Menu
▾
▴
[OMPT-svn-commit] SF.net SVN: modplug:[3755] trunk/OpenMPT
|
From: <sag...@us...> - 2014-02-22 23:47:34
|
Revision: 3755
http://sourceforge.net/p/modplug/code/3755
Author: saga-games
Date: 2014-02-22 23:47:25 +0000 (Sat, 22 Feb 2014)
Log Message:
-----------
[Mod] Add SSE2 implementation for min/max finder. Only used if AMD MMX extensions and SSE code generation are disabled, as those won't work on x64 (since __m64 is not available) and the performance difference is marginal.
Modified Paths:
--------------
trunk/OpenMPT/common/BuildSettings.h
trunk/OpenMPT/common/misc_util.cpp
trunk/OpenMPT/mptrack/View_smp.cpp
Modified: trunk/OpenMPT/common/BuildSettings.h
===================================================================
--- trunk/OpenMPT/common/BuildSettings.h 2014-02-22 22:16:25 UTC (rev 3754)
+++ trunk/OpenMPT/common/BuildSettings.h 2014-02-22 23:47:25 UTC (rev 3755)
@@ -36,8 +36,10 @@
// inline assembly requires MSVC compiler
-#if defined(ENABLE_ASM) && MPT_COMPILER_MSVC && defined(_M_IX86)
+#if defined(ENABLE_ASM) && MPT_COMPILER_MSVC
+#ifdef _M_IX86
+
// Generate general x86 inline assembly.
#define ENABLE_X86
@@ -47,9 +49,22 @@
// Generate inline assembly using SSE instructions (only used when the CPU supports it).
#define ENABLE_SSE
+// Generate inline assembly using SSE2 instructions (only used when the CPU supports it).
+#define ENABLE_SSE2
+
// Generate inline assembly using AMD specific instruction set extensions (only used when the CPU supports it).
#define ENABLE_X86_AMD
+#elif defined(_M_X64)
+
+// Generate general x64 inline assembly.
+#define ENABLE_X64
+
+// Generate inline assembly using SSE2 instructions (only used when the CPU supports it).
+#define ENABLE_SSE2
+
+#endif
+
#endif // ENABLE_ASM
Modified: trunk/OpenMPT/common/misc_util.cpp
===================================================================
--- trunk/OpenMPT/common/misc_util.cpp 2014-02-22 22:16:25 UTC (rev 3754)
+++ trunk/OpenMPT/common/misc_util.cpp 2014-02-22 23:47:25 UTC (rev 3755)
@@ -218,7 +218,7 @@
uint32 ProcSupport = 0;
-#if MPT_COMPILER_MSVC && defined(ENABLE_X86)
+#if MPT_COMPILER_MSVC && (defined(ENABLE_X86) || defined(ENABLE_X64))
#include <intrin.h>
Modified: trunk/OpenMPT/mptrack/View_smp.cpp
===================================================================
--- trunk/OpenMPT/mptrack/View_smp.cpp 2014-02-22 22:16:25 UTC (rev 3754)
+++ trunk/OpenMPT/mptrack/View_smp.cpp 2014-02-22 23:47:25 UTC (rev 3755)
@@ -579,7 +579,6 @@
//-------------------------------------------------------------------------------------------------------
{
scanlen *= channels;
-
__m64 minVal = _mm_cvtsi32_si64(smin);
__m64 maxVal = _mm_cvtsi32_si64(smax);
@@ -593,7 +592,7 @@
if(scanlen4)
{
const __m64 *v = static_cast<const __m64 *>(p);
- p = static_cast<const uint64 *>(p) + scanlen4;
+ p = static_cast<const __m64 *>(p) + scanlen4;
while(scanlen4--)
{
@@ -602,15 +601,16 @@
maxVal = _mm_max_pi16(maxVal, curVals);
}
+ // Now we have 4 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 2 values to the lower half and compute the minima/maxima of that.
__m64 minVal2 = _mm_unpackhi_pi32(minVal, minVal);
__m64 maxVal2 = _mm_unpackhi_pi32(maxVal, maxVal);
-
minVal = _mm_min_pi16(minVal, minVal2);
maxVal = _mm_max_pi16(maxVal, maxVal2);
if(channels < 2)
{
- // Mono
+ // Mono: Compute the minima/maxima of the both remaining values
minVal2 = _mm_sra_pi32(minVal, _mm_cvtsi32_si64(16));
maxVal2 = _mm_sra_pi32(maxVal, _mm_cvtsi32_si64(16));
minVal = _mm_min_pi16(minVal, minVal2);
@@ -665,7 +665,7 @@
if(scanlen8)
{
const __m64 *v = static_cast<const __m64 *>(p);
- p = static_cast<const uint64 *>(p) + scanlen8;
+ p = static_cast<const __m64 *>(p) + scanlen8;
while(scanlen8--)
{
@@ -674,12 +674,15 @@
maxVal = _mm_max_pu8(maxVal, curVals);
}
+ // Now we have 8 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 4 values to the lower half and compute the minima/maxima of that.
__m64 minVal2 = _mm_unpackhi_pi32(minVal, minVal);
__m64 maxVal2 = _mm_unpackhi_pi32(maxVal, maxVal);
-
minVal = _mm_min_pu8(minVal, minVal2);
maxVal = _mm_max_pu8(maxVal, maxVal2);
+ // Now we have 4 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 2 values to the lower half and compute the minima/maxima of that.
minVal2 = _mm_srl_pi32(minVal, _mm_cvtsi32_si64(16));
maxVal2 = _mm_srl_pi32(maxVal, _mm_cvtsi32_si64(16));
minVal = _mm_min_pu8(minVal, minVal2);
@@ -687,7 +690,7 @@
if(channels < 2)
{
- // Mono
+ // Mono: Compute the minima/maxima of the both remaining values
minVal2 = _mm_srl_pi32(minVal, _mm_cvtsi32_si64(8));
maxVal2 = _mm_srl_pi32(maxVal, _mm_cvtsi32_si64(8));
minVal = _mm_min_pu8(minVal, minVal2);
@@ -715,8 +718,146 @@
_mm_empty();
}
-#endif
+#elif defined(ENABLE_SSE2)
+#include <emmintrin.h>
+
+// SSE2 implementation for min/max finder, packs 8*int16 in a 128-bit XMM register.
+// scanlen = How many samples to process on this channel
+static void sse2_findminmax16(const void *p, int scanlen, int channels, int &smin, int &smax)
+//-------------------------------------------------------------------------------------------
+{
+ scanlen *= channels;
+
+ // Put minimum / maximum in 8 packed int16 values
+ __m128i minVal = _mm_set1_epi16(static_cast<int16>(smin));
+ __m128i maxVal = _mm_set1_epi16(static_cast<int16>(smax));
+
+ int scanlen8 = scanlen / 8;
+ if(scanlen8)
+ {
+ const __m128i *v = static_cast<const __m128i *>(p);
+ p = static_cast<const __m128i *>(p) + scanlen8;
+
+ while(scanlen8--)
+ {
+ __m128i curVals = _mm_loadu_si128(v++);
+ minVal = _mm_min_epi16(minVal, curVals);
+ maxVal = _mm_max_epi16(maxVal, curVals);
+ }
+
+ // Now we have 8 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 4 values to the lower half and compute the minima/maxima of that.
+ __m128i minVal2 = _mm_unpackhi_epi64(minVal, minVal);
+ __m128i maxVal2 = _mm_unpackhi_epi64(maxVal, maxVal);
+ minVal = _mm_min_epi16(minVal, minVal2);
+ maxVal = _mm_max_epi16(maxVal, maxVal2);
+
+ // Now we have 4 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 2 values to the lower half and compute the minima/maxima of that.
+ minVal2 = _mm_shuffle_epi32(minVal, _MM_SHUFFLE(1, 1, 1, 1));
+ maxVal2 = _mm_shuffle_epi32(maxVal, _MM_SHUFFLE(1, 1, 1, 1));
+ minVal = _mm_min_epi16(minVal, minVal2);
+ maxVal = _mm_max_epi16(maxVal, maxVal2);
+
+ if(channels < 2)
+ {
+ // Mono: Compute the minima/maxima of the both remaining values
+ minVal2 = _mm_shufflelo_epi16(minVal, _MM_SHUFFLE(1, 1, 1, 1));
+ maxVal2 = _mm_shufflelo_epi16(maxVal, _MM_SHUFFLE(1, 1, 1, 1));
+ minVal = _mm_min_epi16(minVal, minVal2);
+ maxVal = _mm_max_epi16(maxVal, maxVal2);
+ }
+ }
+
+ const int16 *p16 = static_cast<const int16 *>(p);
+ while(scanlen & 7)
+ {
+ scanlen -= channels;
+ __m128i curVals = _mm_set1_epi16(*p16);
+ p16 += channels;
+ minVal = _mm_min_epi16(minVal, curVals);
+ maxVal = _mm_max_epi16(maxVal, curVals);
+ }
+
+ smin = static_cast<int16>(_mm_cvtsi128_si32(minVal));
+ smax = static_cast<int16>(_mm_cvtsi128_si32(maxVal));
+}
+
+
+// SSE2 implementation for min/max finder, packs 16*int8 in a 128-bit XMM register.
+// scanlen = How many samples to process on this channel
+static void sse2_findminmax8(const void *p, int scanlen, int channels, int &smin, int &smax)
+//------------------------------------------------------------------------------------------
+{
+ scanlen *= channels;
+
+ // Put minimum / maximum in 8 packed int16 values
+ __m128i minVal = _mm_set1_epi8(static_cast<int8>(smin));
+ __m128i maxVal = _mm_set1_epi8(static_cast<int8>(smax));
+
+ int scanlen16 = scanlen / 16;
+ if(scanlen16)
+ {
+ const __m128i *v = static_cast<const __m128i *>(p);
+ p = static_cast<const __m128i *>(p) + scanlen16;
+
+ while(scanlen16--)
+ {
+ __m128i curVals = _mm_loadu_si128(v++);
+ minVal = _mm_min_epi8(minVal, curVals);
+ maxVal = _mm_max_epi8(maxVal, curVals);
+ }
+
+ // Now we have 16 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 8 values to the lower half and compute the minima/maxima of that.
+ __m128i minVal2 = _mm_unpackhi_epi64(minVal, minVal);
+ __m128i maxVal2 = _mm_unpackhi_epi64(maxVal, maxVal);
+ minVal = _mm_min_epi8(minVal, minVal2);
+ maxVal = _mm_max_epi8(maxVal, maxVal2);
+
+ // Now we have 8 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 4 values to the lower half and compute the minima/maxima of that.
+ minVal2 = _mm_shuffle_epi32(minVal, _MM_SHUFFLE(1, 1, 1, 1));
+ maxVal2 = _mm_shuffle_epi32(maxVal, _MM_SHUFFLE(1, 1, 1, 1));
+ minVal = _mm_min_epi8(minVal, minVal2);
+ maxVal = _mm_max_epi8(maxVal, maxVal2);
+
+ // Now we have 4 minima and maxima each, in case of stereo they are interleaved L/R values.
+ // Move the upper 2 values to the lower half and compute the minima/maxima of that.
+ minVal2 = _mm_srai_epi32(minVal, 16);
+ maxVal2 = _mm_srai_epi32(maxVal, 16);
+ minVal = _mm_min_epi8(minVal, minVal2);
+ maxVal = _mm_max_epi8(maxVal, maxVal2);
+
+ if(channels < 2)
+ {
+ // Mono: Compute the minima/maxima of the both remaining values
+ minVal2 = _mm_srai_epi16(minVal, 8);
+ maxVal2 = _mm_srai_epi16(maxVal, 8);
+ minVal = _mm_min_epi8(minVal, minVal2);
+ maxVal = _mm_max_epi8(maxVal, maxVal2);
+ }
+ }
+
+ const int8 *p8 = static_cast<const int8 *>(p);
+ while(scanlen & 15)
+ {
+ scanlen -= channels;
+ __m128i curVals = _mm_set1_epi8(*p8);
+ p8 += channels;
+ minVal = _mm_min_epi8(minVal, curVals);
+ maxVal = _mm_max_epi8(maxVal, curVals);
+ }
+
+ smin = static_cast<int8>(_mm_cvtsi128_si32(minVal));
+ smax = static_cast<int8>(_mm_cvtsi128_si32(maxVal));
+}
+
+
+#endif // defined(ENABLE_SSE)
+
+
// Draw one channel of zoomed-out sample data
void CViewSample::DrawSampleData2(HDC hdc, int ymed, int cx, int cy, SmpLength len, int uFlags, const void *pSampleData)
//----------------------------------------------------------------------------------------------------------------------
@@ -776,9 +917,14 @@
{
amdmmxext_or_sse_findminmax16(p, scanlen, numChannels, smin, smax);
} else
+#elif defined(ENABLE_SSE2)
+ if(GetProcSupport() & PROCSUPPORT_SSE2)
+ {
+ sse2_findminmax16(p, scanlen, numChannels, smin, smax);
+ } else
#endif
{
- for (SmpLength i=0; i<scanlen; i++)
+ for (SmpLength i = 0; i < scanlen; i++)
{
int s = *p;
if (s < smin) smin = s;
@@ -799,9 +945,14 @@
{
amdmmxext_or_sse_findminmax8(p, scanlen, numChannels, smin, smax);
} else
+#elif defined(ENABLE_SSE2)
+ if(GetProcSupport() & PROCSUPPORT_SSE2)
+ {
+ sse2_findminmax8(p, scanlen, numChannels, smin, smax);
+ } else
#endif
{
- for (SmpLength i=0; i<scanlen; i++)
+ for (SmpLength i = 0; i < scanlen; i++)
{
int s = *p;
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
