From ef6f6a7fa9a90efa78d0ae557e5546dd6568bb74 Mon Sep 17 00:00:00 2001
From: degasus <wickmarkus@web.de>
Date: Wed, 3 Sep 2014 22:39:26 +0200
Subject: [PATCH] VideoCommon: remove XFReg copy optimization
This code is just ugly and I doubt there is a way that copying twice is faster.
---
Source/Core/VideoCommon/DataReader.h | 40 ----------------
Source/Core/VideoCommon/OpcodeDecoding.cpp | 53 +---------------------
Source/Core/VideoCommon/XFMemory.h | 2 +-
Source/Core/VideoCommon/XFStructs.cpp | 21 +++++----
4 files changed, 15 insertions(+), 101 deletions(-)
diff --git a/Source/Core/VideoCommon/DataReader.h b/Source/Core/VideoCommon/DataReader.h
index 17c927a42b..85beec3a11 100644
--- a/Source/Core/VideoCommon/DataReader.h
+++ b/Source/Core/VideoCommon/DataReader.h
@@ -92,46 +92,6 @@ __forceinline u32 DataReadU32()
return DataRead<u32>();
}
-#if _M_SSE >= 0x301
-const __m128i bs_mask = _mm_set_epi32(0x0C0D0E0FL, 0x08090A0BL, 0x04050607L, 0x00010203L);
-
-template<unsigned int N>
-void DataReadU32xN_SSSE3(u32 *bufx16)
-{
- memcpy(bufx16, g_pVideoData, sizeof(u32) * N);
- __m128i* buf = (__m128i *)bufx16;
- if (N>12) { _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), bs_mask)); buf++; }
- if (N>8) { _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), bs_mask)); buf++; }
- if (N>4) { _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), bs_mask)); buf++; }
- _mm_store_si128(buf, _mm_shuffle_epi8(_mm_load_si128(buf), bs_mask));
- g_pVideoData += (sizeof(u32) * N);
-}
-
-#endif
-
-template<unsigned int N>
-void DataReadU32xN(u32 *bufx16)
-{
- memcpy(bufx16, g_pVideoData, sizeof(u32) * N);
- if (N >= 1) bufx16[0] = Common::swap32(bufx16[0]);
- if (N >= 2) bufx16[1] = Common::swap32(bufx16[1]);
- if (N >= 3) bufx16[2] = Common::swap32(bufx16[2]);
- if (N >= 4) bufx16[3] = Common::swap32(bufx16[3]);
- if (N >= 5) bufx16[4] = Common::swap32(bufx16[4]);
- if (N >= 6) bufx16[5] = Common::swap32(bufx16[5]);
- if (N >= 7) bufx16[6] = Common::swap32(bufx16[6]);
- if (N >= 8) bufx16[7] = Common::swap32(bufx16[7]);
- if (N >= 9) bufx16[8] = Common::swap32(bufx16[8]);
- if (N >= 10) bufx16[9] = Common::swap32(bufx16[9]);
- if (N >= 11) bufx16[10] = Common::swap32(bufx16[10]);
- if (N >= 12) bufx16[11] = Common::swap32(bufx16[11]);
- if (N >= 13) bufx16[12] = Common::swap32(bufx16[12]);
- if (N >= 14) bufx16[13] = Common::swap32(bufx16[13]);
- if (N >= 15) bufx16[14] = Common::swap32(bufx16[14]);
- if (N >= 16) bufx16[15] = Common::swap32(bufx16[15]);
- g_pVideoData += (sizeof(u32) * N);
-}
-
__forceinline u32 DataReadU32Unswapped()
{
u32 tmp = *(u32*)g_pVideoData;
diff --git a/Source/Core/VideoCommon/OpcodeDecoding.cpp b/Source/Core/VideoCommon/OpcodeDecoding.cpp
index fe644db21e..a5bd8a62ee 100644
--- a/Source/Core/VideoCommon/OpcodeDecoding.cpp
+++ b/Source/Core/VideoCommon/OpcodeDecoding.cpp
@@ -34,47 +34,6 @@
u8* g_pVideoData = nullptr;
bool g_bRecordFifoData = false;
-typedef void (*DataReadU32xNfunc)(u32 *buf);
-#if _M_SSE >= 0x301
-static DataReadU32xNfunc DataReadU32xFuncs_SSSE3[16] = {
- DataReadU32xN_SSSE3<1>,
- DataReadU32xN_SSSE3<2>,
- DataReadU32xN_SSSE3<3>,
- DataReadU32xN_SSSE3<4>,
- DataReadU32xN_SSSE3<5>,
- DataReadU32xN_SSSE3<6>,
- DataReadU32xN_SSSE3<7>,
- DataReadU32xN_SSSE3<8>,
- DataReadU32xN_SSSE3<9>,
- DataReadU32xN_SSSE3<10>,
- DataReadU32xN_SSSE3<11>,
- DataReadU32xN_SSSE3<12>,
- DataReadU32xN_SSSE3<13>,
- DataReadU32xN_SSSE3<14>,
- DataReadU32xN_SSSE3<15>,
- DataReadU32xN_SSSE3<16>
-};
-#endif
-
-static DataReadU32xNfunc DataReadU32xFuncs[16] = {
- DataReadU32xN<1>,
- DataReadU32xN<2>,
- DataReadU32xN<3>,
- DataReadU32xN<4>,
- DataReadU32xN<5>,
- DataReadU32xN<6>,
- DataReadU32xN<7>,
- DataReadU32xN<8>,
- DataReadU32xN<9>,
- DataReadU32xN<10>,
- DataReadU32xN<11>,
- DataReadU32xN<12>,
- DataReadU32xN<13>,
- DataReadU32xN<14>,
- DataReadU32xN<15>,
- DataReadU32xN<16>
-};
-
static u32 InterpretDisplayList(u32 address, u32 size)
{
u8* old_pVideoData = g_pVideoData;
@@ -182,9 +141,7 @@ static u32 Decode(u8* end, bool skipped_frame)
return 0;
cycles = 18 + 6 * transfer_size;
u32 xf_address = Cmd2 & 0xFFFF;
- GC_ALIGNED128(u32 data_buffer[16]);
- DataReadU32xFuncs[transfer_size-1](data_buffer);
- LoadXFReg(transfer_size, xf_address, data_buffer);
+ LoadXFReg(transfer_size, xf_address);
INCSTAT(stats.thisFrame.numXFLoads);
}
@@ -296,14 +253,6 @@ static u32 Decode(u8* end, bool skipped_frame)
void OpcodeDecoder_Init()
{
g_pVideoData = GetVideoBufferStartPtr();
-
-#if _M_SSE >= 0x301
- if (cpu_info.bSSSE3)
- {
- for (int i = 0; i < 16; ++i)
- DataReadU32xFuncs[i] = DataReadU32xFuncs_SSSE3[i];
- }
-#endif
}
diff --git a/Source/Core/VideoCommon/XFMemory.h b/Source/Core/VideoCommon/XFMemory.h
index f36378b1e9..3646a70264 100644
--- a/Source/Core/VideoCommon/XFMemory.h
+++ b/Source/Core/VideoCommon/XFMemory.h
@@ -273,5 +273,5 @@ struct XFMemory
extern XFMemory xfmem;
-void LoadXFReg(u32 transferSize, u32 address, u32 *pData);
+void LoadXFReg(u32 transferSize, u32 address);
void LoadIndexedXF(u32 val, int array);
diff --git a/Source/Core/VideoCommon/XFStructs.cpp b/Source/Core/VideoCommon/XFStructs.cpp
index 235c76bea6..650a4a913b 100644
--- a/Source/Core/VideoCommon/XFStructs.cpp
+++ b/Source/Core/VideoCommon/XFStructs.cpp
@@ -5,6 +5,7 @@
#include "Common/Common.h"
#include "Core/HW/Memmap.h"
#include "VideoCommon/CPMemory.h"
+#include "VideoCommon/DataReader.h"
#include "VideoCommon/PixelShaderManager.h"
#include "VideoCommon/VertexManagerBase.h"
#include "VideoCommon/VertexShaderManager.h"
@@ -17,14 +18,14 @@ static void XFMemWritten(u32 transferSize, u32 baseAddress)
VertexShaderManager::InvalidateXFRange(baseAddress, baseAddress + transferSize);
}
-static void XFRegWritten(int transferSize, u32 baseAddress, u32 *pData)
+static void XFRegWritten(int transferSize, u32 baseAddress)
{
u32 address = baseAddress;
u32 dataIndex = 0;
while (transferSize > 0 && address < 0x1058)
{
- u32 newValue = pData[dataIndex];
+ u32 newValue = DataPeek<u32>(dataIndex * sizeof(u32));
u32 nextAddress = address + 1;
switch (address)
@@ -191,7 +192,7 @@ static void XFRegWritten(int transferSize, u32 baseAddress, u32 *pData)
}
}
-void LoadXFReg(u32 transferSize, u32 baseAddress, u32 *pData)
+void LoadXFReg(u32 transferSize, u32 baseAddress)
{
// do not allow writes past registers
if (baseAddress + transferSize > 0x1058)
@@ -225,16 +226,20 @@ void LoadXFReg(u32 transferSize, u32 baseAddress, u32 *pData)
}
XFMemWritten(xfMemTransferSize, xfMemBase);
- memcpy((u32*)(&xfmem) + xfMemBase, pData, xfMemTransferSize * 4);
-
- pData += xfMemTransferSize;
+ for (u32 i = 0; i < xfMemTransferSize; i++)
+ {
+ ((u32*)&xfmem)[xfMemBase + i] = DataRead<u32>();
+ }
}
// write to XF regs
if (transferSize > 0)
{
- XFRegWritten(transferSize, baseAddress, pData);
- memcpy((u32*)(&xfmem) + baseAddress, pData, transferSize * 4);
+ XFRegWritten(transferSize, baseAddress);
+ for (u32 i = 0; i < transferSize; i++)
+ {
+ ((u32*)&xfmem)[baseAddress + i] = DataRead<u32>();
+ }
}
}