Author: post
Date: 2010-03-07 21:40:49 +0100 (Sun, 07 Mar 2010)
New Revision: 201
Modified:
RawSpeed/RawImage.cpp
Log:
Enable SSE2 intrinsics for value scaling on 64 bit GCC.
Modified: RawSpeed/RawImage.cpp
===================================================================
--- RawSpeed/RawImage.cpp 2010-03-07 15:40:09 UTC (rev 200)
+++ RawSpeed/RawImage.cpp 2010-03-07 20:40:49 UTC (rev 201)
@@ -22,6 +22,10 @@
http://www.klauspost.com
*/
+#if defined(__SSE2__)
+#include <emmintrin.h>
+#endif
+
namespace RawSpeed {
RawImageData::RawImageData(void):
@@ -131,14 +135,20 @@
scaleValues(f);
}
-#if _MSC_VER > 1399
+#if _MSC_VER > 1399 || defined(__SSE2__)
void RawImageData::scaleValues(float f) {
+ gboolean use_sse2;
+#ifdef _MSC_VER
int info[4];
__cpuid(info, 1);
+ use_sse2 = !!info[3]&(1 << 26)
+#else
+ use_sse2 = TRUE;
+#endif
// Check SSE2
- if (f >= 0.0f && info[3]&(1 << 26)) {
+ if (f >= 0.0f && use_sse2) {
__m128i ssescale;
__m128i ssesub;
@@ -202,84 +212,19 @@
#else
void RawImageData::scaleValues(float f) {
-#if 0
- //TODO: Check for SSE2 on 32 bit systems and use it there
- guint temp[20];
-
- guint i = (int)(1024.0f * f); // 10 bit fraction
- i |= i << 16;
- guint b = blackLevel | (blackLevel << 16);
-
- for (int j = 0; j < 4; j++) {
- temp[j] = b;
- temp[j+4] = i;
- temp[j+8] = 512;
- temp[j+12] = 32768;
- temp[j+16] = 0x80008000;
- }
-
- asm volatile
- (
- "movdqu 0(%0), %%xmm7\n" // Subtraction
- "movdqu 16(%0), %%xmm6\n" // Multiplication factor
- "movdqu 32(%0), %%xmm5\n" // Fraction
- "movdqu 48(%0), %%xmm4\n" // Sub 32768
- "movdqu 64(%0), %%xmm3\n" // Sign shift
- : // no output registers
- : "r"(temp)
- : // %0
- );
-
+ gint gw = dim.x * cpp;
+ int scale = (int)(16384.0f * f); // 14 bit fraction
for (int y = 0; y < dim.y; y++) {
- guchar* pixel = (guchar*) & data[(mOffset.y+y)*pitch];
- guint gw = pitch >> 4;
- for (guint x = 0; x < gw ; x++) {
- asm volatile(
- "next_pixel:\n"
- "movaps 0(%0), %%xmm0\n"
- "psubusw %%xmm7, %%xmm0\n" // Subtract black
- "movaps %%xmm0, %%xmm1\n"
- "pmullw %%xmm6, %%xmm0\n"
- "pmulhuw %%xmm6, %%xmm1\n"
- "movaps %%xmm0, %%xmm2\n"
- "punpcklwd %%xmm1, %%xmm0\n" // First 4 result
- "punpckhwd %%xmm1, %%xmm2\n" // Last 4 result
- "paddd %%xmm5, %%xmm0\n" // Add fraction
- "paddd %%xmm5, %%xmm2\n"
- "psrad $10, %%xmm0\n"
- "psrad $10, %%xmm2\n"
- "psubd %%xmm4, %%xmm0\n" // Avoid saturation
- "psubd %%xmm4, %%xmm2\n"
- "packssdw %%xmm2, %%xmm0\n"
- "pxor %%xmm3, %%xmm0\n" // Shift sign
- "movaps %%xmm0, 0(%0)\n"
-
- "add $16, %0\n"
- : // no output registers
- : "r"(pixel)
- : // %0
- );
+ gushort *pixel = (gushort*)getData(0, y);
+ for (int x = 0 ; x < gw; x++) {
+ pixel[x] = clampbits(((pixel[x] - blackLevel) * scale + 8192) >> 14, 16);
}
}
-
-#else
-
-gint gw = dim.x * cpp;
-int scale = (int)(16384.0f * f); // 14 bit fraction
-for (int y = 0; y < dim.y; y++) {
- gushort *pixel = (gushort*)getData(0, y);
- for (int x = 0 ; x < gw; x++) {
- pixel[x] = clampbits(((pixel[x] - blackLevel) * scale + 8192) >> 14, 16);
- }
}
#endif
-}
-#endif
-
-
RawImage::RawImage(RawImageData* p) : p_(p) {
pthread_mutex_lock(&p_->mymutex);
++p_->dataRefCount;
_______________________________________________
Rawstudio-commit mailing list
[email protected]
http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-commit
