piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-21 21:30:19 -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

31
Source/Core/VideoCommon/VertexLoaderX64.cpp
View File

@ -26,7 +26,9 @@ static const X64Reg dst_reg = ABI_PARAM2;
static const X64Reg scratch1 = RAX;
static const X64Reg scratch2 = ABI_PARAM3;
static const X64Reg scratch3 = ABI_PARAM4;
static const X64Reg count_reg = R10;
// The remaining number of vertices to be processed. Starts at count - 1, and the final loop has it
// at 0.
static const X64Reg remaining_reg = R10;
static const X64Reg skipped_reg = R11;
static const X64Reg base_reg = RBX;
@ -117,10 +119,11 @@ int VertexLoaderX64::ReadVertex(OpArg data, VertexComponentFormat attribute, Com
const auto write_zfreeze = [&]() { // zfreeze
if (native_format == &m_native_vtx_decl.position)
{
CMP(32, R(count_reg), Imm8(3));
FixupBranch dont_store = J_CC(CC_A);
LEA(32, scratch3,
MScaled(count_reg, SCALE_4, -int(VertexLoaderManager::position_cache[0].size())));
CMP(32, R(remaining_reg), Imm8(3));
FixupBranch dont_store = J_CC(CC_AE);
// The position cache is composed of 3 rows of 4 floats each; since each float is 4 bytes,
// we need to scale by 4 twice to cover the 4 floats.
LEA(32, scratch3, MScaled(remaining_reg, SCALE_4, 0));
MOVUPS(MPIC(VertexLoaderManager::position_cache.data(), scratch3, SCALE_4), coords);
SetJumpTarget(dont_store);
}
@ -380,8 +383,8 @@ void VertexLoaderX64::ReadColor(OpArg data, VertexComponentFormat attribute, Col
void VertexLoaderX64::GenerateVertexLoader()
{
BitSet32 regs = {src_reg, dst_reg, scratch1, scratch2,
scratch3, count_reg, skipped_reg, base_reg};
BitSet32 regs = {src_reg, dst_reg, scratch1, scratch2,
scratch3, remaining_reg, skipped_reg, base_reg};
regs &= ABI_ALL_CALLEE_SAVED;
ABI_PushRegistersAndAdjustStack(regs, 0);
@ -389,7 +392,9 @@ void VertexLoaderX64::GenerateVertexLoader()
PUSH(32, R(ABI_PARAM3));
// ABI_PARAM3 is one of the lower registers, so free it for scratch2.
MOV(32, R(count_reg), R(ABI_PARAM3));
// We also have it end at a value of 0, to simplify indexing for zfreeze;
// this requires subtracting 1 at the start.
LEA(32, remaining_reg, MDisp(ABI_PARAM3, -1));
MOV(64, R(base_reg), R(ABI_PARAM4));
@ -407,9 +412,9 @@ void VertexLoaderX64::GenerateVertexLoader()
MOV(32, MDisp(dst_reg, m_dst_ofs), R(scratch1));
// zfreeze
CMP(32, R(count_reg), Imm8(3));
FixupBranch dont_store = J_CC(CC_A);
MOV(32, MPIC(VertexLoaderManager::position_matrix_index_cache.data(), count_reg, SCALE_4),
CMP(32, R(remaining_reg), Imm8(3));
FixupBranch dont_store = J_CC(CC_AE);
MOV(32, MPIC(VertexLoaderManager::position_matrix_index_cache.data(), remaining_reg, SCALE_4),
R(scratch1));
SetJumpTarget(dont_store);
@ -509,8 +514,8 @@ void VertexLoaderX64::GenerateVertexLoader()
const u8* cont = GetCodePtr();
ADD(64, R(src_reg), Imm32(m_src_ofs));
SUB(32, R(count_reg), Imm8(1));
J_CC(CC_NZ, loop_start);
SUB(32, R(remaining_reg), Imm8(1));
J_CC(CC_AE, loop_start);
// Get the original count.
POP(32, R(ABI_RETURN));