pixman: Branch 'master'
Søren Sandmann Pedersen
sandmann at kemper.freedesktop.org
Tue Jul 21 00:47:08 PDT 2009
pixman/Makefile.am | 2
pixman/pixman-sse2.c | 9 +
pixman/pixman-x64-mmx-emulation.h | 263 ++++++++++++++++++++++++++++++++++++++
3 files changed, 273 insertions(+), 1 deletion(-)
New commits:
commit 0ff5733c16804d5b10782556eeeade7061924846
Author: George Yohng <georgefd at oss3d.com>
Date: Tue Jul 21 03:43:42 2009 -0400
Add implementation of MMX __m64 functions for MSVC x64.
Microsoft C++ does not define __m64 and all related MMX functions in
x64. However, it succeeds in generating object files for SSE2 code
inside pixman.
The real problem happens during linking, when it cannot find MMX functions
(which are not defined as intrinsics for AMD64 platform).
I have implemented those missing functions using general programming.
MMX __m64 is used relatively scarcely within SSE2 implementation, and the
performance impact probably is negligible.
Bug 22390.
diff --git a/pixman/Makefile.am b/pixman/Makefile.am
index 0c528d8..e19fa6e 100644
--- a/pixman/Makefile.am
+++ b/pixman/Makefile.am
@@ -52,7 +52,7 @@ pixman-combine64.h : pixman-combine.h.template make-combine.pl
$(PERL) $(srcdir)/make-combine.pl 16 < $(srcdir)/pixman-combine.h.template > $@ || ($(RM) $@; exit 1)
EXTRA_DIST = Makefile.win32 pixman-combine.c.template make-combine.pl pixman-region.c \
- pixman-combine.h.template solaris-hwcap.mapfile
+ pixman-combine.h.template solaris-hwcap.mapfile pixman-x64-mmx-emulation.h
CLEANFILES = pixman-combine32.c pixman-combine64.c pixman-combine32.h pixman-combine64.h
# mmx code
diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index 3ed8165..2fa956e 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -36,6 +36,15 @@
#include "pixman-private.h"
#include "pixman-combine32.h"
+#if defined(_MSC_VER) && defined(_M_AMD64)
+/* Windows 64 doesn't allow MMX to be used, so
+ * the pixman-x64-mmx-emulation.h file contains
+ * implementations of those MMX intrinsics that
+ * are used in the SSE2 implementation.
+ */
+# include "pixman-x64-mmx-emulation.h"
+#endif
+
#ifdef USE_SSE2
/* --------------------------------------------------------------------
diff --git a/pixman/pixman-x64-mmx-emulation.h b/pixman/pixman-x64-mmx-emulation.h
new file mode 100644
index 0000000..231030f
--- /dev/null
+++ b/pixman/pixman-x64-mmx-emulation.h
@@ -0,0 +1,263 @@
+#ifndef MMX_X64_H_INCLUDED
+#define MMX_X64_H_INCLUDED
+
+/* Implementation of x64 MMX substitition functions, before
+ * pixman is reimplemented not to use __m64 type on Visual C++
+ *
+ * Copyright (C)2009 by George Yohng
+ * Released in public domain.
+ */
+
+#include <intrin.h>
+
+#define M64C(a) (*(const __m64 *)(&a))
+#define M64U(a) (*(const unsigned long long *)(&a))
+
+__inline __m64
+_m_from_int (int a)
+{
+ long long i64 = a;
+
+ return M64C (i64);
+}
+
+__inline __m64
+_mm_setzero_si64 ()
+{
+ long long i64 = 0;
+
+ return M64C (i64);
+}
+
+__inline __m64
+_mm_set_pi32 (int i1, int i0)
+{
+ unsigned long long i64 = ((unsigned)i0) + (((unsigned long long)(unsigned)i1) << 32);
+
+ return M64C (i64);
+}
+
+__inline void
+_m_empty ()
+{
+}
+
+__inline __m64
+_mm_set1_pi16 (short w)
+{
+ unsigned long long i64 = ((unsigned long long)(unsigned short)(w)) * 0x0001000100010001ULL;
+
+ return M64C (i64);
+}
+
+__inline int
+_m_to_int (__m64 m)
+{
+ return m.m64_i32[0];
+}
+
+__inline __m64
+_mm_movepi64_pi64 (__m128i a)
+{
+ return M64C (a.m128i_i64[0]);
+}
+
+__inline __m64
+_m_pand (__m64 a, __m64 b)
+{
+ unsigned long long i64 = M64U (a) & M64U (b);
+
+ return M64C (i64);
+}
+
+__inline __m64
+_m_por (__m64 a, __m64 b)
+{
+ unsigned long long i64 = M64U (a) | M64U (b);
+
+ return M64C (i64);
+}
+
+__inline __m64
+_m_pxor (__m64 a, __m64 b)
+{
+ unsigned long long i64 = M64U (a) ^ M64U (b);
+
+ return M64C (i64);
+}
+
+__inline __m64
+_m_pmulhuw (__m64 a, __m64 b) /* unoptimized */
+{
+ unsigned short d[4] =
+ {
+ (unsigned short)((((unsigned)a.m64_u16[0]) * b.m64_u16[0]) >> 16),
+ (unsigned short)((((unsigned)a.m64_u16[1]) * b.m64_u16[1]) >> 16),
+ (unsigned short)((((unsigned)a.m64_u16[2]) * b.m64_u16[2]) >> 16),
+ (unsigned short)((((unsigned)a.m64_u16[3]) * b.m64_u16[3]) >> 16)
+ };
+
+ return M64C (d[0]);
+}
+
+__inline __m64
+_m_pmullw2 (__m64 a, __m64 b) /* unoptimized */
+{
+ unsigned short d[4] =
+ {
+ (unsigned short)((((unsigned)a.m64_u16[0]) * b.m64_u16[0])),
+ (unsigned short)((((unsigned)a.m64_u16[1]) * b.m64_u16[1])),
+ (unsigned short)((((unsigned)a.m64_u16[2]) * b.m64_u16[2])),
+ (unsigned short)((((unsigned)a.m64_u16[3]) * b.m64_u16[3]))
+ };
+
+ return M64C (d[0]);
+}
+
+__inline __m64
+_m_pmullw (__m64 a, __m64 b) /* unoptimized */
+{
+ unsigned long long x =
+ ((unsigned long long)(unsigned short)((((unsigned)a.m64_u16[0]) * b.m64_u16[0]))) +
+ (((unsigned long long)(unsigned short)((((unsigned)a.m64_u16[1]) * b.m64_u16[1]))) << 16) +
+ (((unsigned long long)(unsigned short)((((unsigned)a.m64_u16[2]) * b.m64_u16[2]))) << 32) +
+ (((unsigned long long)(unsigned short)((((unsigned)a.m64_u16[3]) * b.m64_u16[3]))) << 48);
+
+ return M64C (x);
+}
+
+__inline __m64
+_m_paddusb (__m64 a, __m64 b) /* unoptimized */
+{
+ unsigned long long x = (M64U (a) & 0x00FF00FF00FF00FFULL) +
+ (M64U (b) & 0x00FF00FF00FF00FFULL);
+
+ unsigned long long y = ((M64U (a) >> 8) & 0x00FF00FF00FF00FFULL) +
+ ((M64U (b) >> 8) & 0x00FF00FF00FF00FFULL);
+
+ x | = ((x & 0xFF00FF00FF00FF00ULL) >> 8) * 0xFF;
+ y | = ((y & 0xFF00FF00FF00FF00ULL) >> 8) * 0xFF;
+
+ x = (x & 0x00FF00FF00FF00FFULL) | ((y & 0x00FF00FF00FF00FFULL) << 8);
+
+ return M64C (x);
+}
+
+__inline __m64
+_m_paddusw (__m64 a, __m64 b) /* unoptimized */
+{
+ unsigned long long x = (M64U (a) & 0x0000FFFF0000FFFFULL) +
+ (M64U (b) & 0x0000FFFF0000FFFFULL);
+
+ unsigned long long y = ((M64U (a) >> 16) & 0x0000FFFF0000FFFFULL) +
+ ((M64U (b) >> 16) & 0x0000FFFF0000FFFFULL);
+
+ x | = ((x & 0xFFFF0000FFFF0000) >> 16) * 0xFFFF;
+ y | = ((y & 0xFFFF0000FFFF0000) >> 16) * 0xFFFF;
+
+ x = (x & 0x0000FFFF0000FFFFULL) | ((y & 0x0000FFFF0000FFFFULL) << 16);
+
+ return M64C (x);
+}
+
+__inline __m64
+_m_pshufw (__m64 a, int n) /* unoptimized */
+{
+ unsigned short d[4] =
+ {
+ a.m64_u16[n & 3],
+ a.m64_u16[(n >> 2) & 3],
+ a.m64_u16[(n >> 4) & 3],
+ a.m64_u16[(n >> 6) & 3]
+ };
+
+ return M64C (d[0]);
+}
+
+__inline unsigned char
+sat16 (unsigned short d)
+{
+ if (d > 0xFF) return 0xFF;
+ else return d & 0xFF;
+}
+
+__inline __m64
+_m_packuswb (__m64 m1, __m64 m2) /* unoptimized */
+{
+ unsigned char d[8] =
+ {
+ sat16 (m1.m64_u16[0]),
+ sat16 (m1.m64_u16[1]),
+ sat16 (m1.m64_u16[2]),
+ sat16 (m1.m64_u16[3]),
+ sat16 (m2.m64_u16[0]),
+ sat16 (m2.m64_u16[1]),
+ sat16 (m2.m64_u16[2]),
+ sat16 (m2.m64_u16[3])
+ };
+
+ return M64C (d[0]);
+}
+
+__inline __m64 _m_punpcklbw (__m64 m1, __m64 m2) /* unoptimized */
+{
+ unsigned char d[8] =
+ {
+ m1.m64_u8[0],
+ m2.m64_u8[0],
+ m1.m64_u8[1],
+ m2.m64_u8[1],
+ m1.m64_u8[2],
+ m2.m64_u8[2],
+ m1.m64_u8[3],
+ m2.m64_u8[3],
+ };
+
+ return M64C (d[0]);
+}
+
+__inline __m64 _m_punpckhbw (__m64 m1, __m64 m2) /* unoptimized */
+{
+ unsigned char d[8] =
+ {
+ m1.m64_u8[4],
+ m2.m64_u8[4],
+ m1.m64_u8[5],
+ m2.m64_u8[5],
+ m1.m64_u8[6],
+ m2.m64_u8[6],
+ m1.m64_u8[7],
+ m2.m64_u8[7],
+ };
+
+ return M64C (d[0]);
+}
+
+__inline __m64 _m_psrlwi (__m64 a, int n) /* unoptimized */
+{
+ unsigned short d[4] =
+ {
+ a.m64_u16[0] >> n,
+ a.m64_u16[1] >> n,
+ a.m64_u16[2] >> n,
+ a.m64_u16[3] >> n
+ };
+
+ return M64C (d[0]);
+}
+
+__inline __m64 _m_psrlqi (__m64 m, int n)
+{
+ unsigned long long x = M64U (m) >> n;
+
+ return M64C (x);
+}
+
+__inline __m64 _m_psllqi (__m64 m, int n)
+{
+ unsigned long long x = M64U (m) << n;
+
+ return M64C (x);
+}
+
+#endif /* MMX_X64_H_INCLUDED */
More information about the xorg-commit
mailing list