summaryrefslogtreecommitdiff
path: root/src/libffmpeg/libavcodec/i386
diff options
context:
space:
mode:
authorMiguel Freitas <miguelfreitas@users.sourceforge.net>2004-04-25 18:57:04 +0000
committerMiguel Freitas <miguelfreitas@users.sourceforge.net>2004-04-25 18:57:04 +0000
commita2a44876712f079610f0396fb9a682ea47e05b6e (patch)
tree1f3e328dfe6a5f9fa7c79e7a23bf6310be2827fd /src/libffmpeg/libavcodec/i386
parentacb7dc0f256afc24e875a168da989ef25d86b7b7 (diff)
downloadxine-lib-a2a44876712f079610f0396fb9a682ea47e05b6e.tar.gz
xine-lib-a2a44876712f079610f0396fb9a682ea47e05b6e.tar.bz2
ffmpeg sync
CVS patchset: 6437 CVS date: 2004/04/25 18:57:04
Diffstat (limited to 'src/libffmpeg/libavcodec/i386')
-rw-r--r--src/libffmpeg/libavcodec/i386/Makefile.am3
-rw-r--r--src/libffmpeg/libavcodec/i386/dsputil_mmx.c12
-rw-r--r--src/libffmpeg/libavcodec/i386/fdct_mmx.c21
-rw-r--r--src/libffmpeg/libavcodec/i386/mmx.h24
-rw-r--r--src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c803
-rw-r--r--src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c890
6 files changed, 1338 insertions, 415 deletions
diff --git a/src/libffmpeg/libavcodec/i386/Makefile.am b/src/libffmpeg/libavcodec/i386/Makefile.am
index d7b2bb4f8..eaa8d0f75 100644
--- a/src/libffmpeg/libavcodec/i386/Makefile.am
+++ b/src/libffmpeg/libavcodec/i386/Makefile.am
@@ -18,7 +18,8 @@ libavcodec_mmx_src = \
motion_est_mmx.c \
mpegvideo_mmx.c \
simple_idct_mmx.c \
- vp3dsp_mmx.c
+ vp3dsp_mmx.c \
+ vp3dsp_sse2.c
libavcodec_mmx_dummy = libavcodec_mmx_dummy.c
diff --git a/src/libffmpeg/libavcodec/i386/dsputil_mmx.c b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
index 15dc8eec2..772c9c1f0 100644
--- a/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
@@ -2147,9 +2147,15 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
}
/* VP3 optimized DSP functions */
- c->vp3_dsp_init = vp3_dsp_init_mmx;
- c->vp3_idct_put = vp3_idct_put_mmx;
- c->vp3_idct_add = vp3_idct_add_mmx;
+ if (mm_flags & MM_SSE2) {
+ c->vp3_dsp_init = vp3_dsp_init_sse2;
+ c->vp3_idct_put = vp3_idct_put_sse2;
+ c->vp3_idct_add = vp3_idct_add_sse2;
+ } else {
+ c->vp3_dsp_init = vp3_dsp_init_mmx;
+ c->vp3_idct_put = vp3_idct_put_mmx;
+ c->vp3_idct_add = vp3_idct_add_mmx;
+ }
#ifdef CONFIG_ENCODERS
c->get_pixels = get_pixels_mmx;
diff --git a/src/libffmpeg/libavcodec/i386/fdct_mmx.c b/src/libffmpeg/libavcodec/i386/fdct_mmx.c
index 877160773..7af576971 100644
--- a/src/libffmpeg/libavcodec/i386/fdct_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/fdct_mmx.c
@@ -50,7 +50,14 @@ static const long long fdct_one_corr ATTR_ALIGN(8) = 0x0001000100010001LL;
static const long fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW };
-static const long fdct_r_row_sse2[4] ATTR_ALIGN(16) = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};
+struct
+{
+ const long fdct_r_row_sse2[4] ATTR_ALIGN(16);
+} fdct_r_row_sse2 ATTR_ALIGN(16)=
+{{
+ RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
+}};
+//static const long fdct_r_row_sse2[4] ATTR_ALIGN(16) = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};
static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = { // forward_dct coeff table
16384, 16384, -8867, -21407,
@@ -126,7 +133,12 @@ static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = { // forward_dct coeff
6270, 26722, 6270, -17855,
};
-static const int16_t tab_frw_01234567_sse2[] ATTR_ALIGN(16) = { // forward_dct coeff table
+struct
+{
+ const int16_t tab_frw_01234567_sse2[256] ATTR_ALIGN(16);
+} tab_frw_01234567_sse2 ATTR_ALIGN(16) =
+{{
+//static const int16_t tab_frw_01234567_sse2[] ATTR_ALIGN(16) = { // forward_dct coeff table
#define TABLE_SSE2 C4, C4, C1, C3, -C6, -C2, -C1, -C5, \
C4, C4, C5, C7, C2, C6, C3, -C7, \
-C4, C4, C7, C3, C6, -C2, C7, -C5, \
@@ -252,7 +264,8 @@ TABLE_SSE2
#define C6 12299
#define C7 6270
TABLE_SSE2
-};
+}};
+
static always_inline void fdct_col(const int16_t *in, int16_t *out, int offset)
{
@@ -392,7 +405,7 @@ static always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
"FDCT_ROW_SSE2_H2 80 192 \n\t"
"FDCT_ROW_SSE2 80 \n\t"
:
- : "r" (in), "r" (tab_frw_01234567_sse2), "r" (fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out)
+ : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out)
);
}
diff --git a/src/libffmpeg/libavcodec/i386/mmx.h b/src/libffmpeg/libavcodec/i386/mmx.h
index 7e94cfd9b..ad684bc5a 100644
--- a/src/libffmpeg/libavcodec/i386/mmx.h
+++ b/src/libffmpeg/libavcodec/i386/mmx.h
@@ -240,4 +240,28 @@ typedef union {
#define sfence() __asm__ __volatile__ ("sfence\n\t")
+/* SSE2 */
+#define pshufhw_m2r(var,reg,imm) mmx_m2ri(pshufhw, var, reg, imm)
+#define pshufhw_r2r(regs,regd,imm) mmx_r2ri(pshufhw, regs, regd, imm)
+#define pshuflw_m2r(var,reg,imm) mmx_m2ri(pshuflw, var, reg, imm)
+#define pshuflw_r2r(regs,regd,imm) mmx_r2ri(pshuflw, regs, regd, imm)
+
+#define pshufd_r2r(regs,regd,imm) mmx_r2ri(pshufd, regs, regd, imm)
+
+#define movdqa_m2r(var,reg) mmx_m2r (movdqa, var, reg)
+#define movdqa_r2m(reg,var) mmx_r2m (movdqa, reg, var)
+#define movdqa_r2r(regs,regd) mmx_r2r (movdqa, regs, regd)
+#define movdqu_m2r(var,reg) mmx_m2r (movdqu, var, reg)
+#define movdqu_r2m(reg,var) mmx_r2m (movdqu, reg, var)
+#define movdqu_r2r(regs,regd) mmx_r2r (movdqu, regs, regd)
+
+#define pmullw_r2m(reg,var) mmx_r2m (pmullw, reg, var)
+
+#define pslldq_i2r(imm,reg) mmx_i2r (pslldq, imm, reg)
+#define psrldq_i2r(imm,reg) mmx_i2r (psrldq, imm, reg)
+
+#define punpcklqdq_r2r(regs,regd) mmx_r2r (punpcklqdq, regs, regd)
+#define punpckhqdq_r2r(regs,regd) mmx_r2r (punpckhqdq, regs, regd)
+
+
#endif /* AVCODEC_I386MMX_H */
diff --git a/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c b/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
index 59020466f..76007a1d1 100644
--- a/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
@@ -46,213 +46,216 @@ static uint16_t idct_cosine_table[7] = {
#define r7 mm7
/* from original comments: The Macro does IDct on 4 1-D Dcts */
-#define BeginIDCT() \
+#define BeginIDCT() { \
movq_m2r(*I(3), r2); \
movq_m2r(*C(3), r6); \
movq_r2r(r2, r4); \
movq_m2r(*J(5), r7); \
- pmulhw_r2r(r6, r4); \
+ pmulhw_r2r(r6, r4); /* r4 = c3*i3 - i3 */ \
movq_m2r(*C(5), r1); \
- pmulhw_r2r(r7, r6); \
+ pmulhw_r2r(r7, r6); /* r6 = c3*i5 - i5 */ \
movq_r2r(r1, r5); \
- pmulhw_r2r(r2, r1); \
+ pmulhw_r2r(r2, r1); /* r1 = c5*i3 - i3 */ \
movq_m2r(*I(1), r3); \
- pmulhw_r2r(r7, r5); \
- movq_m2r(*C(1), r0); \
- paddw_r2r(r2, r4); \
- paddw_r2r(r7, r6); \
- paddw_r2r(r1, r2); \
+ pmulhw_r2r(r7, r5); /* r5 = c5*i5 - i5 */ \
+ movq_m2r(*C(1), r0); /* (all registers are in use) */ \
+ paddw_r2r(r2, r4); /* r4 = c3*i3 */ \
+ paddw_r2r(r7, r6); /* r6 = c3*i5 */ \
+ paddw_r2r(r1, r2); /* r2 = c5*i3 */ \
movq_m2r(*J(7), r1); \
- paddw_r2r(r5, r7); \
- movq_r2r(r0, r5); \
- pmulhw_r2r(r3, r0); \
- paddsw_r2r(r7, r4); \
- pmulhw_r2r(r1, r5); \
+ paddw_r2r(r5, r7); /* r7 = c5*i5 */ \
+ movq_r2r(r0, r5); /* r5 = c1 */ \
+ pmulhw_r2r(r3, r0); /* r0 = c1*i1 - i1 */ \
+ paddsw_r2r(r7, r4); /* r4 = C = c3*i3 + c5*i5 */ \
+ pmulhw_r2r(r1, r5); /* r5 = c1*i7 - i7 */ \
movq_m2r(*C(7), r7); \
- psubsw_r2r(r2, r6); \
- paddw_r2r(r3, r0); \
- pmulhw_r2r(r7, r3); \
+ psubsw_r2r(r2, r6); /* r6 = D = c3*i5 - c5*i3 */ \
+ paddw_r2r(r3, r0); /* r0 = c1*i1 */ \
+ pmulhw_r2r(r7, r3); /* r3 = c7*i1 */ \
movq_m2r(*I(2), r2); \
- pmulhw_r2r(r1, r7); \
- paddw_r2r(r1, r5); \
- movq_r2r(r2, r1); \
- pmulhw_m2r(*C(2), r2); \
- psubsw_r2r(r5, r3); \
+ pmulhw_r2r(r1, r7); /* r7 = c7*i7 */ \
+ paddw_r2r(r1, r5); /* r5 = c1*i7 */ \
+ movq_r2r(r2, r1); /* r1 = i2 */ \
+ pmulhw_m2r(*C(2), r2); /* r2 = c2*i2 - i2 */ \
+ psubsw_r2r(r5, r3); /* r3 = B = c7*i1 - c1*i7 */ \
movq_m2r(*J(6), r5); \
- paddsw_r2r(r7, r0); \
- movq_r2r(r5, r7); \
- psubsw_r2r(r4, r0); \
- pmulhw_m2r(*C(2), r5); \
- paddw_r2r(r1, r2); \
- pmulhw_m2r(*C(6), r1); \
- paddsw_r2r(r4, r4); \
- paddsw_r2r(r0, r4); \
- psubsw_r2r(r6, r3); \
- paddw_r2r(r7, r5); \
- paddsw_r2r(r6, r6); \
- pmulhw_m2r(*C(6), r7); \
- paddsw_r2r(r3, r6); \
- movq_r2m(r4, *I(1)); \
- psubsw_r2r(r5, r1); \
+ paddsw_r2r(r7, r0); /* r0 = A = c1*i1 + c7*i7 */ \
+ movq_r2r(r5, r7); /* r7 = i6 */ \
+ psubsw_r2r(r4, r0); /* r0 = A - C */ \
+ pmulhw_m2r(*C(2), r5); /* r5 = c2*i6 - i6 */ \
+ paddw_r2r(r1, r2); /* r2 = c2*i2 */ \
+ pmulhw_m2r(*C(6), r1); /* r1 = c6*i2 */ \
+ paddsw_r2r(r4, r4); /* r4 = C + C */ \
+ paddsw_r2r(r0, r4); /* r4 = C. = A + C */ \
+ psubsw_r2r(r6, r3); /* r3 = B - D */ \
+ paddw_r2r(r7, r5); /* r5 = c2*i6 */ \
+ paddsw_r2r(r6, r6); /* r6 = D + D */ \
+ pmulhw_m2r(*C(6), r7); /* r7 = c6*i6 */ \
+ paddsw_r2r(r3, r6); /* r6 = D. = B + D */ \
+ movq_r2m(r4, *I(1)); /* save C. at I(1) */ \
+ psubsw_r2r(r5, r1); /* r1 = H = c6*i2 - c2*i6 */ \
movq_m2r(*C(4), r4); \
- movq_r2r(r3, r5); \
- pmulhw_r2r(r4, r3); \
- paddsw_r2r(r2, r7); \
- movq_r2m(r6, *I(2)); \
- movq_r2r(r0, r2); \
+ movq_r2r(r3, r5); /* r5 = B - D */ \
+ pmulhw_r2r(r4, r3); /* r3 = (c4 - 1) * (B - D) */ \
+ paddsw_r2r(r2, r7); /* r7 = G = c6*i6 + c2*i2 */ \
+ movq_r2m(r6, *I(2)); /* save D. at I(2) */ \
+ movq_r2r(r0, r2); /* r2 = A - C */ \
movq_m2r(*I(0), r6); \
- pmulhw_r2r(r4, r0); \
- paddw_r2r(r3, r5); \
+ pmulhw_r2r(r4, r0); /* r0 = (c4 - 1) * (A - C) */ \
+ paddw_r2r(r3, r5); /* r5 = B. = c4 * (B - D) */ \
movq_m2r(*J(4), r3); \
- psubsw_r2r(r1, r5); \
- paddw_r2r(r0, r2); \
- psubsw_r2r(r3, r6); \
+ psubsw_r2r(r1, r5); /* r5 = B.. = B. - H */ \
+ paddw_r2r(r0, r2); /* r0 = A. = c4 * (A - C) */ \
+ psubsw_r2r(r3, r6); /* r6 = i0 - i4 */ \
movq_r2r(r6, r0); \
- pmulhw_r2r(r4, r6); \
- paddsw_r2r(r3, r3); \
- paddsw_r2r(r1, r1); \
- paddsw_r2r(r0, r3); \
- paddsw_r2r(r5, r1); \
- pmulhw_r2r(r3, r4); \
- paddsw_r2r(r0, r6); \
- psubsw_r2r(r2, r6); \
- paddsw_r2r(r2, r2); \
- movq_m2r(*I(1), r0); \
- paddsw_r2r(r6, r2); \
- paddw_r2r(r3, r4); \
- psubsw_r2r(r1, r2);
+ pmulhw_r2r(r4, r6); /* r6 = (c4 - 1) * (i0 - i4) */ \
+ paddsw_r2r(r3, r3); /* r3 = i4 + i4 */ \
+ paddsw_r2r(r1, r1); /* r1 = H + H */ \
+ paddsw_r2r(r0, r3); /* r3 = i0 + i4 */ \
+ paddsw_r2r(r5, r1); /* r1 = H. = B + H */ \
+ pmulhw_r2r(r3, r4); /* r4 = (c4 - 1) * (i0 + i4) */ \
+ paddsw_r2r(r0, r6); /* r6 = F = c4 * (i0 - i4) */ \
+ psubsw_r2r(r2, r6); /* r6 = F. = F - A. */ \
+ paddsw_r2r(r2, r2); /* r2 = A. + A. */ \
+ movq_m2r(*I(1), r0); /* r0 = C. */ \
+ paddsw_r2r(r6, r2); /* r2 = A.. = F + A. */ \
+ paddw_r2r(r3, r4); /* r4 = E = c4 * (i0 + i4) */ \
+ psubsw_r2r(r1, r2); /* r2 = R2 = A.. - H. */ \
+}
/* RowIDCT gets ready to transpose */
-#define RowIDCT() \
+#define RowIDCT() { \
\
- BeginIDCT() \
+ BeginIDCT(); \
\
- movq_m2r(*I(2), r3); \
- psubsw_r2r(r7, r4); \
- paddsw_r2r(r1, r1); \
- paddsw_r2r(r7, r7); \
- paddsw_r2r(r2, r1); \
- paddsw_r2r(r4, r7); \
- psubsw_r2r(r3, r4); \
- psubsw_r2r(r5, r6); \
+ movq_m2r(*I(2), r3); /* r3 = D. */ \
+ psubsw_r2r(r7, r4); /* r4 = E. = E - G */ \
+ paddsw_r2r(r1, r1); /* r1 = H. + H. */ \
+ paddsw_r2r(r7, r7); /* r7 = G + G */ \
+ paddsw_r2r(r2, r1); /* r1 = R1 = A.. + H. */ \
+ paddsw_r2r(r4, r7); /* r7 = G. = E + G */ \
+ psubsw_r2r(r3, r4); /* r4 = R4 = E. - D. */ \
+ paddsw_r2r(r3, r3); \
+ psubsw_r2r(r5, r6); /* r6 = R6 = F. - B.. */ \
paddsw_r2r(r5, r5); \
- paddsw_r2r(r4, r3); \
- paddsw_r2r(r6, r5); \
- psubsw_r2r(r0, r7); \
+ paddsw_r2r(r4, r3); /* r3 = R3 = E. + D. */ \
+ paddsw_r2r(r6, r5); /* r5 = R5 = F. + B.. */ \
+ psubsw_r2r(r0, r7); /* r7 = R7 = G. - C. */ \
paddsw_r2r(r0, r0); \
- movq_r2m(r1, *I(1)); \
- paddsw_r2r(r7, r0);
+ movq_r2m(r1, *I(1)); /* save R1 */ \
+ paddsw_r2r(r7, r0); /* r0 = R0 = G. + C. */ \
+}
/* Column IDCT normalizes and stores final results */
-#define ColumnIDCT() \
+#define ColumnIDCT() { \
\
- BeginIDCT() \
+ BeginIDCT(); \
\
- paddsw_m2r(*Eight, r2); \
- paddsw_r2r(r1, r1); \
- paddsw_r2r(r2, r1); \
- psraw_i2r(4, r2); \
- psubsw_r2r(r7, r4); \
- psraw_i2r(4, r1); \
- movq_m2r(*I(2), r3); \
- paddsw_r2r(r7, r7); \
- movq_r2m(r2, *I(2)); \
- paddsw_r2r(r4, r7); \
- movq_r2m(r1, *I(1)); \
- psubsw_r2r(r3, r4); \
- paddsw_m2r(*Eight, r4); \
- paddsw_r2r(r3, r3); \
- paddsw_r2r(r4, r3); \
- psraw_i2r(4, r4); \
- psubsw_r2r(r5, r6); \
- psraw_i2r(4, r3); \
- paddsw_m2r(*Eight, r6); \
- paddsw_r2r(r5, r5); \
- paddsw_r2r(r6, r5); \
- psraw_i2r(4, r6); \
- movq_r2m(r4, *J(4)); \
- psraw_i2r(4, r5); \
- movq_r2m(r3, *I(3)); \
- psubsw_r2r(r0, r7); \
- paddsw_m2r(*Eight, r7); \
- paddsw_r2r(r0, r0); \
- paddsw_r2r(r7, r0); \
- psraw_i2r(4, r7); \
- movq_r2m(r6, *J(6)); \
- psraw_i2r(4, r0); \
- movq_r2m(r5, *J(5)); \
- movq_r2m(r7, *J(7)); \
- movq_r2m(r0, *I(0));
-
+ paddsw_m2r(*Eight, r2); /* adjust R2 (and R1) for shift */ \
+ paddsw_r2r(r1, r1); /* r1 = H. + H. */ \
+ paddsw_r2r(r2, r1); /* r1 = R1 = A.. + H. */ \
+ psraw_i2r(4, r2); /* r2 = NR2 */ \
+ psubsw_r2r(r7, r4); /* r4 = E. = E - G */ \
+ psraw_i2r(4, r1); /* r1 = NR1 */ \
+ movq_m2r(*I(2), r3); /* r3 = D. */ \
+ paddsw_r2r(r7, r7); /* r7 = G + G */ \
+ movq_r2m(r2, *I(2)); /* store NR2 at I2 */ \
+ paddsw_r2r(r4, r7); /* r7 = G. = E + G */ \
+ movq_r2m(r1, *I(1)); /* store NR1 at I1 */ \
+ psubsw_r2r(r3, r4); /* r4 = R4 = E. - D. */ \
+ paddsw_m2r(*Eight, r4); /* adjust R4 (and R3) for shift */ \
+ paddsw_r2r(r3, r3); /* r3 = D. + D. */ \
+ paddsw_r2r(r4, r3); /* r3 = R3 = E. + D. */ \
+ psraw_i2r(4, r4); /* r4 = NR4 */ \
+ psubsw_r2r(r5, r6); /* r6 = R6 = F. - B.. */ \
+ psraw_i2r(4, r3); /* r3 = NR3 */ \
+ paddsw_m2r(*Eight, r6); /* adjust R6 (and R5) for shift */ \
+ paddsw_r2r(r5, r5); /* r5 = B.. + B.. */ \
+ paddsw_r2r(r6, r5); /* r5 = R5 = F. + B.. */ \
+ psraw_i2r(4, r6); /* r6 = NR6 */ \
+ movq_r2m(r4, *J(4)); /* store NR4 at J4 */ \
+ psraw_i2r(4, r5); /* r5 = NR5 */ \
+ movq_r2m(r3, *I(3)); /* store NR3 at I3 */ \
+ psubsw_r2r(r0, r7); /* r7 = R7 = G. - C. */ \
+ paddsw_m2r(*Eight, r7); /* adjust R7 (and R0) for shift */ \
+ paddsw_r2r(r0, r0); /* r0 = C. + C. */ \
+ paddsw_r2r(r7, r0); /* r0 = R0 = G. + C. */ \
+ psraw_i2r(4, r7); /* r7 = NR7 */ \
+ movq_r2m(r6, *J(6)); /* store NR6 at J6 */ \
+ psraw_i2r(4, r0); /* r0 = NR0 */ \
+ movq_r2m(r5, *J(5)); /* store NR5 at J5 */ \
+ movq_r2m(r7, *J(7)); /* store NR7 at J7 */ \
+ movq_r2m(r0, *I(0)); /* store NR0 at I0 */ \
+}
/* Following macro does two 4x4 transposes in place.
At entry (we assume):
- r0 = a3 a2 a1 a0
- I(1) = b3 b2 b1 b0
- r2 = c3 c2 c1 c0
- r3 = d3 d2 d1 d0
-
- r4 = e3 e2 e1 e0
- r5 = f3 f2 f1 f0
- r6 = g3 g2 g1 g0
- r7 = h3 h2 h1 h0
+ r0 = a3 a2 a1 a0
+ I(1) = b3 b2 b1 b0
+ r2 = c3 c2 c1 c0
+ r3 = d3 d2 d1 d0
- At exit, we have:
+ r4 = e3 e2 e1 e0
+ r5 = f3 f2 f1 f0
+ r6 = g3 g2 g1 g0
+ r7 = h3 h2 h1 h0
- I(0) = d0 c0 b0 a0
- I(1) = d1 c1 b1 a1
- I(2) = d2 c2 b2 a2
- I(3) = d3 c3 b3 a3
+ At exit, we have:
- J(4) = h0 g0 f0 e0
- J(5) = h1 g1 f1 e1
- J(6) = h2 g2 f2 e2
- J(7) = h3 g3 f3 e3
+ I(0) = d0 c0 b0 a0
+ I(1) = d1 c1 b1 a1
+ I(2) = d2 c2 b2 a2
+ I(3) = d3 c3 b3 a3
+
+ J(4) = h0 g0 f0 e0
+ J(5) = h1 g1 f1 e1
+ J(6) = h2 g2 f2 e2
+ J(7) = h3 g3 f3 e3
I(0) I(1) I(2) I(3) is the transpose of r0 I(1) r2 r3.
J(4) J(5) J(6) J(7) is the transpose of r4 r5 r6 r7.
Since r1 is free at entry, we calculate the Js first. */
-#define Transpose() \
- movq_r2r(r4, r1); \
- punpcklwd_r2r(r5, r4); \
- movq_r2m(r0, *I(0)); \
- punpckhwd_r2r(r5, r1); \
- movq_r2r(r6, r0); \
- punpcklwd_r2r(r7, r6); \
- movq_r2r(r4, r5); \
- punpckldq_r2r(r6, r4); \
- punpckhdq_r2r(r6, r5); \
- movq_r2r(r1, r6); \
+#define Transpose() { \
+ movq_r2r(r4, r1); /* r1 = e3 e2 e1 e0 */ \
+ punpcklwd_r2r(r5, r4); /* r4 = f1 e1 f0 e0 */ \
+ movq_r2m(r0, *I(0)); /* save a3 a2 a1 a0 */ \
+ punpckhwd_r2r(r5, r1); /* r1 = f3 e3 f2 e2 */ \
+ movq_r2r(r6, r0); /* r0 = g3 g2 g1 g0 */ \
+ punpcklwd_r2r(r7, r6); /* r6 = h1 g1 h0 g0 */ \
+ movq_r2r(r4, r5); /* r5 = f1 e1 f0 e0 */ \
+ punpckldq_r2r(r6, r4); /* r4 = h0 g0 f0 e0 = R4 */ \
+ punpckhdq_r2r(r6, r5); /* r5 = h1 g1 f1 e1 = R5 */ \
+ movq_r2r(r1, r6); /* r6 = f3 e3 f2 e2 */ \
movq_r2m(r4, *J(4)); \
- punpckhwd_r2r(r7, r0); \
+ punpckhwd_r2r(r7, r0); /* r0 = h3 g3 h2 g2 */ \
movq_r2m(r5, *J(5)); \
- punpckhdq_r2r(r0, r6); \
- movq_m2r(*I(0), r4); \
- punpckldq_r2r(r0, r1); \
- movq_m2r(*I(1), r5); \
- movq_r2r(r4, r0); \
+ punpckhdq_r2r(r0, r6); /* r6 = h3 g3 f3 e3 = R7 */ \
+ movq_m2r(*I(0), r4); /* r4 = a3 a2 a1 a0 */ \
+ punpckldq_r2r(r0, r1); /* r1 = h2 g2 f2 e2 = R6 */ \
+ movq_m2r(*I(1), r5); /* r5 = b3 b2 b1 b0 */ \
+ movq_r2r(r4, r0); /* r0 = a3 a2 a1 a0 */ \
movq_r2m(r6, *J(7)); \
- punpcklwd_r2r(r5, r0); \
+ punpcklwd_r2r(r5, r0); /* r0 = b1 a1 b0 a0 */ \
movq_r2m(r1, *J(6)); \
- punpckhwd_r2r(r5, r4); \
- movq_r2r(r2, r5); \
- punpcklwd_r2r(r3, r2); \
- movq_r2r(r0, r1); \
- punpckldq_r2r(r2, r0); \
- punpckhdq_r2r(r2, r1); \
- movq_r2r(r4, r2); \
+ punpckhwd_r2r(r5, r4); /* r4 = b3 a3 b2 a2 */ \
+ movq_r2r(r2, r5); /* r5 = c3 c2 c1 c0 */ \
+ punpcklwd_r2r(r3, r2); /* r2 = d1 c1 d0 c0 */ \
+ movq_r2r(r0, r1); /* r1 = b1 a1 b0 a0 */ \
+ punpckldq_r2r(r2, r0); /* r0 = d0 c0 b0 a0 = R0 */ \
+ punpckhdq_r2r(r2, r1); /* r1 = d1 c1 b1 a1 = R1 */ \
+ movq_r2r(r4, r2); /* r2 = b3 a3 b2 a2 */ \
movq_r2m(r0, *I(0)); \
- punpckhwd_r2r(r3, r5); \
+ punpckhwd_r2r(r3, r5); /* r5 = d3 c3 d2 c2 */ \
movq_r2m(r1, *I(1)); \
- punpckhdq_r2r(r5, r4); \
- punpckldq_r2r(r5, r2); \
+ punpckhdq_r2r(r5, r4); /* r4 = d3 c3 b3 a3 = R3 */ \
+ punpckldq_r2r(r5, r2); /* r2 = d2 c2 b2 a2 = R2 */ \
movq_r2m(r4, *I(3)); \
- movq_r2m(r2, *I(2));
-
+ movq_r2m(r2, *I(2)); \
+}
void vp3_dsp_init_mmx(void)
{
@@ -263,7 +266,7 @@ void vp3_dsp_init_mmx(void)
idct_constants[--j] = 0;
} while (j);
- idct_constants[0] = idct_constants[5] =
+ idct_constants[0] = idct_constants[5] =
idct_constants[10] = idct_constants[15] = 65535;
j = 1;
@@ -272,7 +275,7 @@ void vp3_dsp_init_mmx(void)
p[0] = p[1] = p[2] = p[3] = idct_cosine_table[j - 1];
} while (++j <= 7);
- idct_constants[44] = idct_constants[45] =
+ idct_constants[44] = idct_constants[45] =
idct_constants[46] = idct_constants[47] = IdctAdjustBeforeShift;
}
@@ -292,254 +295,240 @@ static void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
#define C(x) (idct_constants + 16 + (x - 1) * 4)
#define Eight (idct_constants + 44)
- movq_m2r(*input_data, r0);
- pmullw_m2r(*dequant_matrix, r0);
- movq_m2r(*(input_data + 8), r1);
- pmullw_m2r(*(dequant_matrix + 8), r1);
- movq_m2r(*M(0), r2);
- movq_r2r(r0, r3);
- movq_m2r(*(input_data + 4), r4);
- psrlq_i2r(16, r0);
- pmullw_m2r(*(dequant_matrix + 4), r4);
- pand_r2r(r2, r3);
- movq_r2r(r0, r5);
- movq_r2r(r1, r6);
- pand_r2r(r2, r5);
- psllq_i2r(32, r6);
- movq_m2r(*M(3), r7);
- pxor_r2r(r5, r0);
- pand_r2r(r6, r7);
- por_r2r(r3, r0);
- pxor_r2r(r7, r6);
- por_r2r(r7, r0);
- movq_m2r(*M(3), r7);
- movq_r2r(r4, r3);
- movq_r2m(r0, *output_data);
-
- pand_r2r(r2, r3);
- movq_m2r(*(input_data + 16), r0);
- psllq_i2r(16, r3);
- pmullw_m2r(*(dequant_matrix + 16), r0);
- pand_r2r(r1, r7);
- por_r2r(r3, r5);
- por_r2r(r6, r7);
- movq_m2r(*(input_data + 12), r3);
- por_r2r(r5, r7);
- pmullw_m2r(*(dequant_matrix + 12), r3);
- psrlq_i2r(16, r4);
- movq_r2m(r7, *(output_data + 8));
-
- movq_r2r(r4, r5);
- movq_r2r(r0, r7);
- psrlq_i2r(16, r4);
- psrlq_i2r(48, r7);
- movq_r2r(r2, r6);
- pand_r2r(r2, r5);
- pand_r2r(r4, r6);
- movq_r2m(r7, *(output_data + 40));
-
- pxor_r2r(r6, r4);
- psrlq_i2r(32, r1);
- por_r2r(r5, r4);
- movq_m2r(*M(3), r7);
- pand_r2r(r2, r1);
- movq_m2r(*(input_data + 24), r5);
- psllq_i2r(16, r0);
- pmullw_m2r(*(dequant_matrix + 24), r5);
- pand_r2r(r0, r7);
- movq_r2m(r1, *(output_data + 32));
-
- por_r2r(r4, r7);
- movq_r2r(r3, r4);
- pand_r2r(r2, r3);
- movq_m2r(*M(2), r1);
- psllq_i2r(32, r3);
- por_r2r(r3, r7);
- movq_r2r(r5, r3);
- psllq_i2r(48, r3);
- pand_r2r(r0, r1);
- movq_r2m(r7, *(output_data + 16));
-
- por_r2r(r3, r6);
- movq_m2r(*M(1), r7);
- por_r2r(r1, r6);
- movq_m2r(*(input_data + 28), r1);
- pand_r2r(r4, r7);
- pmullw_m2r(*(dequant_matrix + 28), r1);
- por_r2r(r6, r7);
- pand_m2r(*M(1), r0);
- psrlq_i2r(32, r4);
- movq_r2m(r7, *(output_data + 24));
-
- movq_r2r(r4, r6);
- movq_m2r(*M(3), r7);
- pand_r2r(r2, r4);
- movq_m2r(*M(1), r3);
- pand_r2r(r1, r7);
- pand_r2r(r5, r3);
- por_r2r(r4, r0);
- psllq_i2r(16, r3);
- por_r2r(r0, r7);
- movq_m2r(*M(2), r4);
- por_r2r(r3, r7);
- movq_m2r(*(input_data + 40), r0);
- movq_r2r(r4, r3);
- pmullw_m2r(*(dequant_matrix + 40), r0);
- pand_r2r(r5, r4);
- movq_r2m(r7, *(output_data + 4));
-
- por_r2r(r4, r6);
- movq_r2r(r3, r4);
- psrlq_i2r(16, r6);
- movq_r2r(r0, r7);
- pand_r2r(r1, r4);
- psllq_i2r(48, r7);
- por_r2r(r4, r6);
- movq_m2r(*(input_data + 44), r4);
- por_r2r(r6, r7);
- pmullw_m2r(*(dequant_matrix + 44), r4);
- psrlq_i2r(16, r3);
- movq_r2m(r7, *(output_data + 12));
-
- pand_r2r(r1, r3);
- psrlq_i2r(48, r5);
- pand_r2r(r2, r1);
- movq_m2r(*(input_data + 52), r6);
- por_r2r(r3, r5);
- pmullw_m2r(*(input_data + 52), r6);
- psrlq_i2r(16, r0);
- movq_r2r(r4, r7);
- movq_r2r(r2, r3);
- psllq_i2r(48, r7);
- pand_r2r(r0, r3);
- pxor_r2r(r3, r0);
- psllq_i2r(32, r3);
- por_r2r(r5, r7);
- movq_r2r(r6, r5);
- pand_m2r(*M(1), r6);
- por_r2r(r3, r7);
- psllq_i2r(32, r6);
- por_r2r(r1, r0);
- movq_r2m(r7, *(output_data + 20));
-
- por_r2r(r6, r0);
- movq_m2r(*(input_data + 60), r7);
- movq_r2r(r5, r6);
- pmullw_m2r(*(input_data + 60), r7);
- psrlq_i2r(32, r5);
- pand_r2r(r2, r6);
- movq_r2r(r5, r1);
- movq_r2m(r0, *(output_data + 28));
-
- pand_r2r(r2, r1);
- movq_m2r(*(input_data + 56), r0);
- movq_r2r(r7, r3);
- pmullw_m2r(*(dequant_matrix + 56), r0);
- psllq_i2r(16, r3);
- pand_m2r(*M(3), r7);
- pxor_r2r(r1, r5);
- por_r2r(r5, r6);
- movq_r2r(r3, r5);
- pand_m2r(*M(3), r5);
- por_r2r(r1, r7);
- movq_m2r(*(input_data + 48), r1);
- pxor_r2r(r5, r3);
- pmullw_m2r(*(dequant_matrix + 48), r1);
- por_r2r(r3, r7);
- por_r2r(r5, r6);
- movq_r2r(r0, r5);
- movq_r2m(r7, *(output_data + 60));
-
- psrlq_i2r(16, r5);
- pand_m2r(*M(2), r5);
- movq_r2r(r0, r7);
- por_r2r(r5, r6);
- pand_r2r(r2, r0);
- pxor_r2r(r0, r7);
- psllq_i2r(32, r0);
- movq_r2m(r6, *(output_data + 52));
-
- psrlq_i2r(16, r4);
- movq_m2r(*(input_data + 36), r5);
- psllq_i2r(16, r7);
- pmullw_m2r(*(dequant_matrix + 36), r5);
- movq_r2r(r7, r6);
- movq_m2r(*M(2), r3);
- psllq_i2r(16, r6);
- pand_m2r(*M(3), r7);
- pand_r2r(r1, r3);
- por_r2r(r0, r7);
- movq_r2r(r1, r0);
- pand_m2r(*M(3), r1);
- por_r2r(r3, r6);
- movq_r2r(r4, r3);
- psrlq_i2r(32, r1);
- pand_r2r(r2, r3);
- por_r2r(r1, r7);
- por_r2r(r3, r7);
- movq_r2r(r4, r3);
- pand_m2r(*M(1), r3);
- movq_r2r(r5, r1);
- movq_r2m(r7, *(output_data + 44));
-
- psrlq_i2r(48, r5);
- movq_m2r(*(input_data + 32), r7);
- por_r2r(r3, r6);
- pmullw_m2r(*(dequant_matrix + 32), r7);
- por_r2r(r5, r6);
- pand_m2r(*M(2), r4);
- psllq_i2r(32, r0);
- movq_r2m(r6, *(output_data + 36));
-
- movq_r2r(r0, r6);
- pand_m2r(*M(3), r0);
- psllq_i2r(16, r6);
- movq_m2r(*(input_data + 20), r5);
- movq_r2r(r1, r3);
- pmullw_m2r(*(dequant_matrix + 40), r5);
- psrlq_i2r(16, r1);
- pand_m2r(*M(1), r1);
- por_r2r(r4, r0);
- pand_r2r(r7, r2);
- por_r2r(r1, r0);
- por_r2r(r2, r0);
- psllq_i2r(16, r3);
- movq_r2r(r3, r4);
- movq_r2r(r5, r2);
- movq_r2m(r0, *(output_data + 56));
-
- psrlq_i2r(48, r2);
- pand_m2r(*M(2), r4);
- por_r2r(r2, r6);
- movq_m2r(*M(1), r2);
- por_r2r(r4, r6);
- pand_r2r(r7, r2);
- psllq_i2r(32, r3);
- por_m2r(*(output_data + 40), r3);
-
- por_r2r(r2, r6);
- movq_m2r(*M(3), r2);
- psllq_i2r(16, r5);
- movq_r2m(r6, *(output_data + 48));
-
- pand_r2r(r5, r2);
- movq_m2r(*M(2), r6);
- pxor_r2r(r2, r5);
- pand_r2r(r7, r6);
- psrlq_i2r(32, r2);
- pand_m2r(*M(3), r7);
- por_r2r(r2, r3);
- por_m2r(*(output_data + 32), r7);
-
- por_r2r(r3, r6);
- por_r2r(r5, r7);
- movq_r2m(r6, *(output_data + 40));
- movq_r2m(r7, *(output_data + 32));
+ unsigned char *input_bytes = (unsigned char *)input_data;
+ unsigned char *dequant_matrix_bytes = (unsigned char *)dequant_matrix;
+ unsigned char *output_data_bytes = (unsigned char *)output_data;
+
+ movq_m2r(*(input_bytes), r0);
+ pmullw_m2r(*(dequant_matrix_bytes), r0); /* r0 = 03 02 01 00 */
+ movq_m2r(*(input_bytes+16), r1);
+ pmullw_m2r(*(dequant_matrix_bytes+16), r1); /* r1 = 13 12 11 10 */
+ movq_m2r(*M(0), r2); /* r2 = __ __ __ FF */
+ movq_r2r(r0, r3); /* r3 = 03 02 01 00 */
+ movq_m2r(*(input_bytes+8), r4);
+ psrlq_i2r(16, r0); /* r0 = __ 03 02 01 */
+ pmullw_m2r(*(dequant_matrix_bytes+8), r4); /* r4 = 07 06 05 04 */
+ pand_r2r(r2, r3); /* r3 = __ __ __ 00 */
+ movq_r2r(r0, r5); /* r5 = __ 03 02 01 */
+ movq_r2r(r1, r6); /* r6 = 13 12 11 10 */
+ pand_r2r(r2, r5); /* r5 = __ __ __ 01 */
+ psllq_i2r(32, r6); /* r6 = 11 10 __ __ */
+ movq_m2r(*M(3), r7); /* r7 = FF __ __ __ */
+ pxor_r2r(r5, r0); /* r0 = __ 03 02 __ */
+ pand_r2r(r6, r7); /* r7 = 11 __ __ __ */
+ por_r2r(r3, r0); /* r0 = __ 03 02 00 */
+ pxor_r2r(r7, r6); /* r6 = __ 10 __ __ */
+ por_r2r(r7, r0); /* r0 = 11 03 02 00 = R0 */
+ movq_m2r(*M(3), r7); /* r7 = FF __ __ __ */
+ movq_r2r(r4, r3); /* r3 = 07 06 05 04 */
+ movq_r2m(r0, *(output_data_bytes)); /* write R0 = r0 */
+ pand_r2r(r2, r3); /* r3 = __ __ __ 04 */
+ movq_m2r(*(input_bytes+32), r0);
+ psllq_i2r(16, r3); /* r3 = __ __ 04 __ */
+ pmullw_m2r(*(dequant_matrix_bytes+32), r0); /* r0 = 23 22 21 20 */
+ pand_r2r(r1, r7); /* r7 = 13 __ __ __ */
+ por_r2r(r3, r5); /* r5 = __ __ 04 01 */
+ por_r2r(r6, r7); /* r7 = 13 10 __ __ */
+ movq_m2r(*(input_bytes+24), r3);
+ por_r2r(r5, r7); /* r7 = 13 10 04 01 = R1 */
+ pmullw_m2r(*(dequant_matrix_bytes+24), r3); /* r3 = 17 16 15 14 */
+ psrlq_i2r(16, r4); /* r4 = __ 07 06 05 */
+ movq_r2m(r7, *(output_data_bytes+16)); /* write R1 = r7 */
+ movq_r2r(r4, r5); /* r5 = __ 07 06 05 */
+ movq_r2r(r0, r7); /* r7 = 23 22 21 20 */
+ psrlq_i2r(16, r4); /* r4 = __ __ 07 06 */
+ psrlq_i2r(48, r7); /* r7 = __ __ __ 23 */
+ movq_r2r(r2, r6); /* r6 = __ __ __ FF */
+ pand_r2r(r2, r5); /* r5 = __ __ __ 05 */
+ pand_r2r(r4, r6); /* r6 = __ __ __ 06 */
+ movq_r2m(r7, *(output_data_bytes+80)); /* partial R9 = __ __ __ 23 */
+ pxor_r2r(r6, r4); /* r4 = __ __ 07 __ */
+ psrlq_i2r(32, r1); /* r1 = __ __ 13 12 */
+ por_r2r(r5, r4); /* r4 = __ __ 07 05 */
+ movq_m2r(*M(3), r7); /* r7 = FF __ __ __ */
+ pand_r2r(r2, r1); /* r1 = __ __ __ 12 */
+ movq_m2r(*(input_bytes+48), r5);
+ psllq_i2r(16, r0); /* r0 = 22 21 20 __ */
+ pmullw_m2r(*(dequant_matrix_bytes+48), r5); /* r5 = 33 32 31 30 */
+ pand_r2r(r0, r7); /* r7 = 22 __ __ __ */
+ movq_r2m(r1, *(output_data_bytes+64)); /* partial R8 = __ __ __ 12 */
+ por_r2r(r4, r7); /* r7 = 22 __ 07 05 */
+ movq_r2r(r3, r4); /* r4 = 17 16 15 14 */
+ pand_r2r(r2, r3); /* r3 = __ __ __ 14 */
+ movq_m2r(*M(2), r1); /* r1 = __ FF __ __ */
+ psllq_i2r(32, r3); /* r3 = __ 14 __ __ */
+ por_r2r(r3, r7); /* r7 = 22 14 07 05 = R2 */
+ movq_r2r(r5, r3); /* r3 = 33 32 31 30 */
+ psllq_i2r(48, r3); /* r3 = 30 __ __ __ */
+ pand_r2r(r0, r1); /* r1 = __ 21 __ __ */
+ movq_r2m(r7, *(output_data_bytes+32)); /* write R2 = r7 */
+ por_r2r(r3, r6); /* r6 = 30 __ __ 06 */
+ movq_m2r(*M(1), r7); /* r7 = __ __ FF __ */
+ por_r2r(r1, r6); /* r6 = 30 21 __ 06 */
+ movq_m2r(*(input_bytes+56), r1);
+ pand_r2r(r4, r7); /* r7 = __ __ 15 __ */
+ pmullw_m2r(*(dequant_matrix_bytes+56), r1); /* r1 = 37 36 35 34 */
+ por_r2r(r6, r7); /* r7 = 30 21 15 06 = R3 */
+ pand_m2r(*M(1), r0); /* r0 = __ __ 20 __ */
+ psrlq_i2r(32, r4); /* r4 = __ __ 17 16 */
+ movq_r2m(r7, *(output_data_bytes+48)); /* write R3 = r7 */
+ movq_r2r(r4, r6); /* r6 = __ __ 17 16 */
+ movq_m2r(*M(3), r7); /* r7 = FF __ __ __ */
+ pand_r2r(r2, r4); /* r4 = __ __ __ 16 */
+ movq_m2r(*M(1), r3); /* r3 = __ __ FF __ */
+ pand_r2r(r1, r7); /* r7 = 37 __ __ __ */
+ pand_r2r(r5, r3); /* r3 = __ __ 31 __ */
+ por_r2r(r4, r0); /* r0 = __ __ 20 16 */
+ psllq_i2r(16, r3); /* r3 = __ 31 __ __ */
+ por_r2r(r0, r7); /* r7 = 37 __ 20 16 */
+ movq_m2r(*M(2), r4); /* r4 = __ FF __ __ */
+ por_r2r(r3, r7); /* r7 = 37 31 20 16 = R4 */
+ movq_m2r(*(input_bytes+80), r0);
+ movq_r2r(r4, r3); /* r3 = __ __ FF __ */
+ pmullw_m2r(*(dequant_matrix_bytes+80), r0); /* r0 = 53 52 51 50 */
+ pand_r2r(r5, r4); /* r4 = __ 32 __ __ */
+ movq_r2m(r7, *(output_data_bytes+8)); /* write R4 = r7 */
+ por_r2r(r4, r6); /* r6 = __ 32 17 16 */
+ movq_r2r(r3, r4); /* r4 = __ FF __ __ */
+ psrlq_i2r(16, r6); /* r6 = __ __ 32 17 */
+ movq_r2r(r0, r7); /* r7 = 53 52 51 50 */
+ pand_r2r(r1, r4); /* r4 = __ 36 __ __ */
+ psllq_i2r(48, r7); /* r7 = 50 __ __ __ */
+ por_r2r(r4, r6); /* r6 = __ 36 32 17 */
+ movq_m2r(*(input_bytes+88), r4);
+ por_r2r(r6, r7); /* r7 = 50 36 32 17 = R5 */
+ pmullw_m2r(*(dequant_matrix_bytes+88), r4); /* r4 = 57 56 55 54 */
+ psrlq_i2r(16, r3); /* r3 = __ __ FF __ */
+ movq_r2m(r7, *(output_data_bytes+24)); /* write R5 = r7 */
+ pand_r2r(r1, r3); /* r3 = __ __ 35 __ */
+ psrlq_i2r(48, r5); /* r5 = __ __ __ 33 */
+ pand_r2r(r2, r1); /* r1 = __ __ __ 34 */
+ movq_m2r(*(input_bytes+104), r6);
+ por_r2r(r3, r5); /* r5 = __ __ 35 33 */
+ pmullw_m2r(*(dequant_matrix_bytes+104), r6); /* r6 = 67 66 65 64 */
+ psrlq_i2r(16, r0); /* r0 = __ 53 52 51 */
+ movq_r2r(r4, r7); /* r7 = 57 56 55 54 */
+ movq_r2r(r2, r3); /* r3 = __ __ __ FF */
+ psllq_i2r(48, r7); /* r7 = 54 __ __ __ */
+ pand_r2r(r0, r3); /* r3 = __ __ __ 51 */
+ pxor_r2r(r3, r0); /* r0 = __ 53 52 __ */
+ psllq_i2r(32, r3); /* r3 = __ 51 __ __ */
+ por_r2r(r5, r7); /* r7 = 54 __ 35 33 */
+ movq_r2r(r6, r5); /* r5 = 67 66 65 64 */
+ pand_m2r(*M(1), r6); /* r6 = __ __ 65 __ */
+ por_r2r(r3, r7); /* r7 = 54 51 35 33 = R6 */
+ psllq_i2r(32, r6); /* r6 = 65 __ __ __ */
+ por_r2r(r1, r0); /* r0 = __ 53 52 34 */
+ movq_r2m(r7, *(output_data_bytes+40)); /* write R6 = r7 */
+ por_r2r(r6, r0); /* r0 = 65 53 52 34 = R7 */
+ movq_m2r(*(input_bytes+120), r7);
+ movq_r2r(r5, r6); /* r6 = 67 66 65 64 */
+ pmullw_m2r(*(dequant_matrix_bytes+120), r7); /* r7 = 77 76 75 74 */
+ psrlq_i2r(32, r5); /* r5 = __ __ 67 66 */
+ pand_r2r(r2, r6); /* r6 = __ __ __ 64 */
+ movq_r2r(r5, r1); /* r1 = __ __ 67 66 */
+ movq_r2m(r0, *(output_data_bytes+56)); /* write R7 = r0 */
+ pand_r2r(r2, r1); /* r1 = __ __ __ 66 */
+ movq_m2r(*(input_bytes+112), r0);
+ movq_r2r(r7, r3); /* r3 = 77 76 75 74 */
+ pmullw_m2r(*(dequant_matrix_bytes+112), r0); /* r0 = 73 72 71 70 */
+ psllq_i2r(16, r3); /* r3 = 76 75 74 __ */
+ pand_m2r(*M(3), r7); /* r7 = 77 __ __ __ */
+ pxor_r2r(r1, r5); /* r5 = __ __ 67 __ */
+ por_r2r(r5, r6); /* r6 = __ __ 67 64 */
+ movq_r2r(r3, r5); /* r5 = 76 75 74 __ */
+ pand_m2r(*M(3), r5); /* r5 = 76 __ __ __ */
+ por_r2r(r1, r7); /* r7 = 77 __ __ 66 */
+ movq_m2r(*(input_bytes+96), r1);
+ pxor_r2r(r5, r3); /* r3 = __ 75 74 __ */
+ pmullw_m2r(*(dequant_matrix_bytes+96), r1); /* r1 = 63 62 61 60 */
+ por_r2r(r3, r7); /* r7 = 77 75 74 66 = R15 */
+ por_r2r(r5, r6); /* r6 = 76 __ 67 64 */
+ movq_r2r(r0, r5); /* r5 = 73 72 71 70 */
+ movq_r2m(r7, *(output_data_bytes+120)); /* store R15 = r7 */
+ psrlq_i2r(16, r5); /* r5 = __ 73 72 71 */
+ pand_m2r(*M(2), r5); /* r5 = __ 73 __ __ */
+ movq_r2r(r0, r7); /* r7 = 73 72 71 70 */
+ por_r2r(r5, r6); /* r6 = 76 73 67 64 = R14 */
+ pand_r2r(r2, r0); /* r0 = __ __ __ 70 */
+ pxor_r2r(r0, r7); /* r7 = 73 72 71 __ */
+ psllq_i2r(32, r0); /* r0 = __ 70 __ __ */
+ movq_r2m(r6, *(output_data_bytes+104)); /* write R14 = r6 */
+ psrlq_i2r(16, r4); /* r4 = __ 57 56 55 */
+ movq_m2r(*(input_bytes+72), r5);
+ psllq_i2r(16, r7); /* r7 = 72 71 __ __ */
+ pmullw_m2r(*(dequant_matrix_bytes+72), r5); /* r5 = 47 46 45 44 */
+ movq_r2r(r7, r6); /* r6 = 72 71 __ __ */
+ movq_m2r(*M(2), r3); /* r3 = __ FF __ __ */
+ psllq_i2r(16, r6); /* r6 = 71 __ __ __ */
+ pand_m2r(*M(3), r7); /* r7 = 72 __ __ __ */
+ pand_r2r(r1, r3); /* r3 = __ 62 __ __ */
+ por_r2r(r0, r7); /* r7 = 72 70 __ __ */
+ movq_r2r(r1, r0); /* r0 = 63 62 61 60 */
+ pand_m2r(*M(3), r1); /* r1 = 63 __ __ __ */
+ por_r2r(r3, r6); /* r6 = 71 62 __ __ */
+ movq_r2r(r4, r3); /* r3 = __ 57 56 55 */
+ psrlq_i2r(32, r1); /* r1 = __ __ 63 __ */
+ pand_r2r(r2, r3); /* r3 = __ __ __ 55 */
+ por_r2r(r1, r7); /* r7 = 72 70 63 __ */
+ por_r2r(r3, r7); /* r7 = 72 70 63 55 = R13 */
+ movq_r2r(r4, r3); /* r3 = __ 57 56 55 */
+ pand_m2r(*M(1), r3); /* r3 = __ __ 56 __ */
+ movq_r2r(r5, r1); /* r1 = 47 46 45 44 */
+ movq_r2m(r7, *(output_data_bytes+88)); /* write R13 = r7 */
+ psrlq_i2r(48, r5); /* r5 = __ __ __ 47 */
+ movq_m2r(*(input_bytes+64), r7);
+ por_r2r(r3, r6); /* r6 = 71 62 56 __ */
+ pmullw_m2r(*(dequant_matrix_bytes+64), r7); /* r7 = 43 42 41 40 */
+ por_r2r(r5, r6); /* r6 = 71 62 56 47 = R12 */
+ pand_m2r(*M(2), r4); /* r4 = __ 57 __ __ */
+ psllq_i2r(32, r0); /* r0 = 61 60 __ __ */
+ movq_r2m(r6, *(output_data_bytes+72)); /* write R12 = r6 */
+ movq_r2r(r0, r6); /* r6 = 61 60 __ __ */
+ pand_m2r(*M(3), r0); /* r0 = 61 __ __ __ */
+ psllq_i2r(16, r6); /* r6 = 60 __ __ __ */
+ movq_m2r(*(input_bytes+40), r5);
+ movq_r2r(r1, r3); /* r3 = 47 46 45 44 */
+ pmullw_m2r(*(dequant_matrix_bytes+40), r5); /* r5 = 27 26 25 24 */
+ psrlq_i2r(16, r1); /* r1 = __ 47 46 45 */
+ pand_m2r(*M(1), r1); /* r1 = __ __ 46 __ */
+ por_r2r(r4, r0); /* r0 = 61 57 __ __ */
+ pand_r2r(r7, r2); /* r2 = __ __ __ 40 */
+ por_r2r(r1, r0); /* r0 = 61 57 46 __ */
+ por_r2r(r2, r0); /* r0 = 61 57 46 40 = R11 */
+ psllq_i2r(16, r3); /* r3 = 46 45 44 __ */
+ movq_r2r(r3, r4); /* r4 = 46 45 44 __ */
+ movq_r2r(r5, r2); /* r2 = 27 26 25 24 */
+ movq_r2m(r0, *(output_data_bytes+112)); /* write R11 = r0 */
+ psrlq_i2r(48, r2); /* r2 = __ __ __ 27 */
+ pand_m2r(*M(2), r4); /* r4 = __ 45 __ __ */
+ por_r2r(r2, r6); /* r6 = 60 __ __ 27 */
+ movq_m2r(*M(1), r2); /* r2 = __ __ FF __ */
+ por_r2r(r4, r6); /* r6 = 60 45 __ 27 */
+ pand_r2r(r7, r2); /* r2 = __ __ 41 __ */
+ psllq_i2r(32, r3); /* r3 = 44 __ __ __ */
+ por_m2r(*(output_data_bytes+80), r3); /* r3 = 44 __ __ 23 */
+ por_r2r(r2, r6); /* r6 = 60 45 41 27 = R10 */
+ movq_m2r(*M(3), r2); /* r2 = FF __ __ __ */
+ psllq_i2r(16, r5); /* r5 = 26 25 24 __ */
+ movq_r2m(r6, *(output_data_bytes+96)); /* store R10 = r6 */
+ pand_r2r(r5, r2); /* r2 = 26 __ __ __ */
+ movq_m2r(*M(2), r6); /* r6 = __ FF __ __ */
+ pxor_r2r(r2, r5); /* r5 = __ 25 24 __ */
+ pand_r2r(r7, r6); /* r6 = __ 42 __ __ */
+ psrlq_i2r(32, r2); /* r2 = __ __ 26 __ */
+ pand_m2r(*M(3), r7); /* r7 = 43 __ __ __ */
+ por_r2r(r2, r3); /* r3 = 44 __ 26 23 */
+ por_m2r(*(output_data_bytes+64), r7); /* r7 = 43 __ __ 12 */
+ por_r2r(r3, r6); /* r6 = 44 42 26 23 = R9 */
+ por_r2r(r5, r7); /* r7 = 43 25 24 12 = R8 */
+ movq_r2m(r6, *(output_data_bytes+80)); /* store R9 = r6 */
+ movq_r2m(r7, *(output_data_bytes+64)); /* store R8 = r7 */
#undef M
- /* at this point, function has completed dequantization + dezigzag +
+ /* at this point, function has completed dequantization + dezigzag +
* partial transposition; now do the idct itself */
#define I(K) (output_data + K * 8)
diff --git a/src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c b/src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c
new file mode 100644
index 000000000..c8f9158af
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c
@@ -0,0 +1,890 @@
+/*
+ * Copyright (C) 2004 the ffmpeg project
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+/**
+ * @file vp3dsp_sse2.c
+ * SSE2-optimized functions cribbed from the original VP3 source code.
+ */
+
+#include "../dsputil.h"
+#include "mmx.h"
+
+static unsigned short __align16 SSE2_dequant_const[] =
+{
+ 0,65535,65535,0,0,0,0,0, // 0x0000 0000 0000 0000 0000 FFFF FFFF 0000
+ 0,0,0,0,65535,65535,0,0, // 0x0000 0000 FFFF FFFF 0000 0000 0000 0000
+ 65535,65535,65535,0,0,0,0,0,// 0x0000 0000 0000 0000 0000 FFFF FFFF FFFF
+ 0,0,0,65535,0,0,0,0, // 0x0000 0000 0000 0000 FFFF 0000 0000 0000
+ 0,0,0,65535,65535,0,0,0, // 0x0000 0000 0000 FFFF FFFF 0000 0000 0000
+ 65535,0,0,0,0,65535,0,0, // 0x0000 0000 FFFF 0000 0000 0000 0000 FFFF
+ 0,0,65535,65535, 0,0,0,0 // 0x0000 0000 0000 0000 FFFF FFFF 0000 0000
+};
+
+static unsigned int __align16 eight_data[] =
+{
+ 0x00080008,
+ 0x00080008,
+ 0x00080008,
+ 0x00080008
+};
+
+static unsigned short __align16 SSE2_idct_data[7 * 8] =
+{
+ 64277,64277,64277,64277,64277,64277,64277,64277,
+ 60547,60547,60547,60547,60547,60547,60547,60547,
+ 54491,54491,54491,54491,54491,54491,54491,54491,
+ 46341,46341,46341,46341,46341,46341,46341,46341,
+ 36410,36410,36410,36410,36410,36410,36410,36410,
+ 25080,25080,25080,25080,25080,25080,25080,25080,
+ 12785,12785,12785,12785,12785,12785,12785,12785
+};
+
+
+#define SSE2_Column_IDCT() { \
+ \
+ movdqu_m2r(*I(3), xmm2); /* xmm2 = i3 */ \
+ movdqu_m2r(*C(3), xmm6); /* xmm6 = c3 */ \
+ \
+ movdqu_r2r(xmm2, xmm4); /* xmm4 = i3 */ \
+ movdqu_m2r(*I(5), xmm7); /* xmm7 = i5 */ \
+ \
+ pmulhw_r2r(xmm6, xmm4); /* xmm4 = c3 * i3 - i3 */ \
+ movdqu_m2r(*C(5), xmm1); /* xmm1 = c5 */ \
+ \
+ pmulhw_r2r(xmm7, xmm6); /* xmm6 = c3 * i5 - i5 */ \
+ movdqu_r2r(xmm1, xmm5); /* xmm5 = c5 */ \
+ \
+ pmulhw_r2r(xmm2, xmm1); /* xmm1 = c5 * i3 - i3 */ \
+ movdqu_m2r(*I(1), xmm3); /* xmm3 = i1 */ \
+ \
+ pmulhw_r2r(xmm7, xmm5); /* xmm5 = c5 * i5 - i5 */ \
+ movdqu_m2r(*C(1), xmm0); /* xmm0 = c1 */ \
+ \
+ /* all registers are in use */ \
+ \
+ paddw_r2r(xmm2, xmm4); /* xmm4 = c3 * i3 */ \
+ paddw_r2r(xmm7, xmm6); /* xmm6 = c3 * i5 */ \
+ \
+ paddw_r2r(xmm1, xmm2); /* xmm2 = c5 * i3 */ \
+ movdqu_m2r(*I(7), xmm1); /* xmm1 = i7 */ \
+ \
+ paddw_r2r(xmm5, xmm7); /* xmm7 = c5 * i5 */ \
+ movdqu_r2r(xmm0, xmm5); /* xmm5 = c1 */ \
+ \
+ pmulhw_r2r(xmm3, xmm0); /* xmm0 = c1 * i1 - i1 */ \
+ paddsw_r2r(xmm7, xmm4); /* xmm4 = c3 * i3 + c5 * i5 = C */ \
+ \
+ pmulhw_r2r(xmm1, xmm5); /* xmm5 = c1 * i7 - i7 */ \
+ movdqu_m2r(*C(7), xmm7); /* xmm7 = c7 */ \
+ \
+ psubsw_r2r(xmm2, xmm6); /* xmm6 = c3 * i5 - c5 * i3 = D */ \
+ paddw_r2r(xmm3, xmm0); /* xmm0 = c1 * i1 */ \
+ \
+ pmulhw_r2r(xmm7, xmm3); /* xmm3 = c7 * i1 */ \
+ movdqu_m2r(*I(2), xmm2); /* xmm2 = i2 */ \
+ \
+ pmulhw_r2r(xmm1, xmm7); /* xmm7 = c7 * i7 */ \
+ paddw_r2r(xmm1, xmm5); /* xmm5 = c1 * i7 */ \
+ \
+ movdqu_r2r(xmm2, xmm1); /* xmm1 = i2 */ \
+ pmulhw_m2r(*C(2), xmm2); /* xmm2 = i2 * c2 -i2 */ \
+ \
+ psubsw_r2r(xmm5, xmm3); /* xmm3 = c7 * i1 - c1 * i7 = B */ \
+ movdqu_m2r(*I(6), xmm5); /* xmm5 = i6 */ \
+ \
+ paddsw_r2r(xmm7, xmm0); /* xmm0 = c1 * i1 + c7 * i7 = A */ \
+ movdqu_r2r(xmm5, xmm7); /* xmm7 = i6 */ \
+ \
+ psubsw_r2r(xmm4, xmm0); /* xmm0 = A - C */ \
+ pmulhw_m2r(*C(2), xmm5); /* xmm5 = c2 * i6 - i6 */ \
+ \
+ paddw_r2r(xmm1, xmm2); /* xmm2 = i2 * c2 */ \
+ pmulhw_m2r(*C(6), xmm1); /* xmm1 = c6 * i2 */ \
+ \
+ paddsw_r2r(xmm4, xmm4); /* xmm4 = C + C */ \
+ paddsw_r2r(xmm0, xmm4); /* xmm4 = A + C = C. */ \
+ \
+ psubsw_r2r(xmm6, xmm3); /* xmm3 = B - D */ \
+ paddw_r2r(xmm7, xmm5); /* xmm5 = c2 * i6 */ \
+ \
+ paddsw_r2r(xmm6, xmm6); /* xmm6 = D + D */ \
+ pmulhw_m2r(*C(6), xmm7); /* xmm7 = c6 * i6 */ \
+ \
+ paddsw_r2r(xmm3, xmm6); /* xmm6 = B + D = D. */ \
+ movdqu_r2m(xmm4, *I(1)); /* Save C. at I(1) */ \
+ \
+ psubsw_r2r(xmm5, xmm1); /* xmm1 = c6 * i2 - c2 * i6 = H */ \
+ movdqu_m2r(*C(4), xmm4); /* xmm4 = c4 */ \
+ \
+ movdqu_r2r(xmm3, xmm5); /* xmm5 = B - D */ \
+ pmulhw_r2r(xmm4, xmm3); /* xmm3 = ( c4 -1 ) * ( B - D ) */ \
+ \
+ paddsw_r2r(xmm2, xmm7); /* xmm7 = c2 * i2 + c6 * i6 = G */ \
+ movdqu_r2m(xmm6, *I(2)); /* Save D. at I(2) */ \
+ \
+ movdqu_r2r(xmm0, xmm2); /* xmm2 = A - C */ \
+ movdqu_m2r(*I(0), xmm6); /* xmm6 = i0 */ \
+ \
+ pmulhw_r2r(xmm4, xmm0); /* xmm0 = ( c4 - 1 ) * ( A - C ) = A. */ \
+ paddw_r2r(xmm3, xmm5); /* xmm5 = c4 * ( B - D ) = B. */ \
+ \
+ movdqu_m2r(*I(4), xmm3); /* xmm3 = i4 */ \
+ psubsw_r2r(xmm1, xmm5); /* xmm5 = B. - H = B.. */ \
+ \
+ paddw_r2r(xmm0, xmm2); /* xmm2 = c4 * ( A - C) = A. */ \
+ psubsw_r2r(xmm3, xmm6); /* xmm6 = i0 - i4 */ \
+ \
+ movdqu_r2r(xmm6, xmm0); /* xmm0 = i0 - i4 */ \
+ pmulhw_r2r(xmm4, xmm6); /* xmm6 = (c4 - 1) * (i0 - i4) = F */ \
+ \
+ paddsw_r2r(xmm3, xmm3); /* xmm3 = i4 + i4 */ \
+ paddsw_r2r(xmm1, xmm1); /* xmm1 = H + H */ \
+ \
+ paddsw_r2r(xmm0, xmm3); /* xmm3 = i0 + i4 */ \
+ paddsw_r2r(xmm5, xmm1); /* xmm1 = B. + H = H. */ \
+ \
+ pmulhw_r2r(xmm3, xmm4); /* xmm4 = ( c4 - 1 ) * ( i0 + i4 ) */ \
+ paddw_r2r(xmm0, xmm6); /* xmm6 = c4 * ( i0 - i4 ) */ \
+ \
+ psubsw_r2r(xmm2, xmm6); /* xmm6 = F - A. = F. */ \
+ paddsw_r2r(xmm2, xmm2); /* xmm2 = A. + A. */ \
+ \
+ movdqu_m2r(*I(1), xmm0); /* Load C. from I(1) */ \
+ paddsw_r2r(xmm6, xmm2); /* xmm2 = F + A. = A.. */ \
+ \
+ paddw_r2r(xmm3, xmm4); /* xmm4 = c4 * ( i0 + i4 ) = 3 */ \
+ psubsw_r2r(xmm1, xmm2); /* xmm2 = A.. - H. = R2 */ \
+ \
+ paddsw_m2r(*Eight, xmm2); /* Adjust R2 and R1 before shifting */ \
+ paddsw_r2r(xmm1, xmm1); /* xmm1 = H. + H. */ \
+ \
+ paddsw_r2r(xmm2, xmm1); /* xmm1 = A.. + H. = R1 */ \
+ psraw_i2r(4, xmm2); /* xmm2 = op2 */ \
+ \
+ psubsw_r2r(xmm7, xmm4); /* xmm4 = E - G = E. */ \
+ psraw_i2r(4, xmm1); /* xmm1 = op1 */ \
+ \
+ movdqu_m2r(*I(2), xmm3); /* Load D. from I(2) */ \
+ paddsw_r2r(xmm7, xmm7); /* xmm7 = G + G */ \
+ \
+ movdqu_r2m(xmm2, *O(2)); /* Write out op2 */ \
+ paddsw_r2r(xmm4, xmm7); /* xmm7 = E + G = G. */ \
+ \
+ movdqu_r2m(xmm1, *O(1)); /* Write out op1 */ \
+ psubsw_r2r(xmm3, xmm4); /* xmm4 = E. - D. = R4 */ \
+ \
+ paddsw_m2r(*Eight, xmm4); /* Adjust R4 and R3 before shifting */ \
+ paddsw_r2r(xmm3, xmm3); /* xmm3 = D. + D. */ \
+ \
+ paddsw_r2r(xmm4, xmm3); /* xmm3 = E. + D. = R3 */ \
+ psraw_i2r(4, xmm4); /* xmm4 = op4 */ \
+ \
+ psubsw_r2r(xmm5, xmm6); /* xmm6 = F. - B..= R6 */ \
+ psraw_i2r(4, xmm3); /* xmm3 = op3 */ \
+ \
+ paddsw_m2r(*Eight, xmm6); /* Adjust R6 and R5 before shifting */ \
+ paddsw_r2r(xmm5, xmm5); /* xmm5 = B.. + B.. */ \
+ \
+ paddsw_r2r(xmm6, xmm5); /* xmm5 = F. + B.. = R5 */ \
+ psraw_i2r(4, xmm6); /* xmm6 = op6 */ \
+ \
+ movdqu_r2m(xmm4, *O(4)); /* Write out op4 */ \
+ psraw_i2r(4, xmm5); /* xmm5 = op5 */ \
+ \
+ movdqu_r2m(xmm3, *O(3)); /* Write out op3 */ \
+ psubsw_r2r(xmm0, xmm7); /* xmm7 = G. - C. = R7 */ \
+ \
+ paddsw_m2r(*Eight, xmm7); /* Adjust R7 and R0 before shifting */ \
+ paddsw_r2r(xmm0, xmm0); /* xmm0 = C. + C. */ \
+ \
+ paddsw_r2r(xmm7, xmm0); /* xmm0 = G. + C. */ \
+ psraw_i2r(4, xmm7); /* xmm7 = op7 */ \
+ \
+ movdqu_r2m(xmm6, *O(6)); /* Write out op6 */ \
+ psraw_i2r(4, xmm0); /* xmm0 = op0 */ \
+ \
+ movdqu_r2m(xmm5, *O(5)); /* Write out op5 */ \
+ movdqu_r2m(xmm7, *O(7)); /* Write out op7 */ \
+ \
+ movdqu_r2m(xmm0, *O(0)); /* Write out op0 */ \
+ \
+} /* End of SSE2_Column_IDCT macro */
+
+
+#define SSE2_Row_IDCT() { \
+ \
+ movdqu_m2r(*I(3), xmm2); /* xmm2 = i3 */ \
+ movdqu_m2r(*C(3), xmm6); /* xmm6 = c3 */ \
+ \
+ movdqu_r2r(xmm2, xmm4); /* xmm4 = i3 */ \
+ movdqu_m2r(*I(5), xmm7); /* xmm7 = i5 */ \
+ \
+ pmulhw_r2r(xmm6, xmm4); /* xmm4 = c3 * i3 - i3 */ \
+ movdqu_m2r(*C(5), xmm1); /* xmm1 = c5 */ \
+ \
+ pmulhw_r2r(xmm7, xmm6); /* xmm6 = c3 * i5 - i5 */ \
+ movdqu_r2r(xmm1, xmm5); /* xmm5 = c5 */ \
+ \
+ pmulhw_r2r(xmm2, xmm1); /* xmm1 = c5 * i3 - i3 */ \
+ movdqu_m2r(*I(1), xmm3); /* xmm3 = i1 */ \
+ \
+ pmulhw_r2r(xmm7, xmm5); /* xmm5 = c5 * i5 - i5 */ \
+ movdqu_m2r(*C(1), xmm0); /* xmm0 = c1 */ \
+ \
+ /* all registers are in use */ \
+ \
+ paddw_r2r(xmm2, xmm4); /* xmm4 = c3 * i3 */ \
+ paddw_r2r(xmm7, xmm6); /* xmm6 = c3 * i5 */ \
+ \
+ paddw_r2r(xmm1, xmm2); /* xmm2 = c5 * i3 */ \
+ movdqu_m2r(*I(7), xmm1); /* xmm1 = i7 */ \
+ \
+ paddw_r2r(xmm5, xmm7); /* xmm7 = c5 * i5 */ \
+ movdqu_r2r(xmm0, xmm5); /* xmm5 = c1 */ \
+ \
+ pmulhw_r2r(xmm3, xmm0); /* xmm0 = c1 * i1 - i1 */ \
+ paddsw_r2r(xmm7, xmm4); /* xmm4 = c3 * i3 + c5 * i5 = C */ \
+ \
+ pmulhw_r2r(xmm1, xmm5); /* xmm5 = c1 * i7 - i7 */ \
+ movdqu_m2r(*C(7), xmm7); /* xmm7 = c7 */ \
+ \
+ psubsw_r2r(xmm2, xmm6); /* xmm6 = c3 * i5 - c5 * i3 = D */ \
+ paddw_r2r(xmm3, xmm0); /* xmm0 = c1 * i1 */ \
+ \
+ pmulhw_r2r(xmm7, xmm3); /* xmm3 = c7 * i1 */ \
+ movdqu_m2r(*I(2), xmm2); /* xmm2 = i2 */ \
+ \
+ pmulhw_r2r(xmm1, xmm7); /* xmm7 = c7 * i7 */ \
+ paddw_r2r(xmm1, xmm5); /* xmm5 = c1 * i7 */ \
+ \
+ movdqu_r2r(xmm2, xmm1); /* xmm1 = i2 */ \
+ pmulhw_m2r(*C(2), xmm2); /* xmm2 = i2 * c2 -i2 */ \
+ \
+ psubsw_r2r(xmm5, xmm3); /* xmm3 = c7 * i1 - c1 * i7 = B */ \
+ movdqu_m2r(*I(6), xmm5); /* xmm5 = i6 */ \
+ \
+ paddsw_r2r(xmm7, xmm0); /* xmm0 = c1 * i1 + c7 * i7 = A */ \
+ movdqu_r2r(xmm5, xmm7); /* xmm7 = i6 */ \
+ \
+ psubsw_r2r(xmm4, xmm0); /* xmm0 = A - C */ \
+ pmulhw_m2r(*C(2), xmm5); /* xmm5 = c2 * i6 - i6 */ \
+ \
+ paddw_r2r(xmm1, xmm2); /* xmm2 = i2 * c2 */ \
+ pmulhw_m2r(*C(6), xmm1); /* xmm1 = c6 * i2 */ \
+ \
+ paddsw_r2r(xmm4, xmm4); /* xmm4 = C + C */ \
+ paddsw_r2r(xmm0, xmm4); /* xmm4 = A + C = C. */ \
+ \
+ psubsw_r2r(xmm6, xmm3); /* xmm3 = B - D */ \
+ paddw_r2r(xmm7, xmm5); /* xmm5 = c2 * i6 */ \
+ \
+ paddsw_r2r(xmm6, xmm6); /* xmm6 = D + D */ \
+ pmulhw_m2r(*C(6), xmm7); /* xmm7 = c6 * i6 */ \
+ \
+ paddsw_r2r(xmm3, xmm6); /* xmm6 = B + D = D. */ \
+ movdqu_r2m(xmm4, *I(1)); /* Save C. at I(1) */ \
+ \
+ psubsw_r2r(xmm5, xmm1); /* xmm1 = c6 * i2 - c2 * i6 = H */ \
+ movdqu_m2r(*C(4), xmm4); /* xmm4 = c4 */ \
+ \
+ movdqu_r2r(xmm3, xmm5); /* xmm5 = B - D */ \
+ pmulhw_r2r(xmm4, xmm3); /* xmm3 = ( c4 -1 ) * ( B - D ) */ \
+ \
+ paddsw_r2r(xmm2, xmm7); /* xmm7 = c2 * i2 + c6 * i6 = G */ \
+ movdqu_r2m(xmm6, *I(2)); /* Save D. at I(2) */ \
+ \
+ movdqu_r2r(xmm0, xmm2); /* xmm2 = A - C */ \
+ movdqu_m2r(*I(0), xmm6); /* xmm6 = i0 */ \
+ \
+ pmulhw_r2r(xmm4, xmm0); /* xmm0 = ( c4 - 1 ) * ( A - C ) = A. */ \
+ paddw_r2r(xmm3, xmm5); /* xmm5 = c4 * ( B - D ) = B. */ \
+ \
+ movdqu_m2r(*I(4), xmm3); /* xmm3 = i4 */ \
+ psubsw_r2r(xmm1, xmm5); /* xmm5 = B. - H = B.. */ \
+ \
+ paddw_r2r(xmm0, xmm2); /* xmm2 = c4 * ( A - C) = A. */ \
+ psubsw_r2r(xmm3, xmm6); /* xmm6 = i0 - i4 */ \
+ \
+ movdqu_r2r(xmm6, xmm0); /* xmm0 = i0 - i4 */ \
+ pmulhw_r2r(xmm4, xmm6); /* xmm6 = ( c4 - 1 ) * ( i0 - i4 ) = F */ \
+ \
+ paddsw_r2r(xmm3, xmm3); /* xmm3 = i4 + i4 */ \
+ paddsw_r2r(xmm1, xmm1); /* xmm1 = H + H */ \
+ \
+ paddsw_r2r(xmm0, xmm3); /* xmm3 = i0 + i4 */ \
+ paddsw_r2r(xmm5, xmm1); /* xmm1 = B. + H = H. */ \
+ \
+ pmulhw_r2r(xmm3, xmm4); /* xmm4 = ( c4 - 1 ) * ( i0 + i4 ) */ \
+ paddw_r2r(xmm0, xmm6); /* xmm6 = c4 * ( i0 - i4 ) */ \
+ \
+ psubsw_r2r(xmm2, xmm6); /* xmm6 = F - A. = F. */ \
+ paddsw_r2r(xmm2, xmm2); /* xmm2 = A. + A. */ \
+ \
+ movdqu_m2r(*I(1), xmm0); /* Load C. from I(1) */ \
+ paddsw_r2r(xmm6, xmm2); /* xmm2 = F + A. = A.. */ \
+ \
+ paddw_r2r(xmm3, xmm4); /* xmm4 = c4 * ( i0 + i4 ) = 3 */ \
+ psubsw_r2r(xmm1, xmm2); /* xmm2 = A.. - H. = R2 */ \
+ \
+ paddsw_r2r(xmm1, xmm1); /* xmm1 = H. + H. */ \
+ paddsw_r2r(xmm2, xmm1); /* xmm1 = A.. + H. = R1 */ \
+ \
+ psubsw_r2r(xmm7, xmm4); /* xmm4 = E - G = E. */ \
+ \
+ movdqu_m2r(*I(2), xmm3); /* Load D. from I(2) */ \
+ paddsw_r2r(xmm7, xmm7); /* xmm7 = G + G */ \
+ \
+ movdqu_r2m(xmm2, *I(2)); /* Write out op2 */ \
+ paddsw_r2r(xmm4, xmm7); /* xmm7 = E + G = G. */ \
+ \
+ movdqu_r2m(xmm1, *I(1)); /* Write out op1 */ \
+ psubsw_r2r(xmm3, xmm4); /* xmm4 = E. - D. = R4 */ \
+ \
+ paddsw_r2r(xmm3, xmm3); /* xmm3 = D. + D. */ \
+ \
+ paddsw_r2r(xmm4, xmm3); /* xmm3 = E. + D. = R3 */ \
+ \
+ psubsw_r2r(xmm5, xmm6); /* xmm6 = F. - B..= R6 */ \
+ \
+ paddsw_r2r(xmm5, xmm5); /* xmm5 = B.. + B.. */ \
+ \
+ paddsw_r2r(xmm6, xmm5); /* xmm5 = F. + B.. = R5 */ \
+ \
+ movdqu_r2m(xmm4, *I(4)); /* Write out op4 */ \
+ \
+ movdqu_r2m(xmm3, *I(3)); /* Write out op3 */ \
+ psubsw_r2r(xmm0, xmm7); /* xmm7 = G. - C. = R7 */ \
+ \
+ paddsw_r2r(xmm0, xmm0); /* xmm0 = C. + C. */ \
+ \
+ paddsw_r2r(xmm7, xmm0); /* xmm0 = G. + C. */ \
+ \
+ movdqu_r2m(xmm6, *I(6)); /* Write out op6 */ \
+ \
+ movdqu_r2m(xmm5, *I(5)); /* Write out op5 */ \
+ movdqu_r2m(xmm7, *I(7)); /* Write out op7 */ \
+ \
+ movdqu_r2m(xmm0, *I(0)); /* Write out op0 */ \
+ \
+} /* End of SSE2_Row_IDCT macro */
+
+
+#define SSE2_Transpose() { \
+ \
+ movdqu_m2r(*I(4), xmm4); /* xmm4=e7e6e5e4e3e2e1e0 */ \
+ movdqu_m2r(*I(5), xmm0); /* xmm4=f7f6f5f4f3f2f1f0 */ \
+ \
+ movdqu_r2r(xmm4, xmm5); /* make a copy */ \
+ punpcklwd_r2r(xmm0, xmm4); /* xmm4=f3e3f2e2f1e1f0e0 */ \
+ \
+ punpckhwd_r2r(xmm0, xmm5); /* xmm5=f7e7f6e6f5e5f4e4 */ \
+ movdqu_m2r(*I(6), xmm6); /* xmm6=g7g6g5g4g3g2g1g0 */ \
+ \
+ movdqu_m2r(*I(7), xmm0); /* xmm0=h7h6h5h4h3h2h1h0 */ \
+ movdqu_r2r(xmm6, xmm7); /* make a copy */ \
+ \
+ punpcklwd_r2r(xmm0, xmm6); /* xmm6=h3g3h3g2h1g1h0g0 */ \
+ punpckhwd_r2r(xmm0, xmm7); /* xmm7=h7g7h6g6h5g5h4g4 */ \
+ \
+ movdqu_r2r(xmm4, xmm3); /* make a copy */ \
+ punpckldq_r2r(xmm6, xmm4); /* xmm4=h1g1f1e1h0g0f0e0 */ \
+ \
+ punpckhdq_r2r(xmm6, xmm3); /* xmm3=h3g3g3e3h2g2f2e2 */ \
+ movdqu_r2m(xmm3, *I(6)); /* save h3g3g3e3h2g2f2e2 */ \
+ /* Free xmm6 */ \
+ movdqu_r2r(xmm5, xmm6); /* make a copy */ \
+ punpckldq_r2r(xmm7, xmm5); /* xmm5=h5g5f5e5h4g4f4e4 */ \
+ \
+ punpckhdq_r2r(xmm7, xmm6); /* xmm6=h7g7f7e7h6g6f6e6 */ \
+ movdqu_m2r(*I(0), xmm0); /* xmm0=a7a6a5a4a3a2a1a0 */ \
+ /* Free xmm7 */ \
+ movdqu_m2r(*I(1), xmm1); /* xmm1=b7b6b5b4b3b2b1b0 */ \
+ movdqu_r2r(xmm0, xmm7); /* make a copy */ \
+ \
+ punpcklwd_r2r(xmm1, xmm0); /* xmm0=b3a3b2a2b1a1b0a0 */ \
+ punpckhwd_r2r(xmm1, xmm7); /* xmm7=b7a7b6a6b5a5b4a4 */ \
+ /* Free xmm1 */ \
+ movdqu_m2r(*I(2), xmm2); /* xmm2=c7c6c5c4c3c2c1c0 */ \
+ movdqu_m2r(*I(3), xmm3); /* xmm3=d7d6d5d4d3d2d1d0 */ \
+ \
+ movdqu_r2r(xmm2, xmm1); /* make a copy */ \
+ punpcklwd_r2r(xmm3, xmm2); /* xmm2=d3c3d2c2d1c1d0c0 */ \
+ \
+ punpckhwd_r2r(xmm3, xmm1); /* xmm1=d7c7d6c6d5c5d4c4 */ \
+ movdqu_r2r(xmm0, xmm3); /* make a copy */ \
+ \
+ punpckldq_r2r(xmm2, xmm0); /* xmm0=d1c1b1a1d0c0b0a0 */ \
+ punpckhdq_r2r(xmm2, xmm3); /* xmm3=d3c3b3a3d2c2b2a2 */ \
+ /* Free xmm2 */ \
+ movdqu_r2r(xmm7, xmm2); /* make a copy */ \
+ punpckldq_r2r(xmm1, xmm2); /* xmm2=d5c5b5a5d4c4b4a4 */ \
+ \
+ punpckhdq_r2r(xmm1, xmm7); /* xmm7=d7c7b7a7d6c6b6a6 */ \
+ movdqu_r2r(xmm0, xmm1); /* make a copy */ \
+ \
+ punpcklqdq_r2r(xmm4, xmm0); /* xmm0=h0g0f0e0d0c0b0a0 */ \
+ punpckhqdq_r2r(xmm4, xmm1); /* xmm1=h1g1g1e1d1c1b1a1 */ \
+ \
+ movdqu_r2m(xmm0, *I(0)); /* save I(0) */ \
+ movdqu_r2m(xmm1, *I(1)); /* save I(1) */ \
+ \
+ movdqu_m2r(*I(6), xmm0); /* load h3g3g3e3h2g2f2e2 */ \
+ movdqu_r2r(xmm3, xmm1); /* make a copy */ \
+ \
+ punpcklqdq_r2r(xmm0, xmm1); /* xmm1=h2g2f2e2d2c2b2a2 */ \
+ punpckhqdq_r2r(xmm0, xmm3); /* xmm3=h3g3f3e3d3c3b3a3 */ \
+ \
+ movdqu_r2r(xmm2, xmm4); /* make a copy */ \
+ punpcklqdq_r2r(xmm5, xmm4); /* xmm4=h4g4f4e4d4c4b4a4 */ \
+ \
+ punpckhqdq_r2r(xmm5, xmm2); /* xmm2=h5g5f5e5d5c5b5a5 */ \
+ movdqu_r2m(xmm1, *I(2)); /* save I(2) */ \
+ \
+ movdqu_r2m(xmm3, *I(3)); /* save I(3) */ \
+ movdqu_r2m(xmm4, *I(4)); /* save I(4) */ \
+ \
+ movdqu_r2m(xmm2, *I(5)); /* save I(5) */ \
+ movdqu_r2r(xmm7, xmm5); /* make a copy */ \
+ \
+ punpcklqdq_r2r(xmm6, xmm5); /* xmm5=h6g6f6e6d6c6b6a6 */ \
+ punpckhqdq_r2r(xmm6, xmm7); /* xmm7=h7g7f7e7d7c7b7a7 */ \
+ \
+ movdqu_r2m(xmm5, *I(6)); /* save I(6) */ \
+ movdqu_r2m(xmm7, *I(7)); /* save I(7) */ \
+ \
+} /* End of Transpose Macro */
+
+
+#define SSE2_Dequantize() { \
+ movdqu_m2r(*(eax), xmm0); \
+ \
+ pmullw_m2r(*(ebx), xmm0); /* xmm0 = 07 06 05 04 03 02 01 00 */ \
+ movdqu_m2r(*(eax + 16), xmm1); \
+ \
+ pmullw_m2r(*(ebx + 16), xmm1); /* xmm1 = 17 16 15 14 13 12 11 10 */ \
+ pshuflw_r2r(xmm0, xmm3, 0x078); /* xmm3 = 07 06 05 04 01 03 02 00 */ \
+ \
+ movdqu_r2r(xmm1, xmm2); /* xmm2 = 17 16 15 14 13 12 11 10 */ \
+ movdqu_m2r(*(ecx), xmm7); /* xmm7 = -- -- -- -- -- FF FF -- */ \
+ \
+ movdqu_m2r(*(eax + 32), xmm4); \
+ movdqu_m2r(*(eax + 64), xmm5); \
+ \
+ pmullw_m2r(*(ebx + 32), xmm4); /* xmm4 = 27 26 25 24 23 22 21 20 */ \
+ pmullw_m2r(*(ebx + 64), xmm5); /* xmm5 = 47 46 45 44 43 42 41 40 */ \
+ \
+ movdqu_m2r(*(ecx + 16), xmm6); /* xmm6 = -- -- FF FF -- -- -- -- */ \
+ pand_r2r(xmm2, xmm7); /* xmm7 = -- -- -- -- -- 12 11 -- */ \
+ \
+ pand_r2r(xmm4, xmm6); /* xmm6 = -- -- 25 24 -- -- -- -- */ \
+ pxor_r2r(xmm7, xmm2); /* xmm2 = 17 16 15 14 13 -- -- 10 */ \
+ \
+ pxor_r2r(xmm6, xmm4); /* xmm4 = 27 26 -- -- 23 22 21 20 */ \
+ pslldq_i2r(4, xmm7); /* xmm7 = -- -- -- 12 11 -- -- -- */ \
+ \
+ pslldq_i2r(2, xmm6); /* xmm6 = -- 25 24 -- -- -- -- -- */ \
+ por_r2r(xmm6, xmm7); /* xmm7 = -- 25 24 12 11 -- -- -- */ \
+ \
+ movdqu_m2r(*(ecx + 32), xmm0); /* xmm0 = -- -- -- -- -- FF FF FF */ \
+ movdqu_m2r(*(ecx + 48), xmm6); /* xmm6 = -- -- -- -- FF -- -- -- */ \
+ \
+ pand_r2r(xmm3, xmm0); /* xmm0 = -- -- -- -- -- 03 02 00 */ \
+ pand_r2r(xmm5, xmm6); /* xmm6 = -- -- -- -- 43 -- -- -- */ \
+ \
+ pxor_r2r(xmm0, xmm3); /* xmm3 = 07 06 05 04 01 -- -- -- */ \
+ pxor_r2r(xmm6, xmm5); /* xmm5 = 47 46 45 44 -- 42 41 40 */ \
+ \
+ por_r2r(xmm7, xmm0); /* xmm0 = -- 25 24 12 11 03 02 00 */ \
+ pslldq_i2r(8, xmm6); /* xmm6 = 43 -- -- -- -- -- -- -- */ \
+ \
+ por_r2r(xmm6, xmm0); /* xmm0 = 43 25 24 12 11 03 02 00 */ \
+ /* 02345 in use */ \
+ \
+ movdqu_m2r(*(ecx + 64 ), xmm1); /* xmm1 = -- -- -- FF FF -- -- -- */ \
+ pshuflw_r2r(xmm5, xmm5, 0x0B4); /* xmm5 = 47 46 45 44 42 -- 41 40 */ \
+ \
+ movdqu_r2r(xmm1, xmm7); /* xmm7 = -- -- -- FF FF -- -- -- */ \
+ movdqu_r2r(xmm1, xmm6); /* xmm6 = -- -- -- FF FF -- -- -- */ \
+ \
+ movdqu_r2m(xmm0, *(eax)); /* write 43 25 24 12 11 03 02 00 */ \
+ pshufhw_r2r(xmm4, xmm4, 0x0C2); /* xmm4 = 27 -- -- 26 23 22 21 20 */ \
+ \
+ pand_r2r(xmm4, xmm7); /* xmm7 = -- -- -- 26 23 -- -- -- */ \
+ pand_r2r(xmm5, xmm1); /* xmm1 = -- -- -- 44 42 -- -- -- */ \
+ \
+ pxor_r2r(xmm7, xmm4); /* xmm4 = 27 -- -- -- -- 22 21 20 */ \
+ pxor_r2r(xmm1, xmm5); /* xmm5 = 47 46 45 -- -- -- 41 40 */ \
+ \
+ pshuflw_r2r(xmm2, xmm2, 0x0C6); /* xmm2 = 17 16 15 14 13 10 -- -- */ \
+ movdqu_r2r(xmm6, xmm0); /* xmm0 = -- -- -- FF FF -- -- -- */ \
+ \
+ pslldq_i2r(2, xmm7); /* xmm7 = -- -- 26 23 -- -- -- -- */ \
+ pslldq_i2r(6, xmm1); /* xmm1 = 44 42 -- -- -- -- -- -- */ \
+ \
+ psrldq_i2r(2, xmm0); /* xmm0 = -- -- -- -- FF FF -- -- */ \
+ pand_r2r(xmm3, xmm6); /* xmm6 = -- -- -- 04 01 -- -- -- */ \
+ \
+ pand_r2r(xmm2, xmm0); /* xmm0 = -- -- -- -- 13 10 -- -- */ \
+ pxor_r2r(xmm6, xmm3); /* xmm3 = 07 06 05 -- -- -- -- -- */ \
+ \
+ pxor_r2r(xmm0, xmm2); /* xmm2 = 17 16 15 14 -- -- -- -- */ \
+ psrldq_i2r(6, xmm6); /* xmm0 = -- -- -- -- -- -- 04 01 */ \
+ \
+ por_r2r(xmm7, xmm1); /* xmm1 = 44 42 26 23 -- -- -- -- */ \
+ por_r2r(xmm6, xmm0); /* xmm1 = -- -- -- -- 13 10 04 01 */ \
+ /* 12345 in use */ \
+ por_r2r(xmm0, xmm1); /* xmm1 = 44 42 26 23 13 10 04 01 */ \
+ pshuflw_r2r(xmm4, xmm4, 0x093); /* xmm4 = 27 -- -- -- 22 21 20 -- */ \
+ \
+ pshufhw_r2r(xmm4, xmm4, 0x093); /* xmm4 = -- -- -- 27 22 21 20 -- */ \
+ movdqu_r2m(xmm1, *(eax + 16)); /* write 44 42 26 23 13 10 04 01 */ \
+ \
+ pshufhw_r2r(xmm3, xmm3, 0x0D2); /* xmm3 = 07 05 -- 06 -- -- -- -- */ \
+ movdqu_m2r(*(ecx + 64), xmm0); /* xmm0 = -- -- -- FF FF -- -- -- */ \
+ \
+ pand_r2r(xmm3, xmm0); /* xmm0 = -- -- -- 06 -- -- -- -- */ \
+ psrldq_i2r(12, xmm3); /* xmm3 = -- -- -- -- -- -- 07 05 */ \
+ \
+ psrldq_i2r(8, xmm0); /* xmm0 = -- -- -- -- -- -- -- 06 */ \
+ \
+ movdqu_m2r(*(ecx + 64), xmm6); /* xmm6 = -- -- -- FF FF -- -- -- */ \
+ movdqu_m2r(*(ecx + 96), xmm7); /* xmm7 = -- -- -- -- FF FF -- -- */ \
+ \
+ pand_r2r(xmm4, xmm6); /* xmm6 = -- -- -- 27 22 -- -- -- */ \
+ pxor_r2r(xmm6, xmm4); /* xmm4 = -- -- -- -- -- 21 20 -- */ \
+ \
+ por_r2r(xmm6, xmm3); /* xmm3 = -- -- -- 27 22 -- 07 05 */ \
+ pand_r2r(xmm4, xmm7); /* xmm7 = -- -- -- -- -- 21 -- -- */ \
+ \
+ por_r2r(xmm7, xmm0); /* xmm0 = -- -- -- -- -- 21 -- 06 */ \
+ pxor_r2r(xmm7, xmm4); /* xmm4 = -- -- -- -- -- -- 20 -- */ \
+ \
+ movdqu_m2r(*(ecx + 16 ), xmm6); /* xmm6 = -- -- FF FF -- -- -- -- */ \
+ movdqu_m2r(*(ecx + 64 ), xmm1); /* xmm1 = -- -- -- FF FF -- -- -- */ \
+ \
+ pand_r2r(xmm2, xmm6); /* xmm6 = -- -- 15 14 -- -- -- -- */ \
+ pand_r2r(xmm6, xmm1); /* xmm1 = -- -- -- 14 -- -- -- -- */ \
+ \
+ pxor_r2r(xmm6, xmm2); /* xmm2 = 17 16 -- -- -- -- -- -- */ \
+ pxor_r2r(xmm1, xmm6); /* xmm6 = -- -- 15 -- -- -- -- -- */ \
+ \
+ psrldq_i2r(4, xmm1); /* xmm1 = -- -- -- -- -- 14 -- -- */ \
+ \
+ psrldq_i2r(8, xmm6); /* xmm6 = -- -- -- -- -- -- 15 -- */ \
+ por_r2r(xmm1, xmm3); /* xmm3 = -- -- -- 27 22 14 07 05 */ \
+ \
+ por_r2r(xmm6, xmm0); /* xmm0 = -- -- -- -- -- 21 15 06 */ \
+ pshufhw_r2r(xmm5, xmm5, 0x0E1); /* xmm5 = 47 46 -- 45 -- -- 41 40 */ \
+ \
+ movdqu_m2r(*(ecx + 64), xmm1); /* xmm1 = -- -- -- FF FF -- -- -- */ \
+ pshuflw_r2r(xmm5, xmm5, 0x072); /* xmm5 = 47 46 -- 45 41 -- 40 -- */ \
+ \
+ movdqu_r2r(xmm1, xmm6); /* xmm6 = -- -- -- FF FF -- -- -- */ \
+ pand_r2r(xmm5, xmm1); /* xmm1 = -- -- -- 45 41 -- -- -- */ \
+ \
+ pxor_r2r(xmm1, xmm5); /* xmm5 = 47 46 -- -- -- -- 40 -- */ \
+ pslldq_i2r(4, xmm1); /* xmm1 = -- 45 41 -- -- -- -- -- */ \
+ \
+ pshufd_r2r(xmm5, xmm5, 0x09C); /* xmm5 = -- -- -- -- 47 46 40 -- */ \
+ por_r2r(xmm1, xmm3); /* xmm3 = -- 45 41 27 22 14 07 05 */ \
+ \
+ movdqu_m2r(*(eax + 96), xmm1); /* xmm1 = 67 66 65 64 63 62 61 60 */ \
+ pmullw_m2r(*(ebx + 96), xmm1); \
+ \
+ movdqu_m2r(*(ecx), xmm7); /* xmm7 = -- -- -- -- -- FF FF -- */ \
+ \
+ psrldq_i2r(8, xmm6); /* xmm6 = -- -- -- -- -- -- -- FF */ \
+ pand_r2r(xmm5, xmm7); /* xmm7 = -- -- -- -- -- 46 40 -- */ \
+ \
+ pand_r2r(xmm1, xmm6); /* xmm6 = -- -- -- -- -- -- -- 60 */ \
+ pxor_r2r(xmm7, xmm5); /* xmm5 = -- -- -- -- 47 -- -- -- */ \
+ \
+ pxor_r2r(xmm6, xmm1); /* xmm1 = 67 66 65 64 63 62 61 -- */ \
+ pslldq_i2r(2, xmm5); /* xmm5 = -- -- -- 47 -- -- -- -- */ \
+ \
+ pslldq_i2r(14, xmm6); /* xmm6 = 60 -- -- -- -- -- -- -- */ \
+ por_r2r(xmm5, xmm4); /* xmm4 = -- -- -- 47 -- -- 20 -- */ \
+ \
+ por_r2r(xmm6, xmm3); /* xmm3 = 60 45 41 27 22 14 07 05 */ \
+ pslldq_i2r(6, xmm7); /* xmm7 = -- -- 46 40 -- -- -- -- */ \
+ \
+ movdqu_r2m(xmm3, *(eax+32)); /* write 60 45 41 27 22 14 07 05 */ \
+ por_r2r(xmm7, xmm0); /* xmm0 = -- -- 46 40 -- 21 15 06 */ \
+ /* 0, 1, 2, 4 in use */ \
+ movdqu_m2r(*(eax + 48), xmm3); /* xmm3 = 37 36 35 34 33 32 31 30 */ \
+ movdqu_m2r(*(eax + 80), xmm5); /* xmm5 = 57 56 55 54 53 52 51 50 */ \
+ \
+ pmullw_m2r(*(ebx + 48), xmm3); \
+ pmullw_m2r(*(ebx + 80), xmm5); \
+ \
+ movdqu_m2r(*(ecx + 64), xmm6); /* xmm6 = -- -- -- FF FF -- -- -- */ \
+ movdqu_m2r(*(ecx + 64), xmm7); /* xmm7 = -- -- -- FF FF -- -- -- */ \
+ \
+ psrldq_i2r(8, xmm6); /* xmm6 = -- -- -- -- -- -- -- FF */ \
+ pslldq_i2r(8, xmm7); /* xmm7 = FF -- -- -- -- -- -- -- */ \
+ \
+ pand_r2r(xmm3, xmm6); /* xmm6 = -- -- -- -- -- -- -- 30 */ \
+ pand_r2r(xmm5, xmm7); /* xmm7 = 57 -- -- -- -- -- -- -- */ \
+ \
+ pxor_r2r(xmm6, xmm3); /* xmm3 = 37 36 35 34 33 32 31 -- */ \
+ pxor_r2r(xmm7, xmm5); /* xmm5 = __ 56 55 54 53 52 51 50 */ \
+ \
+ pslldq_i2r(6, xmm6); /* xmm6 = -- -- -- -- 30 -- -- -- */ \
+ psrldq_i2r(2, xmm7); /* xmm7 = -- 57 -- -- -- -- -- -- */ \
+ \
+ por_r2r(xmm7, xmm6); /* xmm6 = -- 57 -- -- 30 -- -- -- */ \
+ movdqu_m2r(*(ecx), xmm7); /* xmm7 = -- -- -- -- -- FF FF -- */ \
+ \
+ por_r2r(xmm6, xmm0); /* xmm0 = -- 57 46 40 30 21 15 06 */ \
+ psrldq_i2r(2, xmm7); /* xmm7 = -- -- -- -- -- -- FF FF */ \
+ \
+ movdqu_r2r(xmm2, xmm6); /* xmm6 = 17 16 -- -- -- -- -- -- */ \
+ pand_r2r(xmm1, xmm7); /* xmm7 = -- -- -- -- -- -- 61 -- */ \
+ \
+ pslldq_i2r(2, xmm6); /* xmm6 = 16 -- -- -- -- -- -- -- */ \
+ psrldq_i2r(14, xmm2); /* xmm2 = -- -- -- -- -- -- -- 17 */ \
+ \
+ pxor_r2r(xmm7, xmm1); /* xmm1 = 67 66 65 64 63 62 -- -- */ \
+ pslldq_i2r(12, xmm7); /* xmm7 = 61 -- -- -- -- -- -- -- */ \
+ \
+ psrldq_i2r(14, xmm6); /* xmm6 = -- -- -- -- -- -- -- 16 */ \
+ por_r2r(xmm6, xmm4); /* xmm4 = -- -- -- 47 -- -- 20 16 */ \
+ \
+ por_r2r(xmm7, xmm0); /* xmm0 = 61 57 46 40 30 21 15 06 */ \
+ movdqu_m2r(*(ecx), xmm6); /* xmm6 = -- -- -- -- -- FF FF -- */ \
+ \
+ psrldq_i2r(2, xmm6); /* xmm6 = -- -- -- -- -- -- FF FF */ \
+ movdqu_r2m(xmm0, *(eax+48)); /* write 61 57 46 40 30 21 15 06 */ \
+ /* 1, 2, 3, 4, 5 in use */\
+ movdqu_m2r(*(ecx), xmm0); /* xmm0 = -- -- -- -- -- FF FF -- */ \
+ pand_r2r(xmm3, xmm6); /* xmm6 = -- -- -- -- -- -- 31 -- */ \
+ \
+ movdqu_r2r(xmm3, xmm7); /* xmm7 = 37 36 35 34 33 32 31 -- */ \
+ pxor_r2r(xmm6, xmm3); /* xmm3 = 37 36 35 34 33 32 -- -- */ \
+ \
+ pslldq_i2r(2, xmm3); /* xmm3 = 36 35 34 33 32 -- -- -- */ \
+ pand_r2r(xmm1, xmm0); /* xmm0 = -- -- -- -- -- 62 -- -- */ \
+ \
+ psrldq_i2r(14, xmm7); /* xmm7 = -- -- -- -- -- -- -- 37 */ \
+ pxor_r2r(xmm0, xmm1); /* xmm1 = 67 66 65 64 63 -- -- -- */ \
+ \
+ por_r2r(xmm7, xmm6); /* xmm6 = -- -- -- -- -- -- 31 37 */ \
+ movdqu_m2r(*(ecx + 64), xmm7); /* xmm7 = -- -- -- FF FF -- -- -- */ \
+ \
+ pshuflw_r2r(xmm6, xmm6, 0x01E); /* xmm6 = -- -- -- -- 37 31 -- -- */ \
+ pslldq_i2r(6, xmm7); /* xmm7 = FF FF -- -- -- -- -- -- */ \
+ \
+ por_r2r(xmm6, xmm4); /* xmm4 = -- -- -- 47 37 31 20 16 */ \
+ pand_r2r(xmm5, xmm7); /* xmm7 = -- 56 -- -- -- -- -- -- */ \
+ \
+ pslldq_i2r(8, xmm0); /* xmm0 = -- 62 -- -- -- -- -- -- */ \
+ pxor_r2r(xmm7, xmm5); /* xmm5 = -- -- 55 54 53 52 51 50 */ \
+ \
+ psrldq_i2r(2, xmm7); /* xmm7 = -- -- 56 -- -- -- -- -- */ \
+ \
+ pshufhw_r2r(xmm3, xmm3, 0x087); /* xmm3 = 35 33 34 36 32 -- -- -- */ \
+ por_r2r(xmm7, xmm0); /* xmm0 = -- 62 56 -- -- -- -- -- */ \
+ \
+ movdqu_m2r(*(eax + 112), xmm7); /* xmm7 = 77 76 75 74 73 72 71 70 */ \
+ pmullw_m2r(*(ebx + 112), xmm7); \
+ \
+ movdqu_m2r(*(ecx + 64), xmm6); /* xmm6 = -- -- -- FF FF -- -- -- */ \
+ por_r2r(xmm0, xmm4); /* xmm4 = -- 62 56 47 37 31 20 16 */ \
+ \
+ pshuflw_r2r(xmm7, xmm7, 0x0E1); /* xmm7 = 77 76 75 74 73 72 70 71 */ \
+ psrldq_i2r(8, xmm6); /* xmm6 = -- -- -- -- -- -- -- FF */ \
+ \
+ movdqu_m2r(*(ecx + 64), xmm0); /* xmm0 = -- -- -- FF FF -- -- -- */ \
+ pand_r2r(xmm7, xmm6); /* xmm6 = -- -- -- -- -- -- -- 71 */ \
+ \
+ pand_r2r(xmm3, xmm0); /* xmm0 = -- -- -- 36 32 -- -- -- */ \
+ pxor_r2r(xmm6, xmm7); /* xmm7 = 77 76 75 74 73 72 70 -- */ \
+ \
+ pxor_r2r(xmm0, xmm3); /* xmm3 = 35 33 34 -- -- -- -- -- */ \
+ pslldq_i2r(14, xmm6); /* xmm6 = 71 -- -- -- -- -- -- -- */ \
+ \
+ psrldq_i2r(4, xmm0); /* xmm0 = -- -- -- -- -- 36 32 -- */ \
+ por_r2r(xmm6, xmm4); /* xmm4 = 71 62 56 47 37 31 20 16 */ \
+ \
+ por_r2r(xmm0, xmm2); /* xmm2 = -- -- -- -- -- 36 32 17 */ \
+ movdqu_r2m(xmm4, *(eax + 64)); /* write 71 62 56 47 37 31 20 16 */ \
+ /* 1, 2, 3, 5, 7 in use */ \
+ movdqu_m2r(*(ecx + 80), xmm6); /* xmm6 = -- -- FF -- -- -- -- FF */ \
+ pshufhw_r2r(xmm7, xmm7, 0x0D2); /* xmm7 = 77 75 74 76 73 72 70 __ */ \
+ \
+ movdqu_m2r(*(ecx), xmm4); /* xmm4 = -- -- -- -- -- FF FF -- */ \
+ movdqu_m2r(*(ecx+48), xmm0); /* xmm0 = -- -- -- -- FF -- -- -- */ \
+ \
+ pand_r2r(xmm5, xmm6); /* xmm6 = -- -- 55 -- -- -- -- 50 */ \
+ pand_r2r(xmm7, xmm4); /* xmm4 = -- -- -- -- -- 72 70 -- */ \
+ \
+ pand_r2r(xmm1, xmm0); /* xmm0 = -- -- -- -- 63 -- -- -- */ \
+ pxor_r2r(xmm6, xmm5); /* xmm5 = -- -- -- 54 53 52 51 -- */ \
+ \
+ pxor_r2r(xmm4, xmm7); /* xmm7 = 77 75 74 76 73 -- -- -- */ \
+ pxor_r2r(xmm0, xmm1); /* xmm1 = 67 66 65 64 -- -- -- -- */ \
+ \
+ pshuflw_r2r(xmm6, xmm6, 0x02B); /* xmm6 = -- -- 55 -- 50 -- -- -- */ \
+ pslldq_i2r(10, xmm4); /* xmm4 = 72 20 -- -- -- -- -- -- */ \
+ \
+ pshufhw_r2r(xmm6, xmm6, 0x0B1); /* xmm6 = -- -- -- 55 50 -- -- -- */ \
+ pslldq_i2r(4, xmm0); /* xmm0 = -- -- 63 -- -- -- -- -- */ \
+ \
+ por_r2r(xmm4, xmm6); /* xmm6 = 72 70 -- 55 50 -- -- -- */ \
+ por_r2r(xmm0, xmm2); /* xmm2 = -- -- 63 -- -- 36 32 17 */ \
+ \
+ por_r2r(xmm6, xmm2); /* xmm2 = 72 70 64 55 50 36 32 17 */ \
+ pshufhw_r2r(xmm1, xmm1, 0x0C9); /* xmm1 = 67 64 66 65 -- -- -- -- */ \
+ \
+ movdqu_r2r(xmm3, xmm6); /* xmm6 = 35 33 34 -- -- -- -- -- */ \
+ movdqu_r2m(xmm2, *(eax+80)); /* write 72 70 64 55 50 36 32 17 */ \
+ \
+ psrldq_i2r(12, xmm6); /* xmm6 = -- -- -- -- -- -- 35 33 */ \
+ pslldq_i2r(4, xmm3); /* xmm3 = 34 -- -- -- -- -- -- -- */ \
+ \
+ pshuflw_r2r(xmm5, xmm5, 0x04E); /* xmm5 = -- -- -- 54 51 -- 53 52 */ \
+ movdqu_r2r(xmm7, xmm4); /* xmm4 = 77 75 74 76 73 -- -- -- */ \
+ \
+ movdqu_r2r(xmm5, xmm2); /* xmm2 = -- -- -- 54 51 -- 53 52 */ \
+ psrldq_i2r(10, xmm7); /* xmm7 = -- -- -- -- -- 77 75 74 */ \
+ \
+ pslldq_i2r(6, xmm4); /* xmm4 = 76 73 -- -- -- -- -- -- */ \
+ pslldq_i2r(12, xmm2); /* xmm2 = 53 52 -- -- -- -- -- -- */ \
+ \
+ movdqu_r2r(xmm1, xmm0); /* xmm0 = 67 64 66 65 -- -- -- -- */ \
+ psrldq_i2r(12, xmm1); /* xmm1 = -- -- -- -- -- -- 67 64 */ \
+ \
+ psrldq_i2r(6, xmm5); /* xmm5 = -- -- -- -- -- -- 54 51 */ \
+ psrldq_i2r(14, xmm3); /* xmm3 = -- -- -- -- -- -- -- 34 */ \
+ \
+ pslldq_i2r(10, xmm7); /* xmm7 = 77 75 74 -- -- -- -- -- */ \
+ por_r2r(xmm6, xmm4); /* xmm4 = 76 73 -- -- -- -- 35 33 */ \
+ \
+ psrldq_i2r(10, xmm2); /* xmm2 = -- -- -- -- -- 53 52 -- */ \
+ pslldq_i2r(4, xmm0); /* xmm0 = 66 65 -- -- -- -- -- -- */ \
+ \
+ pslldq_i2r(8, xmm1); /* xmm1 = -- -- 67 64 -- -- -- -- */ \
+ por_r2r(xmm7, xmm3); /* xmm3 = 77 75 74 -- -- -- -- 34 */ \
+ \
+ psrldq_i2r(6, xmm0); /* xmm0 = -- -- -- 66 65 -- -- -- */ \
+ pslldq_i2r(4, xmm5); /* xmm5 = -- -- -- -- 54 51 -- -- */ \
+ \
+ por_r2r(xmm1, xmm4); /* xmm4 = 76 73 67 64 -- -- 35 33 */ \
+ por_r2r(xmm2, xmm3); /* xmm3 = 77 75 74 -- -- 53 52 34 */ \
+ \
+ por_r2r(xmm5, xmm4); /* xmm4 = 76 73 67 64 54 51 35 33 */ \
+ por_r2r(xmm0, xmm3); /* xmm3 = 77 75 74 66 65 53 52 34 */ \
+ \
+ movdqu_r2m(xmm4, *(eax+96)); /* write 76 73 67 64 54 51 35 33 */ \
+ movdqu_r2m(xmm3, *(eax+112)); /* write 77 75 74 66 65 53 52 34 */ \
+ \
+} /* end of SSE2_Dequantize Macro */
+
+
+void vp3_dsp_init_sse2(void)
+{
+ /* nop */
+}
+
+
+static void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix,
+ int16_t *output_data)
+{
+ unsigned char *input_bytes = (unsigned char *)input_data;
+ unsigned char *dequant_matrix_bytes = (unsigned char *)dequant_matrix;
+ unsigned char *dequant_const_bytes = (unsigned char *)SSE2_dequant_const;
+ unsigned char *output_data_bytes = (unsigned char *)output_data;
+ unsigned char *idct_data_bytes = (unsigned char *)SSE2_idct_data;
+ unsigned char *Eight = (unsigned char *)eight_data;
+
+#define eax input_bytes
+#define ebx dequant_matrix_bytes
+#define ecx dequant_const_bytes
+#define edx idct_data_bytes
+
+#define I(i) (eax + 16 * i)
+#define O(i) (ebx + 16 * i)
+#define C(i) (edx + 16 * (i-1))
+
+ SSE2_Dequantize();
+
+#undef ebx
+#define ebx output_data_bytes
+
+ SSE2_Row_IDCT();
+
+ SSE2_Transpose();
+
+ SSE2_Column_IDCT();
+}
+
+
+void vp3_idct_put_sse2(int16_t *input_data, int16_t *dequant_matrix,
+ int coeff_count, uint8_t *dest, int stride)
+{
+ int16_t transformed_data[64];
+ int16_t *op;
+ int i, j;
+
+ vp3_idct_sse2(input_data, dequant_matrix, transformed_data);
+
+ /* place in final output */
+ op = transformed_data;
+ for (i = 0; i < 8; i++) {
+ for (j = 0; j < 8; j++) {
+ if (*op < -128)
+ *dest = 0;
+ else if (*op > 127)
+ *dest = 255;
+ else
+ *dest = (uint8_t)(*op + 128);
+ op++;
+ dest++;
+ }
+ dest += (stride - 8);
+ }
+}
+
+
+void vp3_idct_add_sse2(int16_t *input_data, int16_t *dequant_matrix,
+ int coeff_count, uint8_t *dest, int stride)
+{
+ int16_t transformed_data[64];
+ int16_t *op;
+ int i, j;
+ int16_t sample;
+
+ vp3_idct_sse2(input_data, dequant_matrix, transformed_data);
+
+ /* place in final output */
+ op = transformed_data;
+ for (i = 0; i < 8; i++) {
+ for (j = 0; j < 8; j++) {
+ sample = *dest + *op;
+ if (sample < 0)
+ *dest = 0;
+ else if (sample > 255)
+ *dest = 255;
+ else
+ *dest = (uint8_t)(sample & 0xFF);
+ op++;
+ dest++;
+ }
+ dest += (stride - 8);
+ }
+}
generated by cgit v1.2.3 (git 2.25.1) at 2025-07-24 21:35:43 +0000