summaryrefslogtreecommitdiffstats
path: root/src/video_core/shader/decode
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/shader/decode')
-rw-r--r--src/video_core/shader/decode/arithmetic.cpp3
-rw-r--r--src/video_core/shader/decode/arithmetic_half.cpp1
-rw-r--r--src/video_core/shader/decode/arithmetic_half_immediate.cpp3
-rw-r--r--src/video_core/shader/decode/arithmetic_immediate.cpp2
-rw-r--r--src/video_core/shader/decode/arithmetic_integer_immediate.cpp2
-rw-r--r--src/video_core/shader/decode/bfe.cpp2
-rw-r--r--src/video_core/shader/decode/bfi.cpp2
-rw-r--r--src/video_core/shader/decode/ffma.cpp2
-rw-r--r--src/video_core/shader/decode/float_set.cpp2
-rw-r--r--src/video_core/shader/decode/float_set_predicate.cpp2
-rw-r--r--src/video_core/shader/decode/half_set.cpp3
-rw-r--r--src/video_core/shader/decode/half_set_predicate.cpp2
-rw-r--r--src/video_core/shader/decode/integer_set.cpp3
-rw-r--r--src/video_core/shader/decode/integer_set_predicate.cpp2
-rw-r--r--src/video_core/shader/decode/memory.cpp84
-rw-r--r--src/video_core/shader/decode/other.cpp1
-rw-r--r--src/video_core/shader/decode/predicate_set_predicate.cpp2
-rw-r--r--src/video_core/shader/decode/predicate_set_register.cpp2
-rw-r--r--src/video_core/shader/decode/register_set_predicate.cpp2
-rw-r--r--src/video_core/shader/decode/shift.cpp2
-rw-r--r--src/video_core/shader/decode/video.cpp2
21 files changed, 78 insertions, 48 deletions
diff --git a/src/video_core/shader/decode/arithmetic.cpp b/src/video_core/shader/decode/arithmetic.cpp
index 3190e2d7c..b4859bc1e 100644
--- a/src/video_core/shader/decode/arithmetic.cpp
+++ b/src/video_core/shader/decode/arithmetic.cpp
@@ -4,6 +4,7 @@
#include "common/assert.h"
#include "common/common_types.h"
+#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/shader_ir.h"
@@ -152,4 +153,4 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/arithmetic_half.cpp b/src/video_core/shader/decode/arithmetic_half.cpp
index 2098c1170..3a29c4a46 100644
--- a/src/video_core/shader/decode/arithmetic_half.cpp
+++ b/src/video_core/shader/decode/arithmetic_half.cpp
@@ -4,6 +4,7 @@
#include "common/assert.h"
#include "common/common_types.h"
+#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/shader_ir.h"
diff --git a/src/video_core/shader/decode/arithmetic_half_immediate.cpp b/src/video_core/shader/decode/arithmetic_half_immediate.cpp
index fbcd35b18..5341e460f 100644
--- a/src/video_core/shader/decode/arithmetic_half_immediate.cpp
+++ b/src/video_core/shader/decode/arithmetic_half_immediate.cpp
@@ -4,6 +4,7 @@
#include "common/assert.h"
#include "common/common_types.h"
+#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/shader_ir.h"
@@ -47,4 +48,4 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/arithmetic_immediate.cpp b/src/video_core/shader/decode/arithmetic_immediate.cpp
index 0d139c0d2..3095f2fd4 100644
--- a/src/video_core/shader/decode/arithmetic_immediate.cpp
+++ b/src/video_core/shader/decode/arithmetic_immediate.cpp
@@ -49,4 +49,4 @@ u32 ShaderIR::DecodeArithmeticImmediate(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/arithmetic_integer_immediate.cpp b/src/video_core/shader/decode/arithmetic_integer_immediate.cpp
index 3ed5ccc5a..679ac0d4e 100644
--- a/src/video_core/shader/decode/arithmetic_integer_immediate.cpp
+++ b/src/video_core/shader/decode/arithmetic_integer_immediate.cpp
@@ -93,4 +93,4 @@ void ShaderIR::WriteLogicOperation(NodeBlock& bb, Register dest, LogicOperation
}
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/bfe.cpp b/src/video_core/shader/decode/bfe.cpp
index 6a95dc928..1ae192c6a 100644
--- a/src/video_core/shader/decode/bfe.cpp
+++ b/src/video_core/shader/decode/bfe.cpp
@@ -46,4 +46,4 @@ u32 ShaderIR::DecodeBfe(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/bfi.cpp b/src/video_core/shader/decode/bfi.cpp
index 601d66f1f..0b12a0d08 100644
--- a/src/video_core/shader/decode/bfi.cpp
+++ b/src/video_core/shader/decode/bfi.cpp
@@ -38,4 +38,4 @@ u32 ShaderIR::DecodeBfi(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/ffma.cpp b/src/video_core/shader/decode/ffma.cpp
index 0559cc8de..a1d04c6e5 100644
--- a/src/video_core/shader/decode/ffma.cpp
+++ b/src/video_core/shader/decode/ffma.cpp
@@ -56,4 +56,4 @@ u32 ShaderIR::DecodeFfma(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/float_set.cpp b/src/video_core/shader/decode/float_set.cpp
index 1bd6755dd..cc522f1de 100644
--- a/src/video_core/shader/decode/float_set.cpp
+++ b/src/video_core/shader/decode/float_set.cpp
@@ -55,4 +55,4 @@ u32 ShaderIR::DecodeFloatSet(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/float_set_predicate.cpp b/src/video_core/shader/decode/float_set_predicate.cpp
index 9285b8d05..9d2322a1d 100644
--- a/src/video_core/shader/decode/float_set_predicate.cpp
+++ b/src/video_core/shader/decode/float_set_predicate.cpp
@@ -53,4 +53,4 @@ u32 ShaderIR::DecodeFloatSetPredicate(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/half_set.cpp b/src/video_core/shader/decode/half_set.cpp
index 1dd94bf9d..755f2ec44 100644
--- a/src/video_core/shader/decode/half_set.cpp
+++ b/src/video_core/shader/decode/half_set.cpp
@@ -6,6 +6,7 @@
#include "common/assert.h"
#include "common/common_types.h"
+#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/shader_ir.h"
@@ -64,4 +65,4 @@ u32 ShaderIR::DecodeHalfSet(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/half_set_predicate.cpp b/src/video_core/shader/decode/half_set_predicate.cpp
index 6e59eb650..fba44d714 100644
--- a/src/video_core/shader/decode/half_set_predicate.cpp
+++ b/src/video_core/shader/decode/half_set_predicate.cpp
@@ -59,4 +59,4 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/integer_set.cpp b/src/video_core/shader/decode/integer_set.cpp
index a3bf17eba..a4cdaf74d 100644
--- a/src/video_core/shader/decode/integer_set.cpp
+++ b/src/video_core/shader/decode/integer_set.cpp
@@ -2,7 +2,6 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
-#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/shader_ir.h"
@@ -47,4 +46,4 @@ u32 ShaderIR::DecodeIntegerSet(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/integer_set_predicate.cpp b/src/video_core/shader/decode/integer_set_predicate.cpp
index aad836d24..a6a1fb632 100644
--- a/src/video_core/shader/decode/integer_set_predicate.cpp
+++ b/src/video_core/shader/decode/integer_set_predicate.cpp
@@ -50,4 +50,4 @@ u32 ShaderIR::DecodeIntegerSetPredicate(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index 6a992c543..e6a010a7d 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -12,8 +12,6 @@
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/shader_ir.h"
-#pragma optimize("", off)
-
namespace VideoCommon::Shader {
using Tegra::Shader::Attribute;
@@ -148,12 +146,25 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
}
break;
}
+ case OpCode::Id::LD:
case OpCode::Id::LDG: {
+ const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType {
+ switch (opcode->get().GetId()) {
+ case OpCode::Id::LD:
+ UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended LD is not implemented");
+ return instr.generic.type;
+ case OpCode::Id::LDG:
+ return instr.ldg.type;
+ default:
+ UNREACHABLE();
+ return {};
+ }
+ }();
+
const auto [real_address_base, base_address, descriptor] =
- TrackAndGetGlobalMemory(bb, GetRegister(instr.gpr8),
- static_cast<u32>(instr.ldg.immediate_offset.Value()), false);
+ TrackAndGetGlobalMemory(bb, instr, false);
- const u32 count = GetUniformTypeElementsCount(instr.ldg.type);
+ const u32 count = GetUniformTypeElementsCount(type);
for (u32 i = 0; i < count; ++i) {
const Node it_offset = Immediate(i * 4);
const Node real_address =
@@ -167,28 +178,6 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
}
break;
}
- case OpCode::Id::STG: {
- const auto [real_address_base, base_address, descriptor] =
- TrackAndGetGlobalMemory(bb, GetRegister(instr.gpr8),
- static_cast<u32>(instr.stg.immediate_offset.Value()), true);
-
- // Encode in temporary registers like this: real_base_address, {registers_to_be_written...}
- SetTemporal(bb, 0, real_address_base);
-
- const u32 count = GetUniformTypeElementsCount(instr.stg.type);
- for (u32 i = 0; i < count; ++i) {
- SetTemporal(bb, i + 1, GetRegister(instr.gpr0.Value() + i));
- }
- for (u32 i = 0; i < count; ++i) {
- const Node it_offset = Immediate(i * 4);
- const Node real_address =
- Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
- const Node gmem = StoreNode(GmemNode(real_address, base_address, descriptor));
-
- bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporal(i + 1)));
- }
- break;
- }
case OpCode::Id::ST_A: {
UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex,
"Indirect attribute loads are not supported");
@@ -244,6 +233,41 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
}
break;
}
+ case OpCode::Id::ST:
+ case OpCode::Id::STG: {
+ const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType {
+ switch (opcode->get().GetId()) {
+ case OpCode::Id::ST:
+ UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended ST is not implemented");
+ return instr.generic.type;
+ case OpCode::Id::STG:
+ return instr.stg.type;
+ default:
+ UNREACHABLE();
+ return {};
+ }
+ }();
+
+ const auto [real_address_base, base_address, descriptor] =
+ TrackAndGetGlobalMemory(bb, instr, true);
+
+ // Encode in temporary registers like this: real_base_address, {registers_to_be_written...}
+ SetTemporal(bb, 0, real_address_base);
+
+ const u32 count = GetUniformTypeElementsCount(type);
+ for (u32 i = 0; i < count; ++i) {
+ SetTemporal(bb, i + 1, GetRegister(instr.gpr0.Value() + i));
+ }
+ for (u32 i = 0; i < count; ++i) {
+ const Node it_offset = Immediate(i * 4);
+ const Node real_address =
+ Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
+ const Node gmem = StoreNode(GmemNode(real_address, base_address, descriptor));
+
+ bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporal(i + 1)));
+ }
+ break;
+ }
case OpCode::Id::AL2P: {
// Ignore al2p.direction since we don't care about it.
@@ -267,9 +291,11 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
}
std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackAndGetGlobalMemory(NodeBlock& bb,
- Node addr_register,
- u32 immediate_offset,
+ Instruction instr,
bool is_write) {
+ const auto addr_register{GetRegister(instr.gmem.gpr)};
+ const auto immediate_offset{static_cast<u32>(instr.gmem.offset)};
+
const Node base_address{
TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()))};
const auto cbuf = std::get_if<CbufNode>(base_address);
diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp
index 77c6f9951..a6c123573 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -4,6 +4,7 @@
#include "common/assert.h"
#include "common/common_types.h"
+#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/shader_ir.h"
diff --git a/src/video_core/shader/decode/predicate_set_predicate.cpp b/src/video_core/shader/decode/predicate_set_predicate.cpp
index 83c61680e..71844c42b 100644
--- a/src/video_core/shader/decode/predicate_set_predicate.cpp
+++ b/src/video_core/shader/decode/predicate_set_predicate.cpp
@@ -64,4 +64,4 @@ u32 ShaderIR::DecodePredicateSetPredicate(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/predicate_set_register.cpp b/src/video_core/shader/decode/predicate_set_register.cpp
index d0495995d..387491bd3 100644
--- a/src/video_core/shader/decode/predicate_set_register.cpp
+++ b/src/video_core/shader/decode/predicate_set_register.cpp
@@ -43,4 +43,4 @@ u32 ShaderIR::DecodePredicateSetRegister(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/register_set_predicate.cpp b/src/video_core/shader/decode/register_set_predicate.cpp
index f070e8912..f8659e48e 100644
--- a/src/video_core/shader/decode/register_set_predicate.cpp
+++ b/src/video_core/shader/decode/register_set_predicate.cpp
@@ -48,4 +48,4 @@ u32 ShaderIR::DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/shift.cpp b/src/video_core/shader/decode/shift.cpp
index 951e85f44..44ae87ece 100644
--- a/src/video_core/shader/decode/shift.cpp
+++ b/src/video_core/shader/decode/shift.cpp
@@ -52,4 +52,4 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
return pc;
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/video.cpp b/src/video_core/shader/decode/video.cpp
index 956c01d9b..cb9ab72b1 100644
--- a/src/video_core/shader/decode/video.cpp
+++ b/src/video_core/shader/decode/video.cpp
@@ -108,4 +108,4 @@ Node ShaderIR::GetVideoOperand(Node op, bool is_chunk, bool is_signed,
}
}
-} // namespace VideoCommon::Shader \ No newline at end of file
+} // namespace VideoCommon::Shader
generated by cgit v1.2.3 (git 2.25.1) at 2024-08-03 21:01:10 +0200