diff options
Diffstat (limited to 'libavcodec/ppc/dsputil_altivec.c')
| -rw-r--r-- | libavcodec/ppc/dsputil_altivec.c | 403 |
1 files changed, 181 insertions, 222 deletions
diff --git a/libavcodec/ppc/dsputil_altivec.c b/libavcodec/ppc/dsputil_altivec.c index 2091023ed5..a8985fd81a 100644 --- a/libavcodec/ppc/dsputil_altivec.c +++ b/libavcodec/ppc/dsputil_altivec.c @@ -35,34 +35,30 @@ static int sad16_x2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { - int i; - int s; + int i, s = 0; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); vector unsigned char perm1 = vec_lvsl(0, pix2); vector unsigned char perm2 = vec_add(perm1, vec_splat_u8(1)); - vector unsigned char pix2l, pix2r; - vector unsigned char pix1v, pix2v, pix2iv, avgv, t5; - vector unsigned int sad; + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; - s = 0; - sad = (vector unsigned int) vec_splat_u32(0); for (i = 0; i < h; i++) { /* Read unaligned pixels into our vectors. The vectors are as follows: * pix1v: pix1[0] - pix1[15] * pix2v: pix2[0] - pix2[15] pix2iv: pix2[1] - pix2[16] */ - pix1v = vec_ld(0, pix1); - pix2l = vec_ld(0, pix2); - pix2r = vec_ld(16, pix2); - pix2v = vec_perm(pix2l, pix2r, perm1); - pix2iv = vec_perm(pix2l, pix2r, perm2); + vector unsigned char pix1v = vec_ld(0, pix1); + vector unsigned char pix2l = vec_ld(0, pix2); + vector unsigned char pix2r = vec_ld(16, pix2); + vector unsigned char pix2v = vec_perm(pix2l, pix2r, perm1); + vector unsigned char pix2iv = vec_perm(pix2l, pix2r, perm2); /* Calculate the average vector. */ - avgv = vec_avg(pix2v, pix2iv); + vector unsigned char avgv = vec_avg(pix2v, pix2iv); /* Calculate a sum of abs differences vector. */ - t5 = vec_sub(vec_max(pix1v, avgv), vec_min(pix1v, avgv)); + vector unsigned char t5 = vec_sub(vec_max(pix1v, avgv), + vec_min(pix1v, avgv)); /* Add each 4 pixel group together and put 4 results into sad. */ sad = vec_sum4s(t5, sad); @@ -81,20 +77,15 @@ static int sad16_x2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, static int sad16_y2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { - int i; - int s; + int i, s = 0; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); vector unsigned char perm = vec_lvsl(0, pix2); - vector unsigned char pix2l, pix2r; - vector unsigned char pix1v, pix2v, pix3v, avgv, t5; - vector unsigned int sad; + vector unsigned char pix1v, pix3v, avgv, t5; + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; uint8_t *pix3 = pix2 + line_size; - s = 0; - sad = (vector unsigned int) vec_splat_u32(0); - /* Due to the fact that pix3 = pix2 + line_size, the pix3 of one * iteration becomes pix2 in the next iteration. We can use this * fact to avoid a potentially expensive unaligned read, each @@ -102,9 +93,9 @@ static int sad16_y2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, * Read unaligned pixels into our vectors. The vectors are as follows: * pix2v: pix2[0] - pix2[15] * Split the pixel vectors into shorts. */ - pix2l = vec_ld(0, pix2); - pix2r = vec_ld(15, pix2); - pix2v = vec_perm(pix2l, pix2r, perm); + vector unsigned char pix2l = vec_ld(0, pix2); + vector unsigned char pix2r = vec_ld(15, pix2); + vector unsigned char pix2v = vec_perm(pix2l, pix2r, perm); for (i = 0; i < h; i++) { /* Read unaligned pixels into our vectors. The vectors are as follows: @@ -140,8 +131,7 @@ static int sad16_y2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, static int sad16_xy2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { - int i; - int s; + int i, s = 0; uint8_t *pix3 = pix2 + line_size; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); @@ -150,19 +140,12 @@ static int sad16_xy2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, vector unsigned char avgv, t5; vector unsigned char perm1 = vec_lvsl(0, pix2); vector unsigned char perm2 = vec_add(perm1, vec_splat_u8(1)); - vector unsigned char pix2l, pix2r; - vector unsigned char pix1v, pix2v, pix3v, pix2iv, pix3iv; - vector unsigned short pix2lv, pix2hv, pix2ilv, pix2ihv; + vector unsigned char pix1v, pix3v, pix3iv; vector unsigned short pix3lv, pix3hv, pix3ilv, pix3ihv; vector unsigned short avghv, avglv; - vector unsigned short t1, t2, t3, t4; - vector unsigned int sad; + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; - sad = (vector unsigned int) vec_splat_u32(0); - - s = 0; - /* Due to the fact that pix3 = pix2 + line_size, the pix3 of one * iteration becomes pix2 in the next iteration. We can use this * fact to avoid a potentially expensive unaligned read, as well @@ -170,17 +153,22 @@ static int sad16_xy2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, * Read unaligned pixels into our vectors. The vectors are as follows: * pix2v: pix2[0] - pix2[15] pix2iv: pix2[1] - pix2[16] * Split the pixel vectors into shorts. */ - pix2l = vec_ld(0, pix2); - pix2r = vec_ld(16, pix2); - pix2v = vec_perm(pix2l, pix2r, perm1); - pix2iv = vec_perm(pix2l, pix2r, perm2); - - pix2hv = (vector unsigned short) vec_mergeh(zero, pix2v); - pix2lv = (vector unsigned short) vec_mergel(zero, pix2v); - pix2ihv = (vector unsigned short) vec_mergeh(zero, pix2iv); - pix2ilv = (vector unsigned short) vec_mergel(zero, pix2iv); - t1 = vec_add(pix2hv, pix2ihv); - t2 = vec_add(pix2lv, pix2ilv); + vector unsigned char pix2l = vec_ld(0, pix2); + vector unsigned char pix2r = vec_ld(16, pix2); + vector unsigned char pix2v = vec_perm(pix2l, pix2r, perm1); + vector unsigned char pix2iv = vec_perm(pix2l, pix2r, perm2); + + vector unsigned short pix2hv = + (vector unsigned short) vec_mergeh(zero, pix2v); + vector unsigned short pix2lv = + (vector unsigned short) vec_mergel(zero, pix2v); + vector unsigned short pix2ihv = + (vector unsigned short) vec_mergeh(zero, pix2iv); + vector unsigned short pix2ilv = + (vector unsigned short) vec_mergel(zero, pix2iv); + vector unsigned short t1 = vec_add(pix2hv, pix2ihv); + vector unsigned short t2 = vec_add(pix2lv, pix2ilv); + vector unsigned short t3, t4; for (i = 0; i < h; i++) { /* Read unaligned pixels into our vectors. The vectors are as follows: @@ -238,28 +226,24 @@ static int sad16_xy2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, static int sad16_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { - int i; - int s; + int i, s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); vector unsigned char perm = vec_lvsl(0, pix2); - vector unsigned char t1, t2, t3, t4, t5; - vector unsigned int sad; + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; - sad = (vector unsigned int) vec_splat_u32(0); - for (i = 0; i < h; i++) { /* Read potentially unaligned pixels into t1 and t2. */ vector unsigned char pix2l = vec_ld(0, pix2); vector unsigned char pix2r = vec_ld(15, pix2); - t1 = vec_ld(0, pix1); - t2 = vec_perm(pix2l, pix2r, perm); + vector unsigned char t1 = vec_ld(0, pix1); + vector unsigned char t2 = vec_perm(pix2l, pix2r, perm); /* Calculate a sum of abs differences vector. */ - t3 = vec_max(t1, t2); - t4 = vec_min(t1, t2); - t5 = vec_sub(t3, t4); + vector unsigned char t3 = vec_max(t1, t2); + vector unsigned char t4 = vec_min(t1, t2); + vector unsigned char t5 = vec_sub(t3, t4); /* Add each 4 pixel group together and put 4 results into sad. */ sad = vec_sum4s(t5, sad); @@ -279,8 +263,7 @@ static int sad16_altivec(void *v, uint8_t *pix1, uint8_t *pix2, static int sad8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { - int i; - int s; + int i, s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); const vector unsigned char permclear = @@ -288,12 +271,9 @@ static int sad8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, { 255, 255, 255, 255, 255, 255, 255, 255, 0, 0, 0, 0, 0, 0, 0, 0 }; vector unsigned char perm1 = vec_lvsl(0, pix1); vector unsigned char perm2 = vec_lvsl(0, pix2); - vector unsigned char t1, t2, t3, t4, t5; - vector unsigned int sad; + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; - sad = (vector unsigned int) vec_splat_u32(0); - for (i = 0; i < h; i++) { /* Read potentially unaligned pixels into t1 and t2. * Since we're reading 16 pixels, and actually only want 8, @@ -302,13 +282,15 @@ static int sad8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, vector unsigned char pix1r = vec_ld(7, pix1); vector unsigned char pix2l = vec_ld(0, pix2); vector unsigned char pix2r = vec_ld(7, pix2); - t1 = vec_and(vec_perm(pix1l, pix1r, perm1), permclear); - t2 = vec_and(vec_perm(pix2l, pix2r, perm2), permclear); + vector unsigned char t1 = vec_and(vec_perm(pix1l, pix1r, perm1), + permclear); + vector unsigned char t2 = vec_and(vec_perm(pix2l, pix2r, perm2), + permclear); /* Calculate a sum of abs differences vector. */ - t3 = vec_max(t1, t2); - t4 = vec_min(t1, t2); - t5 = vec_sub(t3, t4); + vector unsigned char t3 = vec_max(t1, t2); + vector unsigned char t4 = vec_min(t1, t2); + vector unsigned char t5 = vec_sub(t3, t4); /* Add each 4 pixel group together and put 4 results into sad. */ sad = vec_sum4s(t5, sad); @@ -327,23 +309,18 @@ static int sad8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, static int pix_norm1_altivec(uint8_t *pix, int line_size) { - int i; - int s; + int i, s = 0; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); vector unsigned char perm = vec_lvsl(0, pix); - vector unsigned char pixv; - vector unsigned int sv; + vector unsigned int sv = (vector unsigned int) vec_splat_u32(0); vector signed int sum; - sv = (vector unsigned int) vec_splat_u32(0); - - s = 0; for (i = 0; i < 16; i++) { /* Read the potentially unaligned pixels. */ vector unsigned char pixl = vec_ld(0, pix); vector unsigned char pixr = vec_ld(15, pix); - pixv = vec_perm(pixl, pixr, perm); + vector unsigned char pixv = vec_perm(pixl, pixr, perm); /* Square the values, and add them to our sum. */ sv = vec_msum(pixv, pixv, sv); @@ -363,8 +340,7 @@ static int pix_norm1_altivec(uint8_t *pix, int line_size) static int sse8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { - int i; - int s; + int i, s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); const vector unsigned char permclear = @@ -372,12 +348,9 @@ static int sse8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, { 255, 255, 255, 255, 255, 255, 255, 255, 0, 0, 0, 0, 0, 0, 0, 0 }; vector unsigned char perm1 = vec_lvsl(0, pix1); vector unsigned char perm2 = vec_lvsl(0, pix2); - vector unsigned char t1, t2, t3, t4, t5; - vector unsigned int sum; + vector unsigned int sum = (vector unsigned int) vec_splat_u32(0); vector signed int sumsqr; - sum = (vector unsigned int) vec_splat_u32(0); - for (i = 0; i < h; i++) { /* Read potentially unaligned pixels into t1 and t2. * Since we're reading 16 pixels, and actually only want 8, @@ -386,16 +359,18 @@ static int sse8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, vector unsigned char pix1r = vec_ld(7, pix1); vector unsigned char pix2l = vec_ld(0, pix2); vector unsigned char pix2r = vec_ld(7, pix2); - t1 = vec_and(vec_perm(pix1l, pix1r, perm1), permclear); - t2 = vec_and(vec_perm(pix2l, pix2r, perm2), permclear); + vector unsigned char t1 = vec_and(vec_perm(pix1l, pix1r, perm1), + permclear); + vector unsigned char t2 = vec_and(vec_perm(pix2l, pix2r, perm2), + permclear); /* Since we want to use unsigned chars, we can take advantage * of the fact that abs(a - b) ^ 2 = (a - b) ^ 2. */ /* Calculate abs differences vector. */ - t3 = vec_max(t1, t2); - t4 = vec_min(t1, t2); - t5 = vec_sub(t3, t4); + vector unsigned char t3 = vec_max(t1, t2); + vector unsigned char t4 = vec_min(t1, t2); + vector unsigned char t5 = vec_sub(t3, t4); /* Square the values and add them to our sum. */ sum = vec_msum(t5, t5, sum); @@ -417,31 +392,27 @@ static int sse8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, static int sse16_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { - int i; - int s; + int i, s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); vector unsigned char perm = vec_lvsl(0, pix2); - vector unsigned char t1, t2, t3, t4, t5; - vector unsigned int sum; + vector unsigned int sum = (vector unsigned int) vec_splat_u32(0); vector signed int sumsqr; - sum = (vector unsigned int) vec_splat_u32(0); - for (i = 0; i < h; i++) { /* Read potentially unaligned pixels into t1 and t2. */ vector unsigned char pix2l = vec_ld(0, pix2); vector unsigned char pix2r = vec_ld(15, pix2); - t1 = vec_ld(0, pix1); - t2 = vec_perm(pix2l, pix2r, perm); + vector unsigned char t1 = vec_ld(0, pix1); + vector unsigned char t2 = vec_perm(pix2l, pix2r, perm); /* Since we want to use unsigned chars, we can take advantage * of the fact that abs(a - b) ^ 2 = (a - b) ^ 2. */ /* Calculate abs differences vector. */ - t3 = vec_max(t1, t2); - t4 = vec_min(t1, t2); - t5 = vec_sub(t3, t4); + vector unsigned char t3 = vec_max(t1, t2); + vector unsigned char t4 = vec_min(t1, t2); + vector unsigned char t5 = vec_sub(t3, t4); /* Square the values and add them to our sum. */ sum = vec_msum(t5, t5, sum); @@ -460,23 +431,18 @@ static int sse16_altivec(void *v, uint8_t *pix1, uint8_t *pix2, static int pix_sum_altivec(uint8_t *pix, int line_size) { + int i, s; const vector unsigned int zero = (const vector unsigned int) vec_splat_u32(0); vector unsigned char perm = vec_lvsl(0, pix); - vector unsigned char t1; - vector unsigned int sad; + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); vector signed int sumdiffs; - int i; - int s; - - sad = (vector unsigned int) vec_splat_u32(0); - for (i = 0; i < 16; i++) { /* Read the potentially unaligned 16 pixels into t1. */ vector unsigned char pixl = vec_ld(0, pix); vector unsigned char pixr = vec_ld(15, pix); - t1 = vec_perm(pixl, pixr, perm); + vector unsigned char t1 = vec_perm(pixl, pixr, perm); /* Add each 4 pixel group together and put 4 results into sad. */ sad = vec_sum4s(t1, sad); @@ -497,10 +463,8 @@ static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, { int i; vector unsigned char perm = vec_lvsl(0, pixels); - vector unsigned char bytes; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); - vector signed short shorts; for (i = 0; i < 8; i++) { /* Read potentially unaligned pixels. @@ -508,10 +472,11 @@ static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, * but we simply ignore the extras. */ vector unsigned char pixl = vec_ld(0, pixels); vector unsigned char pixr = vec_ld(7, pixels); - bytes = vec_perm(pixl, pixr, perm); + vector unsigned char bytes = vec_perm(pixl, pixr, perm); // Convert the bytes into shorts. - shorts = (vector signed short) vec_mergeh(zero, bytes); + vector signed short shorts = (vector signed short) vec_mergeh(zero, + bytes); // Save the data to the block, we assume the block is 16-byte aligned. vec_st(shorts, i * 16, (vector signed short *) block); @@ -526,7 +491,6 @@ static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, int i; vector unsigned char perm1 = vec_lvsl(0, s1); vector unsigned char perm2 = vec_lvsl(0, s2); - vector unsigned char bytes, pixl, pixr; const vector unsigned char zero = (const vector unsigned char) vec_splat_u8(0); vector signed short shorts1, shorts2; @@ -535,9 +499,9 @@ static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, /* Read potentially unaligned pixels. * We're reading 16 pixels, and actually only want 8, * but we simply ignore the extras. */ - pixl = vec_ld(0, s1); - pixr = vec_ld(15, s1); - bytes = vec_perm(pixl, pixr, perm1); + vector unsigned char pixl = vec_ld(0, s1); + vector unsigned char pixr = vec_ld(15, s1); + vector unsigned char bytes = vec_perm(pixl, pixr, perm1); // Convert the bytes into shorts. shorts1 = (vector signed short) vec_mergeh(zero, bytes); @@ -653,29 +617,31 @@ static int hadamard8_diff8x8_altivec(/* MpegEncContext */ void *s, uint8_t *dst, #define ONEITERBUTTERFLY(i, res) \ { \ - register vector unsigned char src1, src2, srcO; \ - register vector unsigned char dst1, dst2, dstO; \ - register vector signed short srcV, dstV; \ - register vector signed short but0, but1, but2, op1, op2, op3; \ - src1 = vec_ld(stride * i, src); \ - src2 = vec_ld(stride * i + 15, src); \ - srcO = vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ - dst1 = vec_ld(stride * i, dst); \ - dst2 = vec_ld(stride * i + 15, dst); \ - dstO = vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ + register vector unsigned char src1 = vec_ld(stride * i, src); \ + register vector unsigned char src2 = vec_ld(stride * i + 15, src); \ + register vector unsigned char srcO = \ + vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ + register vector unsigned char dst1 = vec_ld(stride * i, dst); \ + register vector unsigned char dst2 = vec_ld(stride * i + 15, dst); \ + register vector unsigned char dstO = \ + vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ + \ /* Promote the unsigned chars to signed shorts. */ \ /* We're in the 8x8 function, we only care for the first 8. */ \ - srcV = (vector signed short) vec_mergeh((vector signed char) vzero, \ - (vector signed char) srcO); \ - dstV = (vector signed short) vec_mergeh((vector signed char) vzero, \ - (vector signed char) dstO); \ + register vector signed short srcV = \ + (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed char) srcO); \ + register vector signed short dstV = \ + (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed char) dstO); \ + \ /* subtractions inside the first butterfly */ \ - but0 = vec_sub(srcV, dstV); \ - op1 = vec_perm(but0, but0, perm1); \ - but1 = vec_mladd(but0, vprod1, op1); \ - op2 = vec_perm(but1, but1, perm2); \ - but2 = vec_mladd(but1, vprod2, op2); \ - op3 = vec_perm(but2, but2, perm3); \ + register vector signed short but0 = vec_sub(srcV, dstV); \ + register vector signed short op1 = vec_perm(but0, but0, perm1); \ + register vector signed short but1 = vec_mladd(but0, vprod1, op1); \ + register vector signed short op2 = vec_perm(but1, but1, perm2); \ + register vector signed short but2 = vec_mladd(but1, vprod2, op2); \ + register vector signed short op3 = vec_perm(but2, but2, perm3); \ res = vec_mladd(but2, vprod3, op3); \ } ONEITERBUTTERFLY(0, temp0); @@ -801,62 +767,60 @@ static int hadamard8_diff16x8_altivec(/* MpegEncContext */ void *s, uint8_t *dst #define ONEITERBUTTERFLY(i, res1, res2) \ { \ - register vector unsigned char \ - src1 __asm__ ("v22"), \ - src2 __asm__ ("v23"), \ - dst1 __asm__ ("v24"), \ - dst2 __asm__ ("v25"), \ - srcO __asm__ ("v22"), \ - dstO __asm__ ("v23"); \ + register vector unsigned char src1 __asm__ ("v22") = \ + vec_ld(stride * i, src); \ + register vector unsigned char src2 __asm__ ("v23") = \ + vec_ld(stride * i + 16, src); \ + register vector unsigned char srcO __asm__ ("v22") = \ + vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ + register vector unsigned char dst1 __asm__ ("v24") = \ + vec_ld(stride * i, dst); \ + register vector unsigned char dst2 __asm__ ("v25") = \ + vec_ld(stride * i + 16, dst); \ + register vector unsigned char dstO __asm__ ("v23") = \ + vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ \ - register vector signed short \ - srcV __asm__ ("v24"), \ - dstV __asm__ ("v25"), \ - srcW __asm__ ("v26"), \ - dstW __asm__ ("v27"), \ - but0 __asm__ ("v28"), \ - but0S __asm__ ("v29"), \ - op1 __asm__ ("v30"), \ - but1 __asm__ ("v22"), \ - op1S __asm__ ("v23"), \ - but1S __asm__ ("v24"), \ - op2 __asm__ ("v25"), \ - but2 __asm__ ("v26"), \ - op2S __asm__ ("v27"), \ - but2S __asm__ ("v28"), \ - op3 __asm__ ("v29"), \ - op3S __asm__ ("v30"); \ - \ - src1 = vec_ld(stride * i, src); \ - src2 = vec_ld(stride * i + 16, src); \ - srcO = vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ - dst1 = vec_ld(stride * i, dst); \ - dst2 = vec_ld(stride * i + 16, dst); \ - dstO = vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ /* Promote the unsigned chars to signed shorts. */ \ - srcV = (vector signed short) vec_mergeh((vector signed char) vzero, \ - (vector signed char) srcO); \ - dstV = (vector signed short) vec_mergeh((vector signed char) vzero, \ - (vector signed char) dstO); \ - srcW = (vector signed short) vec_mergel((vector signed char) vzero, \ - (vector signed char) srcO); \ - dstW = (vector signed short) vec_mergel((vector signed char) vzero, \ - (vector signed char) dstO); \ + register vector signed short srcV __asm__ ("v24") = \ + (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed char) srcO); \ + register vector signed short dstV __asm__ ("v25") = \ + (vector signed short) vec_mergeh((vector signed char) vzero, \ + (vector signed char) dstO); \ + register vector signed short srcW __asm__ ("v26") = \ + (vector signed short) vec_mergel((vector signed char) vzero, \ + (vector signed char) srcO); \ + register vector signed short dstW __asm__ ("v27") = \ + (vector signed short) vec_mergel((vector signed char) vzero, \ + (vector signed char) dstO); \ + \ /* subtractions inside the first butterfly */ \ - but0 = vec_sub(srcV, dstV); \ - but0S = vec_sub(srcW, dstW); \ - op1 = vec_perm(but0, but0, perm1); \ - but1 = vec_mladd(but0, vprod1, op1); \ - op1S = vec_perm(but0S, but0S, perm1); \ - but1S = vec_mladd(but0S, vprod1, op1S); \ - op2 = vec_perm(but1, but1, perm2); \ - but2 = vec_mladd(but1, vprod2, op2); \ - op2S = vec_perm(but1S, but1S, perm2); \ - but2S = vec_mladd(but1S, vprod2, op2S); \ - op3 = vec_perm(but2, but2, perm3); \ - res1 = vec_mladd(but2, vprod3, op3); \ - op3S = vec_perm(but2S, but2S, perm3); \ - res2 = vec_mladd(but2S, vprod3, op3S); \ + register vector signed short but0 __asm__ ("v28") = \ + vec_sub(srcV, dstV); \ + register vector signed short but0S __asm__ ("v29") = \ + vec_sub(srcW, dstW); \ + register vector signed short op1 __asm__ ("v30") = \ + vec_perm(but0, but0, perm1); \ + register vector signed short but1 __asm__ ("v22") = \ + vec_mladd(but0, vprod1, op1); \ + register vector signed short op1S __asm__ ("v23") = \ + vec_perm(but0S, but0S, perm1); \ + register vector signed short but1S __asm__ ("v24") = \ + vec_mladd(but0S, vprod1, op1S); \ + register vector signed short op2 __asm__ ("v25") = \ + vec_perm(but1, but1, perm2); \ + register vector signed short but2 __asm__ ("v26") = \ + vec_mladd(but1, vprod2, op2); \ + register vector signed short op2S __asm__ ("v27") = \ + vec_perm(but1S, but1S, perm2); \ + register vector signed short but2S __asm__ ("v28") = \ + vec_mladd(but1S, vprod2, op2S); \ + register vector signed short op3 __asm__ ("v29") = \ + vec_perm(but2, but2, perm3); \ + register vector signed short op3S __asm__ ("v30") = \ + vec_perm(but2S, but2S, perm3); \ + res1 = vec_mladd(but2, vprod3, op3); \ + res2 = vec_mladd(but2S, vprod3, op3S); \ } ONEITERBUTTERFLY(0, temp0, temp0S); ONEITERBUTTERFLY(1, temp1, temp1S); @@ -870,11 +834,6 @@ static int hadamard8_diff16x8_altivec(/* MpegEncContext */ void *s, uint8_t *dst #undef ONEITERBUTTERFLY { register vector signed int vsum; - register vector signed short line0S, line1S, line2S, line3S, line4S, - line5S, line6S, line7S, line0BS, line2BS, - line1BS, line3BS, line4BS, line6BS, line5BS, - line7BS, line0CS, line4CS, line1CS, line5CS, - line2CS, line6CS, line3CS, line7CS; register vector signed short line0 = vec_add(temp0, temp1); register vector signed short line1 = vec_sub(temp0, temp1); @@ -903,6 +862,33 @@ static int hadamard8_diff16x8_altivec(/* MpegEncContext */ void *s, uint8_t *dst register vector signed short line3C = vec_add(line3B, line7B); register vector signed short line7C = vec_sub(line3B, line7B); + register vector signed short line0S = vec_add(temp0S, temp1S); + register vector signed short line1S = vec_sub(temp0S, temp1S); + register vector signed short line2S = vec_add(temp2S, temp3S); + register vector signed short line3S = vec_sub(temp2S, temp3S); + register vector signed short line4S = vec_add(temp4S, temp5S); + register vector signed short line5S = vec_sub(temp4S, temp5S); + register vector signed short line6S = vec_add(temp6S, temp7S); + register vector signed short line7S = vec_sub(temp6S, temp7S); + + register vector signed short line0BS = vec_add(line0S, line2S); + register vector signed short line2BS = vec_sub(line0S, line2S); + register vector signed short line1BS = vec_add(line1S, line3S); + register vector signed short line3BS = vec_sub(line1S, line3S); + register vector signed short line4BS = vec_add(line4S, line6S); + register vector signed short line6BS = vec_sub(line4S, line6S); + register vector signed short line5BS = vec_add(line5S, line7S); + register vector signed short line7BS = vec_sub(line5S, line7S); + + register vector signed short line0CS = vec_add(line0BS, line4BS); + register vector signed short line4CS = vec_sub(line0BS, line4BS); + register vector signed short line1CS = vec_add(line1BS, line5BS); + register vector signed short line5CS = vec_sub(line1BS, line5BS); + register vector signed short line2CS = vec_add(line2BS, line6BS); + register vector signed short line6CS = vec_sub(line2BS, line6BS); + register vector signed short line3CS = vec_add(line3BS, line7BS); + register vector signed short line7CS = vec_sub(line3BS, line7BS); + vsum = vec_sum4s(vec_abs(line0C), vec_splat_s32(0)); vsum = vec_sum4s(vec_abs(line1C), vsum); vsum = vec_sum4s(vec_abs(line2C), vsum); @@ -912,33 +898,6 @@ static int hadamard8_diff16x8_altivec(/* MpegEncContext */ void *s, uint8_t *dst vsum = vec_sum4s(vec_abs(line6C), vsum); vsum = vec_sum4s(vec_abs(line7C), vsum); - line0S = vec_add(temp0S, temp1S); - line1S = vec_sub(temp0S, temp1S); - line2S = vec_add(temp2S, temp3S); - line3S = vec_sub(temp2S, temp3S); - line4S = vec_add(temp4S, temp5S); - line5S = vec_sub(temp4S, temp5S); - line6S = vec_add(temp6S, temp7S); - line7S = vec_sub(temp6S, temp7S); - - line0BS = vec_add(line0S, line2S); - line2BS = vec_sub(line0S, line2S); - line1BS = vec_add(line1S, line3S); - line3BS = vec_sub(line1S, line3S); - line4BS = vec_add(line4S, line6S); - line6BS = vec_sub(line4S, line6S); - line5BS = vec_add(line5S, line7S); - line7BS = vec_sub(line5S, line7S); - - line0CS = vec_add(line0BS, line4BS); - line4CS = vec_sub(line0BS, line4BS); - line1CS = vec_add(line1BS, line5BS); - line5CS = vec_sub(line1BS, line5BS); - line2CS = vec_add(line2BS, line6BS); - line6CS = vec_sub(line2BS, line6BS); - line3CS = vec_add(line3BS, line7BS); - line7CS = vec_sub(line3BS, line7BS); - vsum = vec_sum4s(vec_abs(line0CS), vsum); vsum = vec_sum4s(vec_abs(line1CS), vsum); vsum = vec_sum4s(vec_abs(line2CS), vsum); @@ -957,8 +916,8 @@ static int hadamard8_diff16x8_altivec(/* MpegEncContext */ void *s, uint8_t *dst static int hadamard8_diff16_altivec(/* MpegEncContext */ void *s, uint8_t *dst, uint8_t *src, int stride, int h) { - int score; - score = hadamard8_diff16x8_altivec(s, dst, src, stride, 8); + int score = hadamard8_diff16x8_altivec(s, dst, src, stride, 8); + if (h == 16) { dst += 8 * stride; src += 8 * stride; |