Michael,
Tres asked me to help with the configure.in and Makefile.am bits for his
SSE2 code for x86_64. I've attached patches against Eterm CVS
with changes to: configure.in, src/Makefile.am, src/pixmap.c, and the
new file: src/sse2_cmod.c (changed since Tres's earlier version)
The configure.in changes add "--enable-sse2" which will be on by default
for x86_64. The incomplete tests for MMX_64 have been removed.
There was an odd "dps_snprintf_oflow()" line in configure.in which was
causing problems on my i686 system, and which I think was a typo, so I
removed it.
The src/Makefile.am changes just add the new source file sse2_cmod.c and
conditionally compile it when HAVE_SSE2 is defined.
The src/pixman.c changes add externs for the new sse2 function, and
conditionals to use them if HAVE_SSE2 is defined. Tres also added 8 or
16 byte alignment conditionally to colormod_trans() - this should
perhaps be done only for gcc?
Tres indicated that adding "-mpreferred-stack-boundary=16" might still
be beneficial on x86_64, but that it might consume extra space at run
time. I'm not in a position to make that call, but if you agree its a
good idea I can probably make the configure.in change.
I've tested the changes on xi686 and x86_64 and verified that the
correct routines get compiled in. I verified that the modified Eterms
run and that the Brightness control does something reasonable on both
systems. I have not done any performance tests.
All credit for the sse2_cmod.c code goes to Tres. I just did the easy
bits.
John
Index: configure.in
===================================================================
RCS file: /cvsroot/enlightenment/eterm/Eterm/configure.in,v
retrieving revision 1.92
diff -u -r1.92 configure.in
--- configure.in 1 May 2005 07:16:51 -0000 1.92
+++ configure.in 6 Jun 2005 14:44:01 -0000
@@ -220,7 +220,7 @@
seteuid memmove putenv strsep setresuid setresgid \
memmem usleep snprintf strcasestr strcasechr \
strcasepbrk strrev nl_langinfo)
-dps_snprintf_oflow()
+
AC_CHECK_LIB(m, pow)
dnl# Portability checks for various functions
@@ -500,9 +500,11 @@
AC_DEFINE(PIXMAP_OFFSET, , [Define for pseudo-transparency support.])
])
+dnl#
+dnl# MMX support
+dnl#
AC_MSG_CHECKING(for MMX support)
HAVE_MMX=""
-HAVE_MMX_64=""
AC_ARG_ENABLE(mmx, [ --enable-mmx enable MMX assembly routines], [
test "x$enableval" = "xyes" && HAVE_MMX="yes"
], [
@@ -510,25 +512,39 @@
i*86)
grep mmx /proc/cpuinfo >/dev/null 2>&1 && HAVE_MMX="yes"
;;
- x86_64)
- grep mmx /proc/cpuinfo >/dev/null 2>&1 && HAVE_MMX_64="yes"
- ;;
esac
])
if test "x$HAVE_MMX" = "xyes"; then
AC_MSG_RESULT([yes (32-bit)])
AC_DEFINE(HAVE_MMX, , [Define for 32-bit MMX support.])
-elif test "x$HAVE_MMX_64" = "xyes"; then
- dnl# AC_MSG_RESULT([yes (64-bit)])
- dnl# AC_DEFINE(HAVE_MMX_64, , [Define for 64-bit MMX support.])
- AC_MSG_RESULT([no (64-bit MMX not yet supported)])
else
AC_MSG_RESULT([no (no MMX detected)])
fi
-dnl# AM_CONDITIONAL(HAVE_MMX, test "x$HAVE_MMX" = "xyes" -o "x$HAVE_MMX_64" = "xyes")
AM_CONDITIONAL(HAVE_MMX, test "x$HAVE_MMX" = "xyes")
dnl#
+dnl# SSE2 support
+dnl#
+AC_MSG_CHECKING(for SSE2 support)
+HAVE_SSE2=""
+AC_ARG_ENABLE(sse2, [ --enable-sse2 enable SSE2 assembly routines], [
+ test "x$enableval" = "xyes" && HAVE_SSE2="yes"
+ ], [
+ case $host_cpu in
+ x86_64)
+ grep sse2 /proc/cpuinfo >/dev/null 2>&1 && HAVE_SSE2="yes"
+ ;;
+ esac
+ ])
+if test "x$HAVE_SSE2" = "xyes"; then
+ AC_MSG_RESULT([yes])
+ AC_DEFINE(HAVE_SSE2, , [Define for 64-bit SSE2 support.])
+else
+ AC_MSG_RESULT([no (no SSE2 detected)])
+fi
+AM_CONDITIONAL(HAVE_SSE2, test "x$HAVE_SSE2" = "xyes")
+
+dnl#
dnl# LibAST
dnl#
LIBAST_MIN=5
Index: src/Makefile.am
===================================================================
RCS file: /cvsroot/enlightenment/eterm/Eterm/src/Makefile.am,v
retrieving revision 1.29
diff -u -r1.29 Makefile.am
--- src/Makefile.am 15 Mar 2005 21:48:01 -0000 1.29
+++ src/Makefile.am 6 Jun 2005 14:44:01 -0000
@@ -6,6 +6,9 @@
MMX_SRCS = mmx_cmod.S
MMX_OBJS = mmx_cmod.lo
+SSE2_SRCS = sse2_cmod.c
+SSE2_OBJS = sse2_cmod.lo
+
libEterm_la_SOURCES = actions.c actions.h buttons.c buttons.h command.c \
command.h draw.c draw.h e.c e.h eterm_debug.h eterm_utmp.h \
events.c events.h feature.h font.c font.h grkelot.c \
@@ -16,22 +19,27 @@
timer.c timer.h utmp.c windows.c windows.h defaultfont.c \
defaultfont.h libscream.c scream.h screamcfg.h
-EXTRA_libEterm_la_SOURCES = $(MMX_SRCS)
+EXTRA_libEterm_la_SOURCES = $(MMX_SRCS) $(SSE2_SRCS)
libEterm_la_LDFLAGS = -release $(VERSION)
+if HAVE_SSE2
+libEterm_la_DEPENDENCIES = feature.h $(SSE2_OBJS)
+libEterm_la_LIBADD = $(SSE2_OBJS)
+else
if HAVE_MMX
libEterm_la_DEPENDENCIES = feature.h $(MMX_OBJS)
libEterm_la_LIBADD = $(MMX_OBJS)
else
libEterm_la_DEPENDENCIES = feature.h
endif
+endif
Eterm_SOURCES = main.c
Eterm_DEPENDENCIES = libEterm.la
Eterm_LDFLAGS = -rpath $(libdir):$(pkglibdir)
Eterm_LDADD = libEterm.la
-EXTRA_DIST = mmx_cmod.S
+EXTRA_DIST = mmx_cmod.S sse2_cmod.c
install-exec-hook:
$(mkinstalldirs) $(DESTDIR)$(pkgdatadir)
Index: src/pixmap.c
===================================================================
RCS file: /cvsroot/enlightenment/eterm/Eterm/src/pixmap.c,v
retrieving revision 1.112
diff -u -r1.112 pixmap.c
--- src/pixmap.c 10 May 2005 18:59:50 -0000 1.112
+++ src/pixmap.c 6 Jun 2005 14:44:03 -0000
@@ -57,11 +57,16 @@
# define WORDS_BIGENDIAN 0
#endif
-/* Assembler routines */
+/* Assembler routines for 32 bit cpu with mmx */
extern void shade_ximage_15_mmx(void *data, int bpl, int w, int h, int rm, int gm, int bm);
extern void shade_ximage_16_mmx(void *data, int bpl, int w, int h, int rm, int gm, int bm);
extern void shade_ximage_32_mmx(void *data, int bpl, int w, int h, int rm, int gm, int bm);
+/* Assembler routines for 64 bit cpu with sse2 */
+extern void shade_ximage_15_sse2(void *data, int bpl, int w, int h, int rm, int gm, int bm);
+extern void shade_ximage_16_sse2(void *data, int bpl, int w, int h, int rm, int gm, int bm);
+extern void shade_ximage_32_sse2(void *data, int bpl, int w, int h, int rm, int gm, int bm);
+
#ifdef PIXMAP_SUPPORT
static Imlib_Border bord_none = { 0, 0, 0, 0 };
#endif
@@ -1548,7 +1553,7 @@
/* New optimized routines for tinting XImages written by Willem Monsuwe <[EMAIL PROTECTED]> */
-#ifndef HAVE_MMX
+#if !defined HAVE_MMX && !defined HAVE_SSE2
/* RGB 15 */
static void
shade_ximage_15(void *data, int bpl, int w, int h, int rm, int gm, int bm)
@@ -1740,7 +1745,13 @@
colormod_trans(Pixmap p, imlib_t *iml, GC gc, unsigned short w, unsigned short h)
{
+#ifdef HAVE_SSE2
+ XImage * __attribute__ ((aligned (16))) ximg;
+#elif defined HAVE_MMX
+ XImage * __attribute__ ((aligned (8))) ximg;
+#else
XImage *ximg;
+#endif
register unsigned long i;
#if 0
@@ -1845,7 +1856,10 @@
/* Determine bitshift and bitmask values */
switch (real_depth) {
case 15:
-#ifdef HAVE_MMX
+#ifdef HAVE_SSE2
+ D_PIXMAP(("Using SSE2 - 15 bit\n"));
+ shade_ximage_15_sse2(ximg->data, ximg->bytes_per_line, w, h, rm, gm, bm);
+#elif defined HAVE_MMX
D_PIXMAP(("Using MMX - 15 bit\n"));
shade_ximage_15_mmx(ximg->data, ximg->bytes_per_line, w, h, rm, gm, bm);
#else
@@ -1854,7 +1868,10 @@
#endif
break;
case 16:
-#ifdef HAVE_MMX
+#ifdef HAVE_SSE2
+ D_PIXMAP(("Using SSE2 - 16 bit\n"));
+ shade_ximage_16_sse2(ximg->data, ximg->bytes_per_line, w, h, rm, gm, bm);
+#elif defined HAVE_MMX
D_PIXMAP(("Using MMX - 16 bit\n"));
shade_ximage_16_mmx(ximg->data, ximg->bytes_per_line, w, h, rm, gm, bm);
#else
@@ -1869,7 +1886,10 @@
}
/* drop */
case 32:
-#ifdef HAVE_MMX
+#ifdef HAVE_SSE2
+ D_PIXMAP(("Using SSE2 - 32 bit\n"));
+ shade_ximage_32_sse2(ximg->data, ximg->bytes_per_line, w, h, rm, gm, bm);
+#elif defined HAVE_MMX
D_PIXMAP(("Using MMX - 32 bit\n"));
shade_ximage_32_mmx(ximg->data, ximg->bytes_per_line, w, h, rm, gm, bm);
#else
/* File: sse2_cmod.c
* Written and Copyright (C) 2005 by Tres Melton
*
* Permission is hereby granted to Michael Jennings to license this code as
* he sees fit. I'd prefer the GPL but he will choose the BSD. The debate
* is moot as this is to become a part of the Eterm project, for which he is
* the primary author. For users of this code I ask that any modifications
* be released back into the community but with Michael Jennings chooses the
* BSD license then that request has no backing in law.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
* IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
*
* Much inspiration was drawn from the original x86 MMX port written by
* Willem Monsuwe <[EMAIL PROTECTED]> in pure x86/MMX Assembly. The MMX
* instructions are taken almost verbatim but the memory and parameter
* accessing had to be completely reworked for the x86_64 ABI and to
* ensure they worked with various gcc options. Further the code was
* extended to take advantage of the 128 bit xmm registers in SSE2.
*
* Manuals used in this port:
* The Gnu Assembler
* http://www.gnu.org/software/binutils/manual/gas-2.9.1/html_mono/as.html
* AMD64 Architecture Programmer's Manual Volume 1: Application Programming
* http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/24592.pdf
* AMD64 Architecture Programmer's Manual Volume 2: System Programming
* http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/24593.pdf
* AMD64 Architecture Programmer's Manual Volume 3: General-Purpose and System Instructions
* http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/24594.pdf
* AMD64 Architecture Programmer's Manual Volume 4: 128-Bit Media Instructions
* http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26568.pdf
* AMD64 Architecture Programmer's Manual Volume 5: 64-Bit Media and x87 Floating-Point Instructions
* http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26569.pdf
* AMD64 Application Binary Interface (v. 0.95)
* http://www.x86-64.org/documentation/documentation/abi-0.95.pdf
*
* The 32 bit color modification algorithm is simple but so optimized
* (even the C version) that it is almost unreadable.
* Therefore the pseudo code is:
*
* for each color of each pixel
* new_color = color * modifier
* if ( new_color > max_color_value )
* new_color = max_color_value
* end if
* end for
*
* The AMD64 ABI is at version 0.95 and might change in the future. Further it has changed a
* number of times in the past (although mostly in 2002-2003) as evidenced by the mailing
* list on http://www.x86-64.org. The GCC, Glibc, and Linux kernel have changed as well
* during this time to keep up. The standard C definition states that function parameters
* are to be passed on the stack but that can be very inefficient compared to passing them
* in registers so gcc tries to use registers. This is very different than on the register
* starved i386 architecture (AMD64 adds 8 general purpose registers: %r8-%r15, and
* MMX/SSE2/F87 adds 16 64/128/80bit registers: %xmm0-%xmm15. The x86_64 version of GCC uses
* registers as efficiently as possible and as a result exactly which registers are used
* for which parameters has evolved. Since all of these tools change simultaneously inline
* assembly code in C functions is the only way to ensure that this code will continue to
* function through a (however unlikely) change. If pure assembly were to be used as the
* original MMX author, Willem Monsuwe, did and the ABI changed then this code would cease
* to function properly. After examination of Willem's code I'm wondering if he
* wrote it for GNU/Linux originally. The ENTER and LEAVE macros put all of the
* parameters on the stack so that they can be accessed by references to the Base_Pointer
* the way that ANSI C is defined. If he originally wrote this for GNU/Linux then he most
* likely would have just used the registers instead of unwinding those optimizations
* manually by pushing them to the stack. And if he explicitly wanted to use the stack
* then there are parameters to gcc that would have performed those operations for him:
* -mregparm/-mmemparm. Other gcc options that can tweak with the stack and the number
* of registers available for function parameters are: -fcall-used/-fcall-saved,
* -fcaller-saves, -fstack-protector, -fPIC/-fpic, -mno-push-args, etc.. It might be
* advisable to check for these switches when using the original MMX code and emit a
* warning if any are enabled. I know that the PIC option trashes the BX register and
* that both Willem and I use that register. In other words If you do manage to get it
* to compile & run w/ -fpic it WILL break. On the plus side, you can keep the pieces! :-)
* On the other hand I could be wrong about everything
*
* In Conclusion:
* Using C functions and inline assembly code should alleviate all of the concerns as the
* C compiler will ensure that the parameters get to the function in a gauranteed manner
* and the inline assembly explicitly loads them into the desired registers for the assembly
* code. This might seem like alot of overhead but great care has been taken to adhere to
* the x86_64 ABI so that gcc/gas/ld will not perform any unneeded operations even when no
* optimizations have been enabled (-O[123]).
*/
#include "config.h"
#ifdef HAVE_SSE2
void shade_ximage_15_sse2( volatile void *data, volatile int bpl, volatile int w, volatile int h, volatile int rm, volatile int gm, volatile int bm )
{
__asm__ __volatile__ (
".align 16 \n\t" /* SIMD instructions should be aligned on 16 byte (128 bit) boundraries for performance reasons.*/
"leaq -14(%%rsi, %%rbx, 2), %%rsi\n\t" /* Load the stack index register with a pointer to data + ( width * bytes/pixel ) -6 */
"negq %%rbx \n\t" /* Negate the width to that we can increment the counter */
"jz 10f \n\t" /* Jump to end if the line count is zero */
"movd %[red_mod], %%xmm5 \n\t" /* Load the color modifiers into mmx registers */
"movd %[green_mod], %%xmm6 \n\t" /* " " */
"movd %[blue_mod], %%xmm7 \n\t" /* " " */
"punpcklwd %%xmm5, %%xmm5 \n\t" /* Unpack and Interleave low words. From A64_128bit_Media_Programming (p. 380) */
"punpcklwd %%xmm6, %%xmm6 \n\t" /* Duplicate the bottom 16 bits into the next 16 bits (both operands are the same) */
"punpcklwd %%xmm7, %%xmm7 \n\t"
"punpckldq %%xmm5, %%xmm5 \n\t" /* Unpack and Interleave low double words. From A64_128bit_Media_Programming (p. 376) */
"punpckldq %%xmm6, %%xmm6 \n\t" /* Duplicate the bottom 32 bits into the next 32 bits (both operands are the same) */
"punpckldq %%xmm7, %%xmm7 \n\t"
"punpcklqdq %%xmm5, %%xmm5 \n\t" /* Unpack and Interleave low quad words. From A64_128bit_Media_Programming (p. 378) */
"punpcklqdq %%xmm6, %%xmm6 \n\t" /* Duplicate the bottom 64 bits into the next 64 bits (both operands are the same) */
"punpcklqdq %%xmm7, %%xmm7 \n\t"
"or %[red_mod], %[green_mod] \n\t" /* This, and the following 4 instructions, check to see if all three colormodifiers are */
"or %[blue_mod], %[green_mod] \n\t" /* less than 256. If any of the modifiers are > 256 then they will have the 9th, or higher, */
"sar $8, %[green_mod] \n\t" /* bit set. Then we shift off eight bits, leaving something set if a modifier > 256. */
"movq %%rax, %[blue_mod] \n\t" /* Use the register named blue_mod to now store bytes_per_line. */
"xor %[red_mod], %[red_mod] \n\t" /* zero red so we don't have to load an immediate value for the following compare. */
"cmp %[red_mod], %[green_mod] \n\t" /* Compare the left over bits to zero */
"jg 5f \n\t" /* If one of the colors (might) need saturated then jump to the secondary set of loops. */
"1: \n\t" /* Start of the outer loop (lines). */
"movq %%rbx, %%rcx \n\t" /* Move the width into the count register */
"addq $7, %%rcx \n\t"
"jns 3f \n\t"
"2: \n\t" /* Start of the inner loop (pixels 8 at a time --> 8 * 16 = 128bits/xmm register ) */
"movdqu (%%rsi, %%rcx, 2), %%xmm0\n\t" /* Load the 16 bits of the pixel (5 bits for red, 6 bits for green, 5 bits for blue) */
"movdqu %%xmm0, %%xmm1 \n\t" /* Create a copy of the pixel for the green color */
"movdqu %%xmm0, %%xmm2 \n\t" /* Create a copy of the pixel for the blue color */
"psrlw $5, %%xmm1 \n\t" /* Packed Shift Right Logical Words */
/* From A64_128bit_Media_Programming (p. 347) */
/* Shifts the blue off of the green color */
"psrlw $10, %%xmm0 \n\t" /* Shifts the blue & green off of the red color */
"psllw $11, %%xmm2 \n\t" /* Packed Shift Left Logical Words */
/* From A64_128bit_Media_Programming (p. 330) */
/* Shifts the red & green off of the blue color */
"psllw $11, %%xmm1 \n\t" /* Shifts the red off of the green color */
"psllw $8, %%xmm0 \n\t" /* Shifts the red color into position */
"psrlw $3, %%xmm1 \n\t" /* Shifts the green color into position */
"psrlw $3, %%xmm2 \n\t" /* Shifts the blue color into position */
"pmulhw %%xmm5, %%xmm0 \n\t" /* color *= modifier */
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
"psllw $10, %%xmm0 \n\t" /* Shift red back into its original position */
"psllw $5, %%xmm1 \n\t" /* Shift green back into its original position */
"por %%xmm2, %%xmm0 \n\t" /* Mesh the colors back together */
"por %%xmm1, %%xmm0 \n\t"
"movdqu %%xmm0, (%%rsi, %%rcx, 2)\n\t" /* Place the shaded 8 pixels back into the image map */
"addq $8, %%rcx \n\t"
"js 2b \n\t"
"jmp 4f \n\t"
"3: \n\t" /* Deal with pixels one at a time here. */
"movw (%%rsi, %%rcx, 2), %%ax \n\t"
"movd %%eax, %%xmm0 \n\t"
"movq %%xmm0, %%xmm1 \n\t"
"movq %%xmm0, %%xmm2 \n\t"
"psrlw $5, %%xmm1 \n\t"
"psrlw $10, %%xmm0 \n\t"
"psllw $11, %%xmm2 \n\t"
"psllw $11, %%xmm1 \n\t"
"psllw $8, %%xmm0 \n\t"
"psrlw $3, %%xmm1 \n\t"
"psrlw $3, %%xmm2 \n\t"
"pmulhw %%xmm5, %%xmm0 \n\t"
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
"psllw $10, %%xmm0 \n\t"
"psllw $5, %%xmm1 \n\t"
"por %%xmm2, %%xmm0 \n\t"
"por %%xmm1, %%xmm0 \n\t"
"movd %%xmm0, %%eax \n\t"
"movw %%ax, (%%rsi, %%rcx, 2) \n\t"
"incq %%rcx \n\t"
"4: \n\t"
"cmpq $6, %%rcx \n\t"
"jng 3b \n\t"
"addq %[blue_mod], %%rsi \n\t" /* Blue_mod is the name of a register that now contains bytes_per_line. */
"decq %%rdx \n\t"
"jnz 1b \n\t"
"jmp 10f \n\t" /* We're done! */
"5: \n\t" /* Saturation is required */
"pcmpeqw %%xmm3, %%xmm3 \n\t" /* Packed Compare Equal Words */
/* From A64_128bit_Media_Programming (p. 276) */
/* This sets xmm3 to 128 1's (since mm6 = mm6) */
"psllw $5, %%xmm3 \n\t" /* xmm3 = 8 copies of 1111 1111 1110 0000 */
"6: \n\t"
"movq %%rbx, %%rcx \n\t"
"addq $7, %%rcx \n\t"
"jns 8f \n\t"
"7: \n\t"
"movdqu (%%rsi, %%rcx, 2), %%xmm0\n\t"
"movdqu %%xmm0, %%xmm1 \n\t"
"movdqu %%xmm0, %%xmm2 \n\t"
"psrlw $5, %%xmm1 \n\t"
"psrlw $10, %%xmm0 \n\t"
"psllw $11, %%xmm2 \n\t"
"psllw $11, %%xmm1 \n\t"
"psllw $8, %%xmm0 \n\t"
"psrlw $3, %%xmm1 \n\t"
"psrlw $3, %%xmm2 \n\t"
"pmulhw %%xmm5, %%xmm0 \n\t"
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
"paddusw %%xmm3, %%xmm0 \n\t"
"paddusw %%xmm3, %%xmm1 \n\t"
"paddusw %%xmm3, %%xmm2 \n\t"
"psubw %%xmm3, %%xmm0 \n\t" /* FIXME: This line needs added to the original asm code */
"psubw %%xmm3, %%xmm1 \n\t"
"psubw %%xmm3, %%xmm2 \n\t"
"psllw $10, %%xmm0 \n\t"
"psllw $5, %%xmm1 \n\t"
"por %%xmm2, %%xmm0 \n\t"
"por %%xmm1, %%xmm0 \n\t"
"movdqu %%xmm0, (%%rsi, %%rcx, 2)\n\t"
"addq $8, %%rcx \n\t"
"js 7b \n\t"
"jmp 9f \n\t"
"8: \n\t"
"movw (%%rsi, %%rcx, 2), %%ax \n\t"
"movd %%eax, %%xmm0 \n\t"
"movq %%xmm0, %%xmm1 \n\t"
"movq %%xmm0, %%xmm2 \n\t"
"psrlw $5, %%xmm1 \n\t"
"psrlw $10, %%xmm0 \n\t"
"psllw $11, %%xmm2 \n\t"
"psllw $11, %%xmm1 \n\t"
"psllw $8, %%xmm0 \n\t"
"psrlw $3, %%xmm1 \n\t"
"psrlw $3, %%xmm2 \n\t"
"pmulhw %%xmm5, %%xmm0 \n\t"
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
"paddusw %%xmm3, %%xmm0 \n\t"
"paddusw %%xmm3, %%xmm1 \n\t"
"paddusw %%xmm3, %%xmm2 \n\t"
"psubw %%xmm3, %%xmm0 \n\t" /* FIXME: This line needs added to the original asm code */
"psubw %%xmm3, %%xmm1 \n\t"
"psubw %%xmm3, %%xmm2 \n\t"
"psllw $10, %%xmm0 \n\t"
"psllw $5, %%xmm1 \n\t"
"por %%xmm2, %%xmm0 \n\t"
"por %%xmm1, %%xmm0 \n\t"
"movd %%xmm0, %%eax \n\t"
"movw %%ax, (%%rsi, %%rcx, 2) \n\t"
"incq %%rcx \n\t"
"9: \n\t"
"cmpq $6, %%rcx \n\t"
"jng 8b \n\t"
"addq %[blue_mod], %%rsi \n\t" /* Blue_mod is the name of a register that now contains bytes_per_line. */
"decq %%rdx \n\t"
"jnz 6b \n\t"
"10: \n\t" /* This is the end. Jump here if the line count is zero. */
"emms \n\t" /* exit multi-media state (last asm instruction) */
: /* outputs: none */
/* inputs: (many operations cannot be performed with a mix of 32bit & 64bit operands directly) */
/* (however the compiler/assembler can preload 32bit values into 64bit registers) */
/* (that is why certain variables cannot be referenced by name -- use their register) */
: [data] "S" (data), /* put the pointer data into the rsi register */
[width] "b" (w), /* put the width in the %rbx register (cannot be referenced by name) */
[height] "d" (h), /* put the heigth in the %rdx register (cannot be referenced by name) */
[red_mod] "r" ((unsigned long)(rm)),/* put the red_modifier in a register (referenced by name) */
[green_mod] "r" ((unsigned long)(gm)),/* put the green_modifier in a register (referenced by name) */
[blue_mod] "r" ((unsigned long)(bm)),/* put the blue_modifier in a register (referenced by name) Later store the bytes_line here */
[bytes_line] "a" (bpl) /* put the bytes_per_line in the %rax register (cannot be referenced by name) */
: "memory" /* clobbers: (memory includes all the registers) */
); /* End of Assembly */
}
void shade_ximage_16_sse2( volatile void *data, volatile int bpl, volatile int w, volatile int h, volatile int rm, volatile int gm, volatile int bm )
{
__asm__ __volatile__ (
".align 16 \n\t" /* SIMD instructions should be aligned on 16 byte (128 bit) boundraries for performance reasons.*/
"leaq -14(%%rsi, %%rbx, 2), %%rsi\n\t" /* Load the stack index register with a pointer to data + ( width * bytes/pixel ) -6 */
"negq %%rbx \n\t" /* Negate the width to that we can increment the counter */
"jz 10f \n\t" /* Jump to end if the line count is zero */
"movd %[red_mod], %%xmm5 \n\t" /* Load the color modifiers into mmx registers */
"movd %[green_mod], %%xmm6 \n\t" /* " " */
"movd %[blue_mod], %%xmm7 \n\t" /* " " */
"punpcklwd %%xmm5, %%xmm5 \n\t" /* Unpack and Interleave low words. From A64_128bit_Media_Programming (p. 380) */
"punpcklwd %%xmm6, %%xmm6 \n\t" /* Duplicate the bottom 16 bits into the next 16 bits (both operands are the same) */
"punpcklwd %%xmm7, %%xmm7 \n\t"
"punpckldq %%xmm5, %%xmm5 \n\t" /* Unpack and Interleave low double words. From A64_128bit_Media_Programming (p. 376) */
"punpckldq %%xmm6, %%xmm6 \n\t" /* Duplicate the bottom 32 bits into the next 32 bits (both operands are the same) */
"punpckldq %%xmm7, %%xmm7 \n\t"
"punpcklqdq %%xmm5, %%xmm5 \n\t" /* Unpack and Interleave low quad words. From A64_128bit_Media_Programming (p. 378) */
"punpcklqdq %%xmm6, %%xmm6 \n\t" /* Duplicate the bottom 64 bits into the next 64 bits (both operands are the same) */
"punpcklqdq %%xmm7, %%xmm7 \n\t"
"or %[red_mod], %[green_mod] \n\t" /* This, and the following 4 instructions, check to see if all three colormodifiers are */
"or %[blue_mod], %[green_mod] \n\t" /* less than 256. If any of the modifiers are > 256 then they will have the 9th, or higher, */
"sar $8, %[green_mod] \n\t" /* bit set. Then we shift off eight bits, leaving something set if a modifier > 256. */
"movq %%rax, %[blue_mod] \n\t" /* Use the register named blue_mod to now store bytes_per_line. */
"xor %[red_mod], %[red_mod] \n\t" /* zero red so we don't have to load an immediate value for the following compare. */
"cmp %[red_mod], %[green_mod] \n\t" /* Compare the left over bits to zero */
"jg 5f \n\t" /* If one of the colors (might) need saturated then jump to the secondary set of loops. */
"1: \n\t" /* Start of the outer loop (lines). */
"movq %%rbx, %%rcx \n\t" /* Move the width into the count register */
"addq $7, %%rcx \n\t"
"jns 3f \n\t"
"2: \n\t" /* Start of the inner loop (pixels 8 at a time --> 8 * 16 = 128bits/xmm register ) */
"movdqu (%%rsi, %%rcx, 2), %%xmm0\n\t" /* Load the 16 bits of the pixel (5 bits for red, 6 bits for green, 5 bits for blue) */
"movdqu %%xmm0, %%xmm1 \n\t" /* Create a copy of the pixel for the green color */
"movdqu %%xmm0, %%xmm2 \n\t" /* Create a copy of the pixel for the blue color */
"psrlw $5, %%xmm1 \n\t" /* Packed Shift Right Logical Words */
/* From A64_128bit_Media_Programming (p. 347) */
/* Shifts the blue off of the green color */
"psrlw $11, %%xmm0 \n\t" /* Shifts the blue & green off of the red color */
"psllw $11, %%xmm2 \n\t" /* Packed Shift Left Logical Words */
/* From A64_128bit_Media_Programming (p. 330) */
/* Shifts the red & green off of the blue color */
"psllw $10, %%xmm1 \n\t" /* Shifts the red off of the green color */
"psllw $8, %%xmm0 \n\t" /* Shifts the red color into position */
"psrlw $2, %%xmm1 \n\t" /* Shifts the green color into position */
"psrlw $3, %%xmm2 \n\t" /* Shifts the blue color into position */
"pmulhw %%xmm5, %%xmm0 \n\t" /* color *= modifier */
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
"psllw $11, %%xmm0 \n\t" /* Shift red back into its original position */
"psllw $5, %%xmm1 \n\t" /* Shift green back into its original position */
"por %%xmm2, %%xmm0 \n\t" /* Mesh the colors back together */
"por %%xmm1, %%xmm0 \n\t"
"movdqu %%xmm0, (%%rsi, %%rcx, 2)\n\t" /* Place the shaded 8 pixels back into the image map */
"addq $8, %%rcx \n\t"
"js 2b \n\t"
"jmp 4f \n\t"
"3: \n\t" /* Deal with pixels one at a time here. */
"movw (%%rsi, %%rcx, 2), %%ax \n\t"
"movd %%eax, %%xmm0 \n\t"
"movq %%xmm0, %%xmm1 \n\t"
"movq %%xmm0, %%xmm2 \n\t"
"psrlw $5, %%xmm1 \n\t"
"psrlw $11, %%xmm0 \n\t"
"psllw $11, %%xmm2 \n\t"
"psllw $10, %%xmm1 \n\t"
"psllw $8, %%xmm0 \n\t"
"psrlw $2, %%xmm1 \n\t"
"psrlw $3, %%xmm2 \n\t"
"pmulhw %%xmm5, %%xmm0 \n\t"
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
"psllw $11, %%xmm0 \n\t"
"psllw $5, %%xmm1 \n\t"
"por %%xmm2, %%xmm0 \n\t"
"por %%xmm1, %%xmm0 \n\t"
"movd %%xmm0, %%eax \n\t"
"movw %%ax, (%%rsi, %%rcx, 2) \n\t"
"incq %%rcx \n\t"
"4: \n\t"
"cmpq $6, %%rcx \n\t"
"jng 3b \n\t"
"addq %[blue_mod], %%rsi \n\t" /* Blue_mod is the name of a register that now contains bytes_per_line. */
"decq %%rdx \n\t"
"jnz 1b \n\t"
"jmp 10f \n\t" /* We're done! */
"5: \n\t" /* Saturation is required */
"pcmpeqw %%xmm3, %%xmm3 \n\t" /* Packed Compare Equal Words */
/* From A64_128bit_Media_Programming (p. 276) */
/* This sets xmm3 to 128 1's (since mm6 = mm6) */
"movdqu %%xmm3, %%xmm4 \n\t" /* Make copy of 128 ones */
"psllw $5, %%xmm3 \n\t" /* xmm3 = 8 copies of 1111 1111 1110 0000 */
"psllw $6, %%xmm4 \n\t" /* xmm4 = 8 copies of 1111 1111 1100 0000 */
"6: \n\t"
"movq %%rbx, %%rcx \n\t"
"addq $7, %%rcx \n\t"
"jns 8f \n\t"
"7: \n\t"
"movdqu (%%rsi, %%rcx, 2), %%xmm0\n\t"
"movdqu %%xmm0, %%xmm1 \n\t"
"movdqu %%xmm0, %%xmm2 \n\t"
"psrlw $5, %%xmm1 \n\t"
"psrlw $11, %%xmm0 \n\t"
"psllw $11, %%xmm2 \n\t"
"psllw $10, %%xmm1 \n\t"
"psllw $8, %%xmm0 \n\t"
"psrlw $2, %%xmm1 \n\t"
"psrlw $3, %%xmm2 \n\t"
"pmulhw %%xmm5, %%xmm0 \n\t"
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
"paddusw %%xmm3, %%xmm0 \n\t"
"paddusw %%xmm4, %%xmm1 \n\t"
"paddusw %%xmm3, %%xmm2 \n\t"
"psubw %%xmm4, %%xmm1 \n\t"
"psubw %%xmm3, %%xmm2 \n\t"
"psllw $11, %%xmm0 \n\t"
"psllw $5, %%xmm1 \n\t"
"por %%xmm2, %%xmm0 \n\t"
"por %%xmm1, %%xmm0 \n\t"
"movdqu %%xmm0, (%%rsi, %%rcx, 2)\n\t"
"addq $8, %%rcx \n\t"
"js 7b \n\t"
"jmp 9f \n\t"
"8: \n\t"
"movw (%%rsi, %%rcx, 2), %%ax \n\t"
"movd %%eax, %%xmm0 \n\t"
"movq %%xmm0, %%xmm1 \n\t"
"movq %%xmm0, %%xmm2 \n\t"
"psrlw $5, %%xmm1 \n\t"
"psrlw $11, %%xmm0 \n\t"
"psllw $11, %%xmm2 \n\t"
"psllw $10, %%xmm1 \n\t"
"psllw $8, %%xmm0 \n\t"
"psrlw $2, %%xmm1 \n\t"
"psrlw $3, %%xmm2 \n\t"
" \n\t"
"pmulhw %%xmm5, %%xmm0 \n\t"
"pmulhw %%xmm6, %%xmm1 \n\t"
"pmulhw %%xmm7, %%xmm2 \n\t"
" \n\t"
"paddusw %%xmm3, %%xmm0 \n\t"
"paddusw %%xmm4, %%xmm1 \n\t"
"paddusw %%xmm3, %%xmm2 \n\t"
" \n\t"
"psubw %%xmm4, %%xmm1 \n\t"
"psubw %%xmm3, %%xmm2 \n\t"
" \n\t"
"psllw $11, %%xmm0 \n\t"
"psllw $5, %%xmm1 \n\t"
"por %%xmm2, %%xmm0 \n\t"
"por %%xmm1, %%xmm0 \n\t"
"movd %%xmm0, %%eax \n\t"
"movw %%ax, (%%rsi, %%rcx, 2) \n\t"
"incq %%rcx \n\t"
"9: \n\t"
"cmpq $6, %%rcx \n\t"
"jng 8b \n\t"
"addq %[blue_mod], %%rsi \n\t" /* Blue_mod is the name of a register that now contains bytes_per_line. */
"decq %%rdx \n\t"
"jnz 6b \n\t"
"10: \n\t" /* This is the end. Jump here if the line count is zero. */
"emms \n\t" /* exit multi-media state (last asm instruction) */
: /* outputs: none */
/* inputs: (many operations cannot be performed with a mix of 32bit & 64bit operands directly) */
/* (however the compiler/assembler can preload 32bit values into 64bit registers) */
/* (that is why certain variables cannot be referenced by name -- use their register) */
: [data] "S" (data), /* put the pointer data into the rsi register */
[width] "b" (w), /* put the width in the %rbx register (cannot be referenced by name) */
[height] "d" (h), /* put the heigth in the %rdx register (cannot be referenced by name) */
[red_mod] "r" ((unsigned long)(rm)),/* put the red_modifier in a register (referenced by name) */
[green_mod] "r" ((unsigned long)(gm)),/* put the green_modifier in a register (referenced by name) */
[blue_mod] "r" ((unsigned long)(bm)),/* put the blue_modifier in a register (referenced by name) Later store the bytes_line here */
[bytes_line] "a" (bpl) /* put the bytes_per_line in the %rax register (cannot be referenced by name) */
: "memory" /* clobbers: (memory includes all the registers) */
); /* End of Assembly */
}
void shade_ximage_32_sse2( volatile void *data, volatile int bpl, volatile int w, volatile int h, volatile int rm, volatile int gm, volatile int bm )
{
__asm__ __volatile__ (
".align 16 \n\t" /* SIMD instructions should be aligned on 16 byte (128 bit) boundraries for performance reasons.*/
"leaq -4(%%rsi, %%rbx, 4), %%rsi\n\t" /* From A64_General_Purpose_and_System_Instructions (p. 182) */
/* Intel syntax section:[base + index*scale + disp] (used by AMD manuals) */
/* AT&T syntax section:disp(base, index, scale) (used by gas/gcc) */
/* Load Effective Address of (rsi + (rbx * size)) into rsi */
/* 32 bits per pixel means a multiplier of 4. */
"negq %%rbx \n\t" /* two's compliment negation of ebx (width) and sets the Zero Flag based on the results */
/* From A64_General_Purpose_and_System_Instructions (p. 212) */
"jz 10f \n\t" /* Jump to label 3 forward on Zero */
/* Basically if width = 0 blowout */
/* I don't understand why the height isn't checked (shouldn't matter, zero loop iterations) */
"movd %[red_mod], %%xmm4 \n\t" /* move red modifier into mm4 w/ zero extension to 128bits */
/* RGB's are 8 bit values. regardless of them coming in in 32/64 bit they are zero extended */
"psllq $16, %%xmm4 \n\t" /* Packed Shift Left Logical Quad words (left shift mm4 16bits twice, once for each 64bit value)*/
/* From A64_128bit_Media_Programming (p. 328) */
"movd %[green_mod], %%xmm5 \n\t" /* move green modifier into mm5 w/ zero extension to 128bits */
"por %%xmm5, %%xmm4 \n\t" /* Mesh green modifier into color modifier */
"psllq $16, %%xmm4 \n\t" /* Packed Shift Left Logical Quad words (left shift mm4 16bits twice, once for each 64bit value)*/
"movd %[blue_mod], %%xmm5 \n\t" /* move blue modifier (32 bits) into mm4 w/ zero extension to 128bits */
"por %%xmm5, %%xmm4 \n\t" /* Mesh blue modifier into color modifier */
/* mm4 (color modifier) now contains 00 00 00 00 : 00 00 00 00 :: 00 00 00 rm : 00 gm 00 bm */
"punpcklqdq %%xmm4, %%xmm4 \n\t" /* Unpack and Interleave low quad words. From A64_128bit_Media_Programming (p. 378) */
/* Duplicate the bottom 64 bits into the next 64 bits (both operands are the same) */
"pcmpeqw %%xmm6, %%xmm6 \n\t" /* Packed Compare Equal Words */
/* From A64_128bit_Media_Programming (p. 276) */
/* This sets mm6 to 128 1's (since mm6 = mm6) */
"psllw $15, %%xmm6 \n\t" /* Packed Shift Left Logical Words */
/* From A64_128bit_Media_Programming (p. 330) */
/* This sets 8 16 bit values of 1000 0000 0000 0000 in the 128 bit word */
"movdqu %%xmm6, %%xmm5 \n\t" /* Copy mm6 to mm5 (we need mm6 later) */
"pmulhw %%xmm4, %%xmm5 \n\t" /* Packed Multiply High Signed Word */
/* mm4 = ( mm4 * mm5 ) >> 16 (8 times, once for each 16bit value) */
/* For each color_ modifier (cm) */
/* (( cm * 80 00 ) >> 16 ) = (( cm << 15 ) >> 16 ) = cm >> 1 */
"1: \n\t" /* The start of the outer loop (lines) */
"movq %%rbx, %%rcx \n\t" /* Load the counting register (rcx) with the width of the window to shade */
"incq %%rcx \n\t"
"2: \n\t" /* The start of the inner loop (columns) */
"movq (%%rsi, %%rcx, 4), %%xmm1 \n\t" /* sets mm1 to the 32bit color in the image map (data[ rcx ]) */
/* 32 bit color is still 4 bytes so leave the multiplier alone it is zero extended to 128 bits */
/* only move 32 bits with movd so we don't get two pixels worth of colors */
"pxor %%xmm0, %%xmm0 \n\t" /* 128bit exclusive or (sets mm0 to 0) */
"punpcklbw %%xmm1, %%xmm0 \n\t" /* Unpack and interleave low bytes */
/* For each color of the pixel expand to 16 bits and shift left 8 bits */
/* From A64_128bit_Media_Programming (p. 374) */
/* discard high 64 bits and expand both mm0 and mm1 a byte at a time into mm0 (mm0 first) */
"pxor %%xmm6, %%xmm0 \n\t" /* This flips the sign of the 16 bit red, green, and blue colors. (mm6 ~= 1000:0000 8 times) */
"pmulhw %%xmm4, %%xmm0 \n\t" /* Package Multiply High Signed Word (an SSE2 instruction) 128bit mm0=color mm4=cm */
/* Each 16 bit signed int in mm4 (8) is multiplied by the same in mm0 */
/* and the high 16 bits of the result replace the 16 bits used from mm0 */
/* For (( each 16 bit color * each 16 bit color modifier ) >> 16 ) */
"psubw %%xmm5, %%xmm0 \n\t" /* Packed Subtract Words */
/* From A64_128bit_Media_Programming (p. 364) */
/* mm0=modified color mm5=corrected color modifier. mm0 = ( mm0 - mm5 ) */
/* 16 bit corrected modified color = ( modified color - corrected color modifier ) */
"packuswb %%xmm0, %%xmm0 \n\t" /* Pack with Saturation Signed Word to Unsigned Byte */
/* From A64_128bit_Media_Programming (p. 246) */
/* if mm0 > 255 then mm0=255 elsif mm0 < 0 mm0=0 else mm0=mm0 */
/* The top 64 bits are now trashed. The remaining 64 bits are 2 pixels */
"movq %%xmm0, (%%rsi, %%rcx, 4) \n\t" /* puts the new 32 bit color value back into the data (image map) */
/* 32 bit color is still a double word so movd stays movd */
"addq $2, %%rcx \n\t" /* Increment the count register (more pixels left) */
"js 2b \n\t" /* Jump backwards to label 2 (restart inner loop) on negative (more pixels left) */
"jmp 5f \n\t" /* Jump to single pixel section after pairs are exhausted */
"4: \n\t" /* The start of the inner loop (columns) */
"movd (%%rsi, %%rcx, 4), %%xmm1 \n\t" /* sets mm1 to the 32bit color in the image map (data[ rcx ]) */
/* 32 bit color is still 4 bytes so leave the multiplier alone it is zero extended to 128 bits */
/* only move 32 bits with movd so we don't get two pixels worth of colors */
"pxor %%xmm0, %%xmm0 \n\t" /* 128bit exclusive or (sets mm0 to 0) */
"punpcklbw %%xmm1, %%xmm0 \n\t" /* Unpack and interleave low bytes */
/* For each color of the pixel expand to 16 bits and shift left 8 bits */
/* From A64_128bit_Media_Programming (p. 374) */
/* discard high 64 bits and expand both mm0 and mm1 a byte at a time into mm0 (mm0 first) */
"pxor %%xmm6, %%xmm0 \n\t" /* This flips the sign of the 16 bit red, green, and blue colors. (mm6 ~= 1000:0000 8 times) */
"pmulhw %%xmm4, %%xmm0 \n\t" /* Package Multiply High Signed Word (an SSE2 instruction) 128bit mm0=color mm4=cm */
/* Each 16 bit signed int in mm4 (8) is multiplied by the same in mm0 */
/* and the high 16 bits of the result replace the 16 bits used from mm0 */
/* For (( each 16 bit color * each 16 bit color modifier ) >> 16 ) */
"psubw %%xmm5, %%xmm0 \n\t" /* Packed Subtract Words */
/* From A64_128bit_Media_Programming (p. 364) */
/* mm0=modified color mm5=corrected color modifier. mm0 = ( mm0 - mm5 ) */
/* 16 bit corrected modified color = ( modified color - corrected color modifier ) */
"packuswb %%xmm0, %%xmm0 \n\t" /* Pack with Saturation Signed Word to Unsigned Byte */
/* From A64_128bit_Media_Programming (p. 246) */
/* if mm0 > 255 then mm0=255 elsif mm0 < 0 mm0=0 else mm0=mm0 */
"movd %%xmm0, (%%rsi, %%rcx, 4) \n\t" /* puts the new 32 bit color value back into the data (image map) */
/* 32 bit color is still a double word so movd stays movd */
"incq %%rcx \n\t" /* Increment the count register (more pixels left) */
"5: \n\t" /* Jump here after all pairs of pixels are exhausted */
"cmpq $0, %%rcx \n\t" /* Increment the count register (more pixels left) */
"jng 4b \n\t" /* Jump backwards to label 2 (restart inner loop) on NOT zero (more pixels left) */
"addq %%rax, %%rsi \n\t" /* Add bytes per line to the data pointer (advance the pointer to the next line) */
"decq %%rdx \n\t" /* Decrement the dx register (row count) */
"jnz 1b \n\t" /* Jump backwards to label 1 (restart outer loop) if not zero (more rows left) */
"10: \n\t" /* End of function (jump here to clean up and return to caller */
"emms \n\t" /* exit multi-media state (last asm instruction) */
: /* outputs: none */
/* inputs: (many operations cannot be performed with a mix of 32bit & 64bit operands directly) */
/* (however the compiler/assembler can preload 32bit values into 64bit registers) */
/* (that is why certain variables cannot be referenced by name -- use their register) */
: [data] "S" (data), /* put the pointer data into the rsi register */
[width] "b" (w), /* put the width in the %rbx register (cannot be referenced by name) */
[height] "d" (h), /* put the heigth in the %rdx register (cannot be referenced by name) */
[red_mod] "r" (rm), /* put the red_modifier in a register (referenced by name) */
[green_mod] "r" (gm), /* put the green_modifier in a register (referenced by name) */
[blue_mod] "r" (bm), /* put the blue_modifier in a register (referenced by name) */
[bytes_line] "a" (bpl) /* put the bytes_per_line in the %rax register (cannot be referenced by name) */
: "memory" /* clobbers: (memory includes all the registers) */
); /* End of Assembly */
}
#endif
