--- libmpeg2-0.5.1.orig/libmpeg2/idct_mmx.c 2008-07-09 23:49:47.000000000 +0200 +++ libmpeg2-0.5.1/libmpeg2/idct_mmx.c 2012-02-21 01:10:05.363747630 +0100 @@ -39,6 +39,9 @@ #define rounder(bias) {round (bias), round (bias)} #define rounder_sse2(bias) {round (bias), round (bias), round (bias), round (bias)} +struct x4 {int32_t x[4];}; +struct x16_8 {int16_t x[8];}; +struct x16_4 {int16_t x[4];}; #if 0 /* C row IDCT - it is just here to document the MMXEXT and MMX versions */ @@ -81,14 +84,14 @@ /* SSE2 row IDCT */ -#define sse2_table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \ - c4, -c6, c4, -c2, \ - c4, c6, -c4, -c2, \ - -c4, c2, c4, -c6, \ - c1, c3, c3, -c7, \ - c5, -c1, c7, -c5, \ - c5, c7, -c1, -c5, \ - c7, c3, c3, -c1 } +#define sse2_table(c1,c2,c3,c4,c5,c6,c7) {{{ c4, c2, c4, c6, \ + c4, -c6, c4, -c2}}, \ + {{ c4, c6, -c4, -c2, \ + -c4, c2, c4, -c6}}, \ + {{ c1, c3, c3, -c7, \ + c5, -c1, c7, -c5}}, \ + {{ c5, c7, -c1, -c5, \ + c7, c3, c3, -c1}} } #define SSE2_IDCT_2ROW(table, row1, row2, round1, round2) do { \ /* no scheduling: trust in out of order execution */ \ @@ -99,27 +102,27 @@ pshufd_r2r (row1, xmm1, 0); /* 1: xmm1= x2 x0 x2 x0 x2 x0 x2 x0 */ \ pmaddwd_m2r (table[0], xmm1); /* 1: xmm1= x2*C + x0*C ... */ \ pshufd_r2r (row1, xmm3, 0xaa); /* 1: xmm3= x3 x1 x3 x1 x3 x1 x3 x1 */ \ - pmaddwd_m2r (table[2*8], xmm3); /* 1: xmm3= x3*C + x1*C ... */ \ + pmaddwd_m2r (table[2], xmm3); /* 1: xmm3= x3*C + x1*C ... */ \ pshufd_r2r (row1, xmm2, 0x55); /* 1: xmm2= x6 x4 x6 x4 x6 x4 x6 x4 */ \ pshufd_r2r (row1, row1, 0xff); /* 1: row1= x7 x5 x7 x5 x7 x5 x7 x5 */ \ - pmaddwd_m2r (table[1*8], xmm2); /* 1: xmm2= x6*C + x4*C ... */ \ + pmaddwd_m2r (table[1], xmm2); /* 1: xmm2= x6*C + x4*C ... */ \ paddd_m2r (round1, xmm1); /* 1: xmm1= x2*C + x0*C + round ... */ \ - pmaddwd_m2r (table[3*8], row1); /* 1: row1= x7*C + x5*C ... */ \ + pmaddwd_m2r (table[3], row1); /* 1: row1= x7*C + x5*C ... */ \ pshufd_r2r (row2, xmm5, 0); /* 2: */ \ pshufd_r2r (row2, xmm6, 0x55); /* 2: */ \ pmaddwd_m2r (table[0], xmm5); /* 2: */ \ paddd_r2r (xmm2, xmm1); /* 1: xmm1= a[] */ \ movdqa_r2r (xmm1, xmm2); /* 1: xmm2= a[] */ \ pshufd_r2r (row2, xmm7, 0xaa); /* 2: */ \ - pmaddwd_m2r (table[1*8], xmm6); /* 2: */ \ + pmaddwd_m2r (table[1], xmm6); /* 2: */ \ paddd_r2r (xmm3, row1); /* 1: row1= b[]= 7*C+5*C+3*C+1*C ... */ \ pshufd_r2r (row2, row2, 0xff); /* 2: */ \ psubd_r2r (row1, xmm2); /* 1: xmm2= a[] - b[] */ \ - pmaddwd_m2r (table[2*8], xmm7); /* 2: */ \ + pmaddwd_m2r (table[2], xmm7); /* 2: */ \ paddd_r2r (xmm1, row1); /* 1: row1= a[] + b[] */ \ psrad_i2r (ROW_SHIFT, xmm2); /* 1: xmm2= result 4...7 */ \ paddd_m2r (round2, xmm5); /* 2: */ \ - pmaddwd_m2r (table[3*8], row2); /* 2: */ \ + pmaddwd_m2r (table[3], row2); /* 2: */ \ paddd_r2r (xmm6, xmm5); /* 2: */ \ movdqa_r2r (xmm5, xmm6); /* 2: */ \ psrad_i2r (ROW_SHIFT, row1); /* 1: row1= result 0...4 */ \ @@ -137,54 +140,54 @@ /* MMXEXT row IDCT */ -#define mmxext_table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, -c4, -c2, \ - c4, c6, c4, c6, \ - c1, c3, -c1, -c5, \ - c5, c7, c3, -c7, \ - c4, -c6, c4, -c6, \ - -c4, c2, c4, -c2, \ - c5, -c1, c3, -c1, \ - c7, c3, c7, -c5 } +#define mmxext_table(c1,c2,c3,c4,c5,c6,c7) {{{ c4, c2, -c4, -c2}}, \ + {{ c4, c6, c4, c6}}, \ + {{ c1, c3, -c1, -c5}}, \ + {{ c5, c7, c3, -c7}}, \ + {{ c4, -c6, c4, -c6}}, \ + {{-c4, c2, c4, -c2}}, \ + {{ c5, -c1, c3, -c1}}, \ + {{ c7, c3, c7, -c5}}} static inline void mmxext_row_head (int16_t * const row, const int offset, - const int16_t * const table) + const struct x16_4 * const table) { movq_m2r (*(row+offset), mm2); /* mm2 = x6 x4 x2 x0 */ movq_m2r (*(row+offset+4), mm5); /* mm5 = x7 x5 x3 x1 */ movq_r2r (mm2, mm0); /* mm0 = x6 x4 x2 x0 */ - movq_m2r (*table, mm3); /* mm3 = -C2 -C4 C2 C4 */ + movq_m2r (table[0], mm3); /* mm3 = -C2 -C4 C2 C4 */ movq_r2r (mm5, mm6); /* mm6 = x7 x5 x3 x1 */ - movq_m2r (*(table+4), mm4); /* mm4 = C6 C4 C6 C4 */ + movq_m2r (table[1], mm4); /* mm4 = C6 C4 C6 C4 */ pmaddwd_r2r (mm0, mm3); /* mm3 = -C4*x4-C2*x6 C4*x0+C2*x2 */ pshufw_r2r (mm2, mm2, 0x4e); /* mm2 = x2 x0 x6 x4 */ } -static inline void mmxext_row (const int16_t * const table, - const int32_t * const rounder) +static inline void mmxext_row (const struct x16_4 * const table, + const struct x4 * const rounder) { - movq_m2r (*(table+8), mm1); /* mm1 = -C5 -C1 C3 C1 */ + movq_m2r (table[2], mm1); /* mm1 = -C5 -C1 C3 C1 */ pmaddwd_r2r (mm2, mm4); /* mm4 = C4*x0+C6*x2 C4*x4+C6*x6 */ - pmaddwd_m2r (*(table+16), mm0); /* mm0 = C4*x4-C6*x6 C4*x0-C6*x2 */ + pmaddwd_m2r (table[4], mm0); /* mm0 = C4*x4-C6*x6 C4*x0-C6*x2 */ pshufw_r2r (mm6, mm6, 0x4e); /* mm6 = x3 x1 x7 x5 */ - movq_m2r (*(table+12), mm7); /* mm7 = -C7 C3 C7 C5 */ + movq_m2r (table[3], mm7); /* mm7 = -C7 C3 C7 C5 */ pmaddwd_r2r (mm5, mm1); /* mm1 = -C1*x5-C5*x7 C1*x1+C3*x3 */ paddd_m2r (*rounder, mm3); /* mm3 += rounder */ pmaddwd_r2r (mm6, mm7); /* mm7 = C3*x1-C7*x3 C5*x5+C7*x7 */ - pmaddwd_m2r (*(table+20), mm2); /* mm2 = C4*x0-C2*x2 -C4*x4+C2*x6 */ + pmaddwd_m2r (table[5], mm2); /* mm2 = C4*x0-C2*x2 -C4*x4+C2*x6 */ paddd_r2r (mm4, mm3); /* mm3 = a1 a0 + rounder */ - pmaddwd_m2r (*(table+24), mm5); /* mm5 = C3*x5-C1*x7 C5*x1-C1*x3 */ + pmaddwd_m2r (table[6], mm5); /* mm5 = C3*x5-C1*x7 C5*x1-C1*x3 */ movq_r2r (mm3, mm4); /* mm4 = a1 a0 + rounder */ - pmaddwd_m2r (*(table+28), mm6); /* mm6 = C7*x1-C5*x3 C7*x5+C3*x7 */ + pmaddwd_m2r (table[7], mm6); /* mm6 = C7*x1-C5*x3 C7*x5+C3*x7 */ paddd_r2r (mm7, mm1); /* mm1 = b1 b0 */ paddd_m2r (*rounder, mm0); /* mm0 += rounder */ @@ -223,7 +226,7 @@ static inline void mmxext_row_mid (int16_t * const row, const int store, const int offset, - const int16_t * const table) + const struct x16_4 * const table) { movq_m2r (*(row+offset), mm2); /* mm2 = x6 x4 x2 x0 */ psrad_i2r (ROW_SHIFT, mm0); /* mm0 = y3 y2 */ @@ -240,69 +243,69 @@ movq_r2m (mm1, *(row+store)); /* save y3 y2 y1 y0 */ pshufw_r2r (mm4, mm4, 0xb1); /* mm4 = y7 y6 y5 y4 */ - movq_m2r (*table, mm3); /* mm3 = -C2 -C4 C2 C4 */ + movq_m2r (table[0], mm3); /* mm3 = -C2 -C4 C2 C4 */ movq_r2m (mm4, *(row+store+4)); /* save y7 y6 y5 y4 */ pmaddwd_r2r (mm0, mm3); /* mm3 = -C4*x4-C2*x6 C4*x0+C2*x2 */ - movq_m2r (*(table+4), mm4); /* mm4 = C6 C4 C6 C4 */ + movq_m2r (table[1], mm4); /* mm4 = C6 C4 C6 C4 */ pshufw_r2r (mm2, mm2, 0x4e); /* mm2 = x2 x0 x6 x4 */ } /* MMX row IDCT */ -#define mmx_table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \ - c4, c6, -c4, -c2, \ - c1, c3, c3, -c7, \ - c5, c7, -c1, -c5, \ - c4, -c6, c4, -c2, \ - -c4, c2, c4, -c6, \ - c5, -c1, c7, -c5, \ - c7, c3, c3, -c1 } +#define mmx_table(c1,c2,c3,c4,c5,c6,c7) {{{ c4, c2, c4, c6}}, \ + {{ c4, c6, -c4, -c2}}, \ + {{ c1, c3, c3, -c7}}, \ + {{ c5, c7, -c1, -c5}}, \ + {{ c4, -c6, c4, -c2}}, \ + {{-c4, c2, c4, -c6}}, \ + {{ c5, -c1, c7, -c5}}, \ + {{ c7, c3, c3, -c1}}} static inline void mmx_row_head (int16_t * const row, const int offset, - const int16_t * const table) + const struct x16_4 * const table) { movq_m2r (*(row+offset), mm2); /* mm2 = x6 x4 x2 x0 */ movq_m2r (*(row+offset+4), mm5); /* mm5 = x7 x5 x3 x1 */ movq_r2r (mm2, mm0); /* mm0 = x6 x4 x2 x0 */ - movq_m2r (*table, mm3); /* mm3 = C6 C4 C2 C4 */ + movq_m2r (table[0], mm3); /* mm3 = C6 C4 C2 C4 */ movq_r2r (mm5, mm6); /* mm6 = x7 x5 x3 x1 */ punpckldq_r2r (mm0, mm0); /* mm0 = x2 x0 x2 x0 */ - movq_m2r (*(table+4), mm4); /* mm4 = -C2 -C4 C6 C4 */ + movq_m2r (table[1], mm4); /* mm4 = -C2 -C4 C6 C4 */ pmaddwd_r2r (mm0, mm3); /* mm3 = C4*x0+C6*x2 C4*x0+C2*x2 */ - movq_m2r (*(table+8), mm1); /* mm1 = -C7 C3 C3 C1 */ + movq_m2r (table[2], mm1); /* mm1 = -C7 C3 C3 C1 */ punpckhdq_r2r (mm2, mm2); /* mm2 = x6 x4 x6 x4 */ } -static inline void mmx_row (const int16_t * const table, - const int32_t * const rounder) +static inline void mmx_row (const struct x16_4 * const table, + const struct x4 * const rounder) { pmaddwd_r2r (mm2, mm4); /* mm4 = -C4*x4-C2*x6 C4*x4+C6*x6 */ punpckldq_r2r (mm5, mm5); /* mm5 = x3 x1 x3 x1 */ - pmaddwd_m2r (*(table+16), mm0); /* mm0 = C4*x0-C2*x2 C4*x0-C6*x2 */ + pmaddwd_m2r (table[4], mm0); /* mm0 = C4*x0-C2*x2 C4*x0-C6*x2 */ punpckhdq_r2r (mm6, mm6); /* mm6 = x7 x5 x7 x5 */ - movq_m2r (*(table+12), mm7); /* mm7 = -C5 -C1 C7 C5 */ + movq_m2r (table[3], mm7); /* mm7 = -C5 -C1 C7 C5 */ pmaddwd_r2r (mm5, mm1); /* mm1 = C3*x1-C7*x3 C1*x1+C3*x3 */ paddd_m2r (*rounder, mm3); /* mm3 += rounder */ pmaddwd_r2r (mm6, mm7); /* mm7 = -C1*x5-C5*x7 C5*x5+C7*x7 */ - pmaddwd_m2r (*(table+20), mm2); /* mm2 = C4*x4-C6*x6 -C4*x4+C2*x6 */ + pmaddwd_m2r (table[5], mm2); /* mm2 = C4*x4-C6*x6 -C4*x4+C2*x6 */ paddd_r2r (mm4, mm3); /* mm3 = a1 a0 + rounder */ - pmaddwd_m2r (*(table+24), mm5); /* mm5 = C7*x1-C5*x3 C5*x1-C1*x3 */ + pmaddwd_m2r (table[6], mm5); /* mm5 = C7*x1-C5*x3 C5*x1-C1*x3 */ movq_r2r (mm3, mm4); /* mm4 = a1 a0 + rounder */ - pmaddwd_m2r (*(table+28), mm6); /* mm6 = C3*x5-C1*x7 C7*x5+C3*x7 */ + pmaddwd_m2r (table[7], mm6); /* mm6 = C3*x5-C1*x7 C7*x5+C3*x7 */ paddd_r2r (mm7, mm1); /* mm1 = b1 b0 */ paddd_m2r (*rounder, mm0); /* mm0 += rounder */ @@ -346,7 +349,7 @@ } static inline void mmx_row_mid (int16_t * const row, const int store, - const int offset, const int16_t * const table) + const int offset, const struct x16_4 * const table) { movq_m2r (*(row+offset), mm2); /* mm2 = x6 x4 x2 x0 */ psrad_i2r (ROW_SHIFT, mm0); /* mm0 = y3 y2 */ @@ -366,13 +369,13 @@ punpckldq_r2r (mm0, mm0); /* mm0 = x2 x0 x2 x0 */ psrld_i2r (16, mm7); /* mm7 = 0 y6 0 y4 */ - movq_m2r (*table, mm3); /* mm3 = C6 C4 C2 C4 */ + movq_m2r (table[0], mm3); /* mm3 = C6 C4 C2 C4 */ pslld_i2r (16, mm1); /* mm1 = y7 0 y5 0 */ - movq_m2r (*(table+4), mm4); /* mm4 = -C2 -C4 C6 C4 */ + movq_m2r (table[1], mm4); /* mm4 = -C2 -C4 C6 C4 */ por_r2r (mm1, mm7); /* mm7 = y7 y6 y5 y4 */ - movq_m2r (*(table+8), mm1); /* mm1 = -C7 C3 C3 C1 */ + movq_m2r (table[2], mm1); /* mm1 = -C7 C3 C3 C1 */ punpckhdq_r2r (mm2, mm2); /* mm2 = x6 x4 x6 x4 */ movq_r2m (mm7, *(row+store+4)); /* save y7 y6 y5 y4 */ @@ -457,6 +460,11 @@ #define T3 43790 #define C4 23170 +/* MMX-Ext. and SSE can share the constants */ +static const struct x16_8 t1_vector ATTR_ALIGN(16) = {{T1,T1,T1,T1,T1,T1,T1,T1}}; +static const struct x16_8 t2_vector ATTR_ALIGN(16) = {{T2,T2,T2,T2,T2,T2,T2,T2}}; +static const struct x16_8 t3_vector ATTR_ALIGN(16) = {{T3,T3,T3,T3,T3,T3,T3,T3}}; +static const struct x16_8 c4_vector ATTR_ALIGN(16) = {{C4,C4,C4,C4,C4,C4,C4,C4}}; /* SSE2 column IDCT */ static inline void sse2_idct_col (int16_t * const col) @@ -464,28 +472,23 @@ /* Almost identical to mmxext version: */ /* just do both 4x8 columns in paraller */ - static const short t1_vector[] ATTR_ALIGN(16) = {T1,T1,T1,T1,T1,T1,T1,T1}; - static const short t2_vector[] ATTR_ALIGN(16) = {T2,T2,T2,T2,T2,T2,T2,T2}; - static const short t3_vector[] ATTR_ALIGN(16) = {T3,T3,T3,T3,T3,T3,T3,T3}; - static const short c4_vector[] ATTR_ALIGN(16) = {C4,C4,C4,C4,C4,C4,C4,C4}; - #if defined(__x86_64__) /* INPUT: block in xmm8 ... xmm15 */ - movdqa_m2r (*t1_vector, xmm0); /* xmm0 = T1 */ + movdqa_m2r (t1_vector, xmm0); /* xmm0 = T1 */ movdqa_r2r (xmm9, xmm1); /* xmm1 = x1 */ movdqa_r2r (xmm0, xmm2); /* xmm2 = T1 */ pmulhw_r2r (xmm1, xmm0); /* xmm0 = T1*x1 */ - movdqa_m2r (*t3_vector, xmm5); /* xmm5 = T3 */ + movdqa_m2r (t3_vector, xmm5); /* xmm5 = T3 */ pmulhw_r2r (xmm15, xmm2); /* xmm2 = T1*x7 */ movdqa_r2r (xmm5, xmm7); /* xmm7 = T3-1 */ psubsw_r2r (xmm15, xmm0); /* xmm0 = v17 */ - movdqa_m2r (*t2_vector, xmm9); /* xmm9 = T2 */ + movdqa_m2r (t2_vector, xmm9); /* xmm9 = T2 */ pmulhw_r2r (xmm11, xmm5); /* xmm5 = (T3-1)*x3 */ paddsw_r2r (xmm2, xmm1); /* xmm1 = u17 */ @@ -518,7 +521,7 @@ movdqa_r2r (xmm1, xmm7); /* xmm7 = u12 */ paddsw_r2r (xmm5, xmm1); /* xmm1 = u12+v12 */ - movdqa_m2r (*c4_vector, xmm0); /* xmm0 = C4/2 */ + movdqa_m2r (c4_vector, xmm0); /* xmm0 = C4/2 */ psubsw_r2r (xmm5, xmm7); /* xmm7 = u12-v12 */ movdqa_r2r (xmm6, xmm4); /* xmm4 = b0 */ @@ -577,7 +580,7 @@ /* OUTPUT: block in xmm8 ... xmm15 */ #else - movdqa_m2r (*t1_vector, xmm0); /* xmm0 = T1 */ + movdqa_m2r (t1_vector, xmm0); /* xmm0 = T1 */ movdqa_m2r (*(col+1*8), xmm1); /* xmm1 = x1 */ movdqa_r2r (xmm0, xmm2); /* xmm2 = T1 */ @@ -585,7 +588,7 @@ movdqa_m2r (*(col+7*8), xmm4); /* xmm4 = x7 */ pmulhw_r2r (xmm1, xmm0); /* xmm0 = T1*x1 */ - movdqa_m2r (*t3_vector, xmm5); /* xmm5 = T3 */ + movdqa_m2r (t3_vector, xmm5); /* xmm5 = T3 */ pmulhw_r2r (xmm4, xmm2); /* xmm2 = T1*x7 */ movdqa_m2r (*(col+5*8), xmm6); /* xmm6 = x5 */ @@ -594,7 +597,7 @@ movdqa_m2r (*(col+3*8), xmm3); /* xmm3 = x3 */ psubsw_r2r (xmm4, xmm0); /* xmm0 = v17 */ - movdqa_m2r (*t2_vector, xmm4); /* xmm4 = T2 */ + movdqa_m2r (t2_vector, xmm4); /* xmm4 = T2 */ pmulhw_r2r (xmm3, xmm5); /* xmm5 = (T3-1)*x3 */ paddsw_r2r (xmm2, xmm1); /* xmm1 = u17 */ @@ -632,7 +635,7 @@ movdqa_m2r (*(col+0*8), xmm3); /* xmm3 = x0 */ paddsw_r2r (xmm5, xmm1); /* xmm1 = u12+v12 */ - movdqa_m2r (*c4_vector, xmm0); /* xmm0 = C4/2 */ + movdqa_m2r (c4_vector, xmm0); /* xmm0 = C4/2 */ psubsw_r2r (xmm5, xmm7); /* xmm7 = u12-v12 */ movdqa_r2m (xmm6, *(col+5*8)); /* save b0 in scratch1 */ @@ -710,15 +713,10 @@ /* MMX column IDCT */ static inline void idct_col (int16_t * const col, const int offset) { - static const short t1_vector[] ATTR_ALIGN(8) = {T1,T1,T1,T1}; - static const short t2_vector[] ATTR_ALIGN(8) = {T2,T2,T2,T2}; - static const short t3_vector[] ATTR_ALIGN(8) = {T3,T3,T3,T3}; - static const short c4_vector[] ATTR_ALIGN(8) = {C4,C4,C4,C4}; - /* column code adapted from peter gubanov */ /* http://www.elecard.com/peter/idct.shtml */ - movq_m2r (*t1_vector, mm0); /* mm0 = T1 */ + movq_m2r (t1_vector, mm0); /* mm0 = T1 */ movq_m2r (*(col+offset+1*8), mm1); /* mm1 = x1 */ movq_r2r (mm0, mm2); /* mm2 = T1 */ @@ -726,7 +724,7 @@ movq_m2r (*(col+offset+7*8), mm4); /* mm4 = x7 */ pmulhw_r2r (mm1, mm0); /* mm0 = T1*x1 */ - movq_m2r (*t3_vector, mm5); /* mm5 = T3 */ + movq_m2r (t3_vector, mm5); /* mm5 = T3 */ pmulhw_r2r (mm4, mm2); /* mm2 = T1*x7 */ movq_m2r (*(col+offset+5*8), mm6); /* mm6 = x5 */ @@ -735,7 +733,7 @@ movq_m2r (*(col+offset+3*8), mm3); /* mm3 = x3 */ psubsw_r2r (mm4, mm0); /* mm0 = v17 */ - movq_m2r (*t2_vector, mm4); /* mm4 = T2 */ + movq_m2r (t2_vector, mm4); /* mm4 = T2 */ pmulhw_r2r (mm3, mm5); /* mm5 = (T3-1)*x3 */ paddsw_r2r (mm2, mm1); /* mm1 = u17 */ @@ -773,7 +771,7 @@ movq_m2r (*(col+offset+0*8), mm3); /* mm3 = x0 */ paddsw_r2r (mm5, mm1); /* mm1 = u12+v12 */ - movq_m2r (*c4_vector, mm0); /* mm0 = C4/2 */ + movq_m2r (c4_vector, mm0); /* mm0 = C4/2 */ psubsw_r2r (mm5, mm7); /* mm7 = u12-v12 */ movq_r2m (mm6, *(col+offset+5*8)); /* save b0 in scratch1 */ @@ -846,52 +844,51 @@ movq_r2m (mm4, *(col+offset+3*8)); /* save y3 */ } - -static const int32_t rounder0[] ATTR_ALIGN(8) = - rounder ((1 << (COL_SHIFT - 1)) - 0.5); -static const int32_t rounder4[] ATTR_ALIGN(8) = rounder (0); -static const int32_t rounder1[] ATTR_ALIGN(8) = - rounder (1.25683487303); /* C1*(C1/C4+C1+C7)/2 */ -static const int32_t rounder7[] ATTR_ALIGN(8) = - rounder (-0.25); /* C1*(C7/C4+C7-C1)/2 */ -static const int32_t rounder2[] ATTR_ALIGN(8) = - rounder (0.60355339059); /* C2 * (C6+C2)/2 */ -static const int32_t rounder6[] ATTR_ALIGN(8) = - rounder (-0.25); /* C2 * (C6-C2)/2 */ -static const int32_t rounder3[] ATTR_ALIGN(8) = - rounder (0.087788325588); /* C3*(-C3/C4+C3+C5)/2 */ -static const int32_t rounder5[] ATTR_ALIGN(8) = - rounder (-0.441341716183); /* C3*(-C5/C4+C5-C3)/2 */ - +/* MMX and SSE can share these constants */ +static const struct x4 rounder0 ATTR_ALIGN(16) = + {rounder_sse2 ((1 << (COL_SHIFT - 1)) - 0.5)}; +static const struct x4 rounder4 ATTR_ALIGN(16) = {rounder_sse2 (0)}; +static const struct x4 rounder1 ATTR_ALIGN(16) = + {rounder_sse2 (1.25683487303)}; /* C1*(C1/C4+C1+C7)/2 */ +static const struct x4 rounder7 ATTR_ALIGN(16) = + {rounder_sse2 (-0.25)}; /* C1*(C7/C4+C7-C1)/2 */ +static const struct x4 rounder2 ATTR_ALIGN(16) = + {rounder_sse2 (0.60355339059)}; /* C2 * (C6+C2)/2 */ +static const struct x4 rounder6 ATTR_ALIGN(16) = + {rounder_sse2 (-0.25)}; /* C2 * (C6-C2)/2 */ +static const struct x4 rounder3 ATTR_ALIGN(16) = + {rounder_sse2 (0.087788325588)}; /* C3*(-C3/C4+C3+C5)/2 */ +static const struct x4 rounder5 ATTR_ALIGN(16) = + {rounder_sse2 (-0.441341716183)}; /* C3*(-C5/C4+C5-C3)/2 */ #define declare_idct(idct,table,idct_row_head,idct_row,idct_row_tail,idct_row_mid) \ static inline void idct (int16_t * const block) \ { \ - static const int16_t table04[] ATTR_ALIGN(16) = \ + static const struct x16_4 table04[8] ATTR_ALIGN(16) = \ table (22725, 21407, 19266, 16384, 12873, 8867, 4520); \ - static const int16_t table17[] ATTR_ALIGN(16) = \ + static const struct x16_4 table17[8] ATTR_ALIGN(16) = \ table (31521, 29692, 26722, 22725, 17855, 12299, 6270); \ - static const int16_t table26[] ATTR_ALIGN(16) = \ + static const struct x16_4 table26[8] ATTR_ALIGN(16) = \ table (29692, 27969, 25172, 21407, 16819, 11585, 5906); \ - static const int16_t table35[] ATTR_ALIGN(16) = \ + static const struct x16_4 table35[8] ATTR_ALIGN(16) = \ table (26722, 25172, 22654, 19266, 15137, 10426, 5315); \ \ idct_row_head (block, 0*8, table04); \ - idct_row (table04, rounder0); \ + idct_row (table04, &rounder0); \ idct_row_mid (block, 0*8, 4*8, table04); \ - idct_row (table04, rounder4); \ + idct_row (table04, &rounder4); \ idct_row_mid (block, 4*8, 1*8, table17); \ - idct_row (table17, rounder1); \ + idct_row (table17, &rounder1); \ idct_row_mid (block, 1*8, 7*8, table17); \ - idct_row (table17, rounder7); \ + idct_row (table17, &rounder7); \ idct_row_mid (block, 7*8, 2*8, table26); \ - idct_row (table26, rounder2); \ + idct_row (table26, &rounder2); \ idct_row_mid (block, 2*8, 6*8, table26); \ - idct_row (table26, rounder6); \ + idct_row (table26, &rounder6); \ idct_row_mid (block, 6*8, 3*8, table35); \ - idct_row (table35, rounder3); \ + idct_row (table35, &rounder3); \ idct_row_mid (block, 3*8, 5*8, table35); \ - idct_row (table35, rounder5); \ + idct_row (table35, &rounder5); \ idct_row_tail (block, 5*8); \ \ idct_col (block, 0); \ @@ -900,72 +897,56 @@ static inline void sse2_idct (int16_t * const block) { - static const int16_t table04[] ATTR_ALIGN(16) = + static const struct x16_8 table04[4] ATTR_ALIGN(16) = sse2_table (22725, 21407, 19266, 16384, 12873, 8867, 4520); - static const int16_t table17[] ATTR_ALIGN(16) = + static const struct x16_8 table17[4] ATTR_ALIGN(16) = sse2_table (31521, 29692, 26722, 22725, 17855, 12299, 6270); - static const int16_t table26[] ATTR_ALIGN(16) = + static const struct x16_8 table26[4] ATTR_ALIGN(16) = sse2_table (29692, 27969, 25172, 21407, 16819, 11585, 5906); - static const int16_t table35[] ATTR_ALIGN(16) = + static const struct x16_8 table35[4] ATTR_ALIGN(16) = sse2_table (26722, 25172, 22654, 19266, 15137, 10426, 5315); - static const int32_t rounder0_128[] ATTR_ALIGN(16) = - rounder_sse2 ((1 << (COL_SHIFT - 1)) - 0.5); - static const int32_t rounder4_128[] ATTR_ALIGN(16) = rounder_sse2 (0); - static const int32_t rounder1_128[] ATTR_ALIGN(16) = - rounder_sse2 (1.25683487303); /* C1*(C1/C4+C1+C7)/2 */ - static const int32_t rounder7_128[] ATTR_ALIGN(16) = - rounder_sse2 (-0.25); /* C1*(C7/C4+C7-C1)/2 */ - static const int32_t rounder2_128[] ATTR_ALIGN(16) = - rounder_sse2 (0.60355339059); /* C2 * (C6+C2)/2 */ - static const int32_t rounder6_128[] ATTR_ALIGN(16) = - rounder_sse2 (-0.25); /* C2 * (C6-C2)/2 */ - static const int32_t rounder3_128[] ATTR_ALIGN(16) = - rounder_sse2 (0.087788325588); /* C3*(-C3/C4+C3+C5)/2 */ - static const int32_t rounder5_128[] ATTR_ALIGN(16) = - rounder_sse2 (-0.441341716183); /* C3*(-C5/C4+C5-C3)/2 */ - #if defined(__x86_64__) movdqa_m2r (block[0*8], xmm8); movdqa_m2r (block[4*8], xmm12); - SSE2_IDCT_2ROW (table04, xmm8, xmm12, *rounder0_128, *rounder4_128); + SSE2_IDCT_2ROW (table04, xmm8, xmm12, rounder0, rounder4); movdqa_m2r (block[1*8], xmm9); movdqa_m2r (block[7*8], xmm15); - SSE2_IDCT_2ROW (table17, xmm9, xmm15, *rounder1_128, *rounder7_128); + SSE2_IDCT_2ROW (table17, xmm9, xmm15, rounder1, rounder7); movdqa_m2r (block[2*8], xmm10); movdqa_m2r (block[6*8], xmm14); - SSE2_IDCT_2ROW (table26, xmm10, xmm14, *rounder2_128, *rounder6_128); + SSE2_IDCT_2ROW (table26, xmm10, xmm14, rounder2, rounder6); movdqa_m2r (block[3*8], xmm11); movdqa_m2r (block[5*8], xmm13); - SSE2_IDCT_2ROW (table35, xmm11, xmm13, *rounder3_128, *rounder5_128); + SSE2_IDCT_2ROW (table35, xmm11, xmm13, rounder3, rounder5); /* OUTPUT: block in xmm8 ... xmm15 */ #else movdqa_m2r (block[0*8], xmm0); movdqa_m2r (block[4*8], xmm4); - SSE2_IDCT_2ROW (table04, xmm0, xmm4, *rounder0_128, *rounder4_128); + SSE2_IDCT_2ROW (table04, xmm0, xmm4, rounder0, rounder4); movdqa_r2m (xmm0, block[0*8]); movdqa_r2m (xmm4, block[4*8]); movdqa_m2r (block[1*8], xmm0); movdqa_m2r (block[7*8], xmm4); - SSE2_IDCT_2ROW (table17, xmm0, xmm4, *rounder1_128, *rounder7_128); + SSE2_IDCT_2ROW (table17, xmm0, xmm4, rounder1, rounder7); movdqa_r2m (xmm0, block[1*8]); movdqa_r2m (xmm4, block[7*8]); movdqa_m2r (block[2*8], xmm0); movdqa_m2r (block[6*8], xmm4); - SSE2_IDCT_2ROW (table26, xmm0, xmm4, *rounder2_128, *rounder6_128); + SSE2_IDCT_2ROW (table26, xmm0, xmm4, rounder2, rounder6); movdqa_r2m (xmm0, block[2*8]); movdqa_r2m (xmm4, block[6*8]); movdqa_m2r (block[3*8], xmm0); movdqa_m2r (block[5*8], xmm4); - SSE2_IDCT_2ROW (table35, xmm0, xmm4, *rounder3_128, *rounder5_128); + SSE2_IDCT_2ROW (table35, xmm0, xmm4, rounder3, rounder5); movdqa_r2m (xmm0, block[3*8]); movdqa_r2m (xmm4, block[5*8]); #endif