piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-21 05:09:34 -06:00
Code Issues Packages Projects Releases Wiki Activity

VertexLoader: Convert count register to remaining register

Browse Source 
This more accurately represents what's going on, and also ends at 0 instead of 1, making some indexing operations easier.  This also changes it so that position_matrix_index_cache actually starts from index 0 instead of index 1.
This commit is contained in:
Pokechu22
2022-04-14 12:01:57 -07:00
parent 97d0ff58c8
commit 39b2854b98
8 changed files with 44 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
Source/Core/VideoCommon/VertexLoaderARM64.cpp
View File

@ -14,7 +14,7 @@ using namespace Arm64Gen;
constexpr ARM64Reg src_reg = ARM64Reg::X0;
constexpr ARM64Reg dst_reg = ARM64Reg::X1;
constexpr ARM64Reg count_reg = ARM64Reg::W2;
constexpr ARM64Reg remaining_reg = ARM64Reg::W2;
constexpr ARM64Reg skipped_reg = ARM64Reg::W17;
constexpr ARM64Reg scratch1_reg = ARM64Reg::W16;
constexpr ARM64Reg scratch2_reg = ARM64Reg::W15;
@ -209,13 +209,10 @@ int VertexLoaderARM64::ReadVertex(VertexComponentFormat attribute, ComponentForm
// Z-Freeze
if (native_format == &m_native_vtx_decl.position)
{
CMP(count_reg, 3);
FixupBranch dont_store = B(CC_GT);
CMP(remaining_reg, 3);
FixupBranch dont_store = B(CC_GE);
MOVP2R(EncodeRegTo64(scratch2_reg), VertexLoaderManager::position_cache.data());
ADD(EncodeRegTo64(scratch1_reg), EncodeRegTo64(scratch2_reg), EncodeRegTo64(count_reg),
ArithOption(EncodeRegTo64(count_reg), ShiftType::LSL, 4));
m_float_emit.STUR(write_size, coords, EncodeRegTo64(scratch1_reg),
-int(sizeof(decltype(VertexLoaderManager::position_cache[0]))));
m_float_emit.STR(128, coords, EncodeRegTo64(scratch2_reg), ArithOption(remaining_reg, true));
SetJumpTarget(dont_store);
}
@ -404,7 +401,7 @@ void VertexLoaderARM64::GenerateVertexLoader()
AlignCode16();
if (IsIndexed(m_VtxDesc.low.Position))
MOV(skipped_reg, ARM64Reg::WZR);
MOV(saved_count, count_reg);
ADD(saved_count, remaining_reg, 1);
MOVP2R(stride_reg, g_main_cp_state.array_strides.data());
MOVP2R(arraybase_reg, VertexLoaderManager::cached_arraybases.data());
@ -421,10 +418,10 @@ void VertexLoaderARM64::GenerateVertexLoader()
STR(IndexType::Unsigned, scratch1_reg, dst_reg, m_dst_ofs);
// Z-Freeze
CMP(count_reg, 3);
FixupBranch dont_store = B(CC_GT);
CMP(remaining_reg, 3);
FixupBranch dont_store = B(CC_GE);
MOVP2R(EncodeRegTo64(scratch2_reg), VertexLoaderManager::position_matrix_index_cache.data());
STR(scratch1_reg, EncodeRegTo64(scratch2_reg), ArithOption(count_reg, true));
STR(scratch1_reg, EncodeRegTo64(scratch2_reg), ArithOption(remaining_reg, true));
SetJumpTarget(dont_store);
m_native_vtx_decl.posmtx.components = 4;
@ -584,8 +581,8 @@ void VertexLoaderARM64::GenerateVertexLoader()
const u8* cont = GetCodePtr();
ADD(src_reg, src_reg, m_src_ofs);
SUB(count_reg, count_reg, 1);
CBNZ(count_reg, loop_start);
SUBS(remaining_reg, remaining_reg, 1);
B(CCFlags::CC_GE, loop_start);
if (IsIndexed(m_VtxDesc.low.Position))
{
@ -612,5 +609,5 @@ int VertexLoaderARM64::RunVertices(DataReader src, DataReader dst, int count)
{
m_numLoadedVertices += count;
return ((int (*)(u8 * src, u8 * dst, int count)) region)(src.GetPointer(), dst.GetPointer(),
count);
count - 1);
}