);
- return {result_out};
+ return {std::move(result_out)};
}
+
// If it's not decompilable, merge the unlabelled blocks together
auto back = result_out.blocks.begin();
auto next = std::next(back);
@@ -469,8 +474,8 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
continue;
}
back = next;
- next++;
+ ++next;
}
- return {result_out};
+ return {std::move(result_out)};
}
} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/control_flow.h b/src/video_core/shader/control_flow.h
index 5e8ea3271..b0a5e4f8c 100644
--- a/src/video_core/shader/control_flow.h
+++ b/src/video_core/shader/control_flow.h
@@ -4,7 +4,6 @@
#pragma once
-#include <cstring>
#include <list>
#include <optional>
#include <unordered_set>
@@ -26,27 +25,44 @@ struct Condition {
bool IsUnconditional() const {
return predicate == Pred::UnusedIndex && cc == ConditionCode::T;
}
+
bool operator==(const Condition& other) const {
return std::tie(predicate, cc) == std::tie(other.predicate, other.cc);
}
+
+ bool operator!=(const Condition& other) const {
+ return !operator==(other);
+ }
};
struct ShaderBlock {
- u32 start{};
- u32 end{};
- bool ignore_branch{};
struct Branch {
Condition cond{};
bool kills{};
s32 address{};
+
bool operator==(const Branch& b) const {
return std::tie(cond, kills, address) == std::tie(b.cond, b.kills, b.address);
}
- } branch{};
+
+ bool operator!=(const Branch& b) const {
+ return !operator==(b);
+ }
+ };
+
+ u32 start{};
+ u32 end{};
+ bool ignore_branch{};
+ Branch branch{};
+
bool operator==(const ShaderBlock& sb) const {
return std::tie(start, end, ignore_branch, branch) ==
std::tie(sb.start, sb.end, sb.ignore_branch, sb.branch);
}
+
+ bool operator!=(const ShaderBlock& sb) const {
+ return !operator==(sb);
+ }
};
struct ShaderCharacteristics {
@@ -57,7 +73,7 @@ struct ShaderCharacteristics {
std::unordered_set<u32> labels{};
};
-std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size,
- u32 start_address);
+std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
+ std::size_t program_size, u32 start_address);
} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode.cpp b/src/video_core/shader/decode.cpp
index afffd157f..b547d8323 100644
--- a/src/video_core/shader/decode.cpp
+++ b/src/video_core/shader/decode.cpp
@@ -47,14 +47,14 @@ void ShaderIR::Decode() {
if (shader_info.decompilable) {
disable_flow_stack = true;
const auto insert_block = [this](NodeBlock& nodes, u32 label) {
- if (label == exit_branch) {
+ if (label == static_cast<u32>(exit_branch)) {
return;
}
basic_blocks.insert({label, nodes});
};
const auto& blocks = shader_info.blocks;
NodeBlock current_block;
- u32 current_label = exit_branch;
+ u32 current_label = static_cast<u32>(exit_branch);
for (auto& block : blocks) {
if (shader_info.labels.count(block.start) != 0) {
insert_block(current_block, current_label);
diff --git a/src/video_core/shader/decode/arithmetic.cpp b/src/video_core/shader/decode/arithmetic.cpp
index 87d8fecaa..1473c282a 100644
--- a/src/video_core/shader/decode/arithmetic.cpp
+++ b/src/video_core/shader/decode/arithmetic.cpp
@@ -42,11 +42,14 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
case OpCode::Id::FMUL_R:
case OpCode::Id::FMUL_IMM: {
// FMUL does not have 'abs' bits and only the second operand has a 'neg' bit.
- UNIMPLEMENTED_IF_MSG(instr.fmul.tab5cb8_2 != 0, "FMUL tab5cb8_2({}) is not implemented",
- instr.fmul.tab5cb8_2.Value());
- UNIMPLEMENTED_IF_MSG(
- instr.fmul.tab5c68_0 != 1, "FMUL tab5cb8_0({}) is not implemented",
- instr.fmul.tab5c68_0.Value()); // SMO typical sends 1 here which seems to be the default
+ if (instr.fmul.tab5cb8_2 != 0) {
+ LOG_WARNING(HW_GPU, "FMUL tab5cb8_2({}) is not implemented",
+ instr.fmul.tab5cb8_2.Value());
+ }
+ if (instr.fmul.tab5c68_0 != 1) {
+ LOG_WARNING(HW_GPU, "FMUL tab5cb8_0({}) is not implemented",
+ instr.fmul.tab5c68_0.Value());
+ }
op_b = GetOperandAbsNegFloat(op_b, false, instr.fmul.negate_b);
diff --git a/src/video_core/shader/decode/arithmetic_half_immediate.cpp b/src/video_core/shader/decode/arithmetic_half_immediate.cpp
index 7bcf38f23..6466fc011 100644
--- a/src/video_core/shader/decode/arithmetic_half_immediate.cpp
+++ b/src/video_core/shader/decode/arithmetic_half_immediate.cpp
@@ -23,7 +23,9 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) {
LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName());
}
} else {
- UNIMPLEMENTED_IF(instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None);
+ if (instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None) {
+ LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName());
+ }
}
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half_imm.type_a);
diff --git a/src/video_core/shader/decode/ffma.cpp b/src/video_core/shader/decode/ffma.cpp
index 29be25ca3..ca2f39e8d 100644
--- a/src/video_core/shader/decode/ffma.cpp
+++ b/src/video_core/shader/decode/ffma.cpp
@@ -18,10 +18,12 @@ u32 ShaderIR::DecodeFfma(NodeBlock& bb, u32 pc) {
const auto opcode = OpCode::Decode(instr);
UNIMPLEMENTED_IF_MSG(instr.ffma.cc != 0, "FFMA cc not implemented");
- UNIMPLEMENTED_IF_MSG(instr.ffma.tab5980_0 != 1, "FFMA tab5980_0({}) not implemented",
- instr.ffma.tab5980_0.Value()); // Seems to be 1 by default based on SMO
- UNIMPLEMENTED_IF_MSG(instr.ffma.tab5980_1 != 0, "FFMA tab5980_1({}) not implemented",
- instr.ffma.tab5980_1.Value());
+ if (instr.ffma.tab5980_0 != 1) {
+ LOG_WARNING(HW_GPU, "FFMA tab5980_0({}) not implemented", instr.ffma.tab5980_0.Value());
+ }
+ if (instr.ffma.tab5980_1 != 0) {
+ LOG_WARNING(HW_GPU, "FFMA tab5980_1({}) not implemented", instr.ffma.tab5980_1.Value());
+ }
const Node op_a = GetRegister(instr.gpr8);
diff --git a/src/video_core/shader/decode/half_set_predicate.cpp b/src/video_core/shader/decode/half_set_predicate.cpp
index ad180d6df..a6c082cc9 100644
--- a/src/video_core/shader/decode/half_set_predicate.cpp
+++ b/src/video_core/shader/decode/half_set_predicate.cpp
@@ -18,7 +18,7 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
- UNIMPLEMENTED_IF(instr.hsetp2.ftz != 0);
+ DEBUG_ASSERT(instr.hsetp2.ftz == 0);
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hsetp2.type_a);
op_a = GetOperandAbsNegHalf(op_a, instr.hsetp2.abs_a, instr.hsetp2.negate_a);
@@ -52,15 +52,15 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) {
}
const OperationCode combiner = GetPredicateCombiner(instr.hsetp2.op);
- const Node pred39 = GetPredicate(instr.hsetp2.pred39, instr.hsetp2.neg_pred);
+ const Node combined_pred = GetPredicate(instr.hsetp2.pred3, instr.hsetp2.neg_pred);
const auto Write = [&](u64 dest, Node src) {
- SetPredicate(bb, dest, Operation(combiner, std::move(src), pred39));
+ SetPredicate(bb, dest, Operation(combiner, std::move(src), combined_pred));
};
const Node comparison = GetPredicateComparisonHalf(cond, op_a, op_b);
const u64 first = instr.hsetp2.pred0;
- const u64 second = instr.hsetp2.pred3;
+ const u64 second = instr.hsetp2.pred39;
if (h_and) {
const Node joined = Operation(OperationCode::LogicalAnd2, comparison);
Write(first, joined);
diff --git a/src/video_core/shader/decode/hfma2.cpp b/src/video_core/shader/decode/hfma2.cpp
index c3bcf1ae9..5b44cb79c 100644
--- a/src/video_core/shader/decode/hfma2.cpp
+++ b/src/video_core/shader/decode/hfma2.cpp
@@ -22,9 +22,9 @@ u32 ShaderIR::DecodeHfma2(NodeBlock& bb, u32 pc) {
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::HFMA2_RR) {
- UNIMPLEMENTED_IF(instr.hfma2.rr.precision != HalfPrecision::None);
+ DEBUG_ASSERT(instr.hfma2.rr.precision == HalfPrecision::None);
} else {
- UNIMPLEMENTED_IF(instr.hfma2.precision != HalfPrecision::None);
+ DEBUG_ASSERT(instr.hfma2.precision == HalfPrecision::None);
}
constexpr auto identity = HalfType::H0_H1;
diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp
index c0f64d7a0..ac0e764d6 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -22,6 +22,12 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
const auto opcode = OpCode::Decode(instr);
switch (opcode->get().GetId()) {
+ case OpCode::Id::NOP: {
+ UNIMPLEMENTED_IF(instr.nop.cc != Tegra::Shader::ConditionCode::T);
+ UNIMPLEMENTED_IF(instr.nop.trigger != 0);
+ // With the previous preconditions, this instruction is a no-operation.
+ break;
+ }
case OpCode::Id::EXIT: {
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "EXIT condition code used: {}",
diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp
index a53e02253..55f5949e4 100644
--- a/src/video_core/shader/track.cpp
+++ b/src/video_core/shader/track.cpp
@@ -59,8 +59,8 @@ std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& co
return TrackCbuf(source, code, new_cursor);
}
if (const auto operation = std::get_if<OperationNode>(&*tracked)) {
- for (std::size_t i = 0; i < operation->GetOperandsCount(); ++i) {
- if (auto found = TrackCbuf((*operation)[i], code, cursor); std::get<0>(found)) {
+ for (std::size_t i = operation->GetOperandsCount(); i > 0; --i) {
+ if (auto found = TrackCbuf((*operation)[i - 1], code, cursor); std::get<0>(found)) {
// Cbuf found in operand.
return found;
}
generated by cgit v1.2.3 (git 2.25.1) at 2024-09-13 08:35:27 +0200