summaryrefslogtreecommitdiffstats
path: root/src/shader_recompiler
diff options
context:
space:
mode:
Diffstat (limited to 'src/shader_recompiler')
-rw-r--r--src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp25
-rw-r--r--src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp6
-rw-r--r--src/shader_recompiler/ir_opt/texture_pass.cpp98
-rw-r--r--src/shader_recompiler/shader_info.h4
4 files changed, 116 insertions, 17 deletions
diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp
index b5c08d611..7e8f37563 100644
--- a/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp
+++ b/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp
@@ -13,9 +13,6 @@ namespace Shader::Backend::GLASM {
namespace {
void GetCbuf(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, ScalarU32 offset,
std::string_view size) {
- if (!binding.IsImmediate()) {
- throw NotImplementedException("Indirect constant buffer loading");
- }
const Register ret{ctx.reg_alloc.Define(inst)};
if (offset.type == Type::U32) {
// Avoid reading arrays out of bounds, matching hardware's behavior
@@ -24,7 +21,27 @@ void GetCbuf(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, ScalarU
return;
}
}
- ctx.Add("LDC.{} {},c{}[{}];", size, ret, binding.U32(), offset);
+
+ if (binding.IsImmediate()) {
+ ctx.Add("LDC.{} {},c{}[{}];", size, ret, binding.U32(), offset);
+ return;
+ }
+
+ const ScalarU32 idx{ctx.reg_alloc.Consume(binding)};
+ for (u32 i = 0; i < Info::MAX_INDIRECT_CBUFS; i++) {
+ ctx.Add("SEQ.S.CC RC.x,{},{};"
+ "IF NE.x;"
+ "LDC.{} {},c{}[{}];",
+ idx, i, size, ret, i, offset);
+
+ if (i != Info::MAX_INDIRECT_CBUFS - 1) {
+ ctx.Add("ELSE;");
+ }
+ }
+
+ for (u32 i = 0; i < Info::MAX_INDIRECT_CBUFS; i++) {
+ ctx.Add("ENDIF;");
+ }
}
bool IsInputArray(Stage stage) {
diff --git a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp
index 578bc8c1b..ce42475d4 100644
--- a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp
+++ b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp
@@ -964,9 +964,9 @@ private:
demote_endif_node.type = Type::EndIf;
demote_endif_node.data.end_if.merge = return_block_it->data.block;
- asl.insert(return_block_it, demote_endif_node);
- asl.insert(return_block_it, demote_node);
- asl.insert(return_block_it, demote_if_node);
+ const auto next_it_1 = asl.insert(return_block_it, demote_endif_node);
+ const auto next_it_2 = asl.insert(next_it_1, demote_node);
+ asl.insert(next_it_2, demote_if_node);
}
ObjectPool<Statement>& stmt_pool;
diff --git a/src/shader_recompiler/ir_opt/texture_pass.cpp b/src/shader_recompiler/ir_opt/texture_pass.cpp
index 597112ba4..e8be58357 100644
--- a/src/shader_recompiler/ir_opt/texture_pass.cpp
+++ b/src/shader_recompiler/ir_opt/texture_pass.cpp
@@ -19,8 +19,10 @@ namespace {
struct ConstBufferAddr {
u32 index;
u32 offset;
+ u32 shift_left;
u32 secondary_index;
u32 secondary_offset;
+ u32 secondary_shift_left;
IR::U32 dynamic_offset;
u32 count;
bool has_secondary;
@@ -172,19 +174,41 @@ bool IsTextureInstruction(const IR::Inst& inst) {
return IndexedInstruction(inst) != IR::Opcode::Void;
}
-std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst);
+std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env);
-std::optional<ConstBufferAddr> Track(const IR::Value& value) {
- return IR::BreadthFirstSearch(value, TryGetConstBuffer);
+std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env) {
+ return IR::BreadthFirstSearch(
+ value, [&env](const IR::Inst* inst) { return TryGetConstBuffer(inst, env); });
}
-std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) {
+std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) {
+ const IR::Inst* inst = value.InstRecursive();
+ if (inst->GetOpcode() != IR::Opcode::GetCbufU32) {
+ return std::nullopt;
+ }
+ const IR::Value index{inst->Arg(0)};
+ const IR::Value offset{inst->Arg(1)};
+ if (!index.IsImmediate()) {
+ return std::nullopt;
+ }
+ if (!offset.IsImmediate()) {
+ return std::nullopt;
+ }
+ const auto index_number = index.U32();
+ if (index_number != 1) {
+ return std::nullopt;
+ }
+ const auto offset_number = offset.U32();
+ return env.ReadCbufValue(index_number, offset_number);
+}
+
+std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env) {
switch (inst->GetOpcode()) {
default:
return std::nullopt;
case IR::Opcode::BitwiseOr32: {
- std::optional lhs{Track(inst->Arg(0))};
- std::optional rhs{Track(inst->Arg(1))};
+ std::optional lhs{Track(inst->Arg(0), env)};
+ std::optional rhs{Track(inst->Arg(1), env)};
if (!lhs || !rhs) {
return std::nullopt;
}
@@ -194,19 +218,62 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) {
if (lhs->count > 1 || rhs->count > 1) {
return std::nullopt;
}
- if (lhs->index > rhs->index || lhs->offset > rhs->offset) {
+ if (lhs->shift_left > 0 || lhs->index > rhs->index || lhs->offset > rhs->offset) {
std::swap(lhs, rhs);
}
return ConstBufferAddr{
.index = lhs->index,
.offset = lhs->offset,
+ .shift_left = lhs->shift_left,
.secondary_index = rhs->index,
.secondary_offset = rhs->offset,
+ .secondary_shift_left = rhs->shift_left,
.dynamic_offset = {},
.count = 1,
.has_secondary = true,
};
}
+ case IR::Opcode::ShiftLeftLogical32: {
+ const IR::Value shift{inst->Arg(1)};
+ if (!shift.IsImmediate()) {
+ return std::nullopt;
+ }
+ std::optional lhs{Track(inst->Arg(0), env)};
+ if (lhs) {
+ lhs->shift_left = shift.U32();
+ }
+ return lhs;
+ break;
+ }
+ case IR::Opcode::BitwiseAnd32: {
+ IR::Value op1{inst->Arg(0)};
+ IR::Value op2{inst->Arg(1)};
+ if (op1.IsImmediate()) {
+ std::swap(op1, op2);
+ }
+ if (!op2.IsImmediate() && !op1.IsImmediate()) {
+ do {
+ auto try_index = TryGetConstant(op1, env);
+ if (try_index) {
+ op1 = op2;
+ op2 = IR::Value{*try_index};
+ break;
+ }
+ auto try_index_2 = TryGetConstant(op2, env);
+ if (try_index_2) {
+ op2 = IR::Value{*try_index_2};
+ break;
+ }
+ return std::nullopt;
+ } while (false);
+ }
+ std::optional lhs{Track(op1, env)};
+ if (lhs) {
+ lhs->shift_left = static_cast<u32>(std::countr_zero(op2.U32()));
+ }
+ return lhs;
+ break;
+ }
case IR::Opcode::GetCbufU32x2:
case IR::Opcode::GetCbufU32:
break;
@@ -222,8 +289,10 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) {
return ConstBufferAddr{
.index = index.U32(),
.offset = offset.U32(),
+ .shift_left = 0,
.secondary_index = 0,
.secondary_offset = 0,
+ .secondary_shift_left = 0,
.dynamic_offset = {},
.count = 1,
.has_secondary = false,
@@ -247,8 +316,10 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) {
return ConstBufferAddr{
.index = index.U32(),
.offset = base_offset,
+ .shift_left = 0,
.secondary_index = 0,
.secondary_offset = 0,
+ .secondary_shift_left = 0,
.dynamic_offset = dynamic_offset,
.count = 8,
.has_secondary = false,
@@ -258,7 +329,7 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) {
TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) {
ConstBufferAddr addr;
if (IsBindless(inst)) {
- const std::optional<ConstBufferAddr> track_addr{Track(inst.Arg(0))};
+ const std::optional<ConstBufferAddr> track_addr{Track(inst.Arg(0), env)};
if (!track_addr) {
throw NotImplementedException("Failed to track bindless texture constant buffer");
}
@@ -267,8 +338,10 @@ TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) {
addr = ConstBufferAddr{
.index = env.TextureBoundBuffer(),
.offset = inst.Arg(0).U32(),
+ .shift_left = 0,
.secondary_index = 0,
.secondary_offset = 0,
+ .secondary_shift_left = 0,
.dynamic_offset = {},
.count = 1,
.has_secondary = false,
@@ -284,8 +357,9 @@ TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) {
TextureType ReadTextureType(Environment& env, const ConstBufferAddr& cbuf) {
const u32 secondary_index{cbuf.has_secondary ? cbuf.secondary_index : cbuf.index};
const u32 secondary_offset{cbuf.has_secondary ? cbuf.secondary_offset : cbuf.offset};
- const u32 lhs_raw{env.ReadCbufValue(cbuf.index, cbuf.offset)};
- const u32 rhs_raw{env.ReadCbufValue(secondary_index, secondary_offset)};
+ const u32 lhs_raw{env.ReadCbufValue(cbuf.index, cbuf.offset) << cbuf.shift_left};
+ const u32 rhs_raw{env.ReadCbufValue(secondary_index, secondary_offset)
+ << cbuf.secondary_shift_left};
return env.ReadTextureType(lhs_raw | rhs_raw);
}
@@ -487,8 +561,10 @@ void TexturePass(Environment& env, IR::Program& program) {
.has_secondary = cbuf.has_secondary,
.cbuf_index = cbuf.index,
.cbuf_offset = cbuf.offset,
+ .shift_left = cbuf.shift_left,
.secondary_cbuf_index = cbuf.secondary_index,
.secondary_cbuf_offset = cbuf.secondary_offset,
+ .secondary_shift_left = cbuf.secondary_shift_left,
.count = cbuf.count,
.size_shift = DESCRIPTOR_SIZE_SHIFT,
});
@@ -499,8 +575,10 @@ void TexturePass(Environment& env, IR::Program& program) {
.has_secondary = cbuf.has_secondary,
.cbuf_index = cbuf.index,
.cbuf_offset = cbuf.offset,
+ .shift_left = cbuf.shift_left,
.secondary_cbuf_index = cbuf.secondary_index,
.secondary_cbuf_offset = cbuf.secondary_offset,
+ .secondary_shift_left = cbuf.secondary_shift_left,
.count = cbuf.count,
.size_shift = DESCRIPTOR_SIZE_SHIFT,
});
diff --git a/src/shader_recompiler/shader_info.h b/src/shader_recompiler/shader_info.h
index f5690805c..cc596da4f 100644
--- a/src/shader_recompiler/shader_info.h
+++ b/src/shader_recompiler/shader_info.h
@@ -61,8 +61,10 @@ struct TextureBufferDescriptor {
bool has_secondary;
u32 cbuf_index;
u32 cbuf_offset;
+ u32 shift_left;
u32 secondary_cbuf_index;
u32 secondary_cbuf_offset;
+ u32 secondary_shift_left;
u32 count;
u32 size_shift;
};
@@ -85,8 +87,10 @@ struct TextureDescriptor {
bool has_secondary;
u32 cbuf_index;
u32 cbuf_offset;
+ u32 shift_left;
u32 secondary_cbuf_index;
u32 secondary_cbuf_offset;
+ u32 secondary_shift_left;
u32 count;
u32 size_shift;
};
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-21 09:00:58 +0200