1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
|
#include <xmmintrin.h>
#include "qcmsint.h"
/* pre-shuffled: just load these into XMM reg instead of load-scalar/shufps sequence */
#define FLOATSCALE (float)(PRECACHE_OUTPUT_SIZE)
#define CLAMPMAXVAL ( ((float) (PRECACHE_OUTPUT_SIZE - 1)) / PRECACHE_OUTPUT_SIZE )
static const ALIGN float floatScaleX4[4] =
{ FLOATSCALE, FLOATSCALE, FLOATSCALE, FLOATSCALE};
static const ALIGN float clampMaxValueX4[4] =
{ CLAMPMAXVAL, CLAMPMAXVAL, CLAMPMAXVAL, CLAMPMAXVAL};
void qcms_transform_data_rgb_out_lut_sse1(qcms_transform *transform,
unsigned char *src,
unsigned char *dest,
size_t length)
{
unsigned int i;
float (*mat)[4] = transform->matrix;
char input_back[32];
/* Ensure we have a buffer that's 16 byte aligned regardless of the original
* stack alignment. We can't use __attribute__((aligned(16))) or __declspec(align(32))
* because they don't work on stack variables. gcc 4.4 does do the right thing
* on x86 but that's too new for us right now. For more info: gcc bug #16660 */
float const * input = (float*)(((uintptr_t)&input_back[16]) & ~0xf);
/* share input and output locations to save having to keep the
* locations in separate registers */
uint32_t const * output = (uint32_t*)input;
/* deref *transform now to avoid it in loop */
const float *igtbl_r = transform->input_gamma_table_r;
const float *igtbl_g = transform->input_gamma_table_g;
const float *igtbl_b = transform->input_gamma_table_b;
/* deref *transform now to avoid it in loop */
const uint8_t *otdata_r = &transform->output_table_r->data[0];
const uint8_t *otdata_g = &transform->output_table_g->data[0];
const uint8_t *otdata_b = &transform->output_table_b->data[0];
/* input matrix values never change */
const __m128 mat0 = _mm_load_ps(mat[0]);
const __m128 mat1 = _mm_load_ps(mat[1]);
const __m128 mat2 = _mm_load_ps(mat[2]);
/* these values don't change, either */
const __m128 max = _mm_load_ps(clampMaxValueX4);
const __m128 min = _mm_setzero_ps();
const __m128 scale = _mm_load_ps(floatScaleX4);
/* working variables */
__m128 vec_r, vec_g, vec_b, result;
/* CYA */
if (!length)
return;
/* one pixel is handled outside of the loop */
length--;
/* setup for transforming 1st pixel */
vec_r = _mm_load_ss(&igtbl_r[src[0]]);
vec_g = _mm_load_ss(&igtbl_g[src[1]]);
vec_b = _mm_load_ss(&igtbl_b[src[2]]);
src += 3;
/* transform all but final pixel */
for (i=0; i<length; i++)
{
/* position values from gamma tables */
vec_r = _mm_shuffle_ps(vec_r, vec_r, 0);
vec_g = _mm_shuffle_ps(vec_g, vec_g, 0);
vec_b = _mm_shuffle_ps(vec_b, vec_b, 0);
/* gamma * matrix */
vec_r = _mm_mul_ps(vec_r, mat0);
vec_g = _mm_mul_ps(vec_g, mat1);
vec_b = _mm_mul_ps(vec_b, mat2);
/* crunch, crunch, crunch */
vec_r = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b));
vec_r = _mm_max_ps(min, vec_r);
vec_r = _mm_min_ps(max, vec_r);
result = _mm_mul_ps(vec_r, scale);
/* store calc'd output tables indices */
*((__m64 *)&output[0]) = _mm_cvtps_pi32(result);
result = _mm_movehl_ps(result, result);
*((__m64 *)&output[2]) = _mm_cvtps_pi32(result) ;
/* load for next loop while store completes */
vec_r = _mm_load_ss(&igtbl_r[src[0]]);
vec_g = _mm_load_ss(&igtbl_g[src[1]]);
vec_b = _mm_load_ss(&igtbl_b[src[2]]);
src += 3;
/* use calc'd indices to output RGB values */
dest[0] = otdata_r[output[0]];
dest[1] = otdata_g[output[1]];
dest[2] = otdata_b[output[2]];
dest += 3;
}
/* handle final (maybe only) pixel */
vec_r = _mm_shuffle_ps(vec_r, vec_r, 0);
vec_g = _mm_shuffle_ps(vec_g, vec_g, 0);
vec_b = _mm_shuffle_ps(vec_b, vec_b, 0);
vec_r = _mm_mul_ps(vec_r, mat0);
vec_g = _mm_mul_ps(vec_g, mat1);
vec_b = _mm_mul_ps(vec_b, mat2);
vec_r = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b));
vec_r = _mm_max_ps(min, vec_r);
vec_r = _mm_min_ps(max, vec_r);
result = _mm_mul_ps(vec_r, scale);
*((__m64 *)&output[0]) = _mm_cvtps_pi32(result);
result = _mm_movehl_ps(result, result);
*((__m64 *)&output[2]) = _mm_cvtps_pi32(result);
dest[0] = otdata_r[output[0]];
dest[1] = otdata_g[output[1]];
dest[2] = otdata_b[output[2]];
_mm_empty();
}
void qcms_transform_data_rgba_out_lut_sse1(qcms_transform *transform,
unsigned char *src,
unsigned char *dest,
size_t length)
{
unsigned int i;
float (*mat)[4] = transform->matrix;
char input_back[32];
/* Ensure we have a buffer that's 16 byte aligned regardless of the original
* stack alignment. We can't use __attribute__((aligned(16))) or __declspec(align(32))
* because they don't work on stack variables. gcc 4.4 does do the right thing
* on x86 but that's too new for us right now. For more info: gcc bug #16660 */
float const * input = (float*)(((uintptr_t)&input_back[16]) & ~0xf);
/* share input and output locations to save having to keep the
* locations in separate registers */
uint32_t const * output = (uint32_t*)input;
/* deref *transform now to avoid it in loop */
const float *igtbl_r = transform->input_gamma_table_r;
const float *igtbl_g = transform->input_gamma_table_g;
const float *igtbl_b = transform->input_gamma_table_b;
/* deref *transform now to avoid it in loop */
const uint8_t *otdata_r = &transform->output_table_r->data[0];
const uint8_t *otdata_g = &transform->output_table_g->data[0];
const uint8_t *otdata_b = &transform->output_table_b->data[0];
/* input matrix values never change */
const __m128 mat0 = _mm_load_ps(mat[0]);
const __m128 mat1 = _mm_load_ps(mat[1]);
const __m128 mat2 = _mm_load_ps(mat[2]);
/* these values don't change, either */
const __m128 max = _mm_load_ps(clampMaxValueX4);
const __m128 min = _mm_setzero_ps();
const __m128 scale = _mm_load_ps(floatScaleX4);
/* working variables */
__m128 vec_r, vec_g, vec_b, result;
unsigned char alpha;
/* CYA */
if (!length)
return;
/* one pixel is handled outside of the loop */
length--;
/* setup for transforming 1st pixel */
vec_r = _mm_load_ss(&igtbl_r[src[0]]);
vec_g = _mm_load_ss(&igtbl_g[src[1]]);
vec_b = _mm_load_ss(&igtbl_b[src[2]]);
alpha = src[3];
src += 4;
/* transform all but final pixel */
for (i=0; i<length; i++)
{
/* position values from gamma tables */
vec_r = _mm_shuffle_ps(vec_r, vec_r, 0);
vec_g = _mm_shuffle_ps(vec_g, vec_g, 0);
vec_b = _mm_shuffle_ps(vec_b, vec_b, 0);
/* gamma * matrix */
vec_r = _mm_mul_ps(vec_r, mat0);
vec_g = _mm_mul_ps(vec_g, mat1);
vec_b = _mm_mul_ps(vec_b, mat2);
/* store alpha for this pixel; load alpha for next */
dest[3] = alpha;
alpha = src[3];
/* crunch, crunch, crunch */
vec_r = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b));
vec_r = _mm_max_ps(min, vec_r);
vec_r = _mm_min_ps(max, vec_r);
result = _mm_mul_ps(vec_r, scale);
/* store calc'd output tables indices */
*((__m64 *)&output[0]) = _mm_cvtps_pi32(result);
result = _mm_movehl_ps(result, result);
*((__m64 *)&output[2]) = _mm_cvtps_pi32(result);
/* load gamma values for next loop while store completes */
vec_r = _mm_load_ss(&igtbl_r[src[0]]);
vec_g = _mm_load_ss(&igtbl_g[src[1]]);
vec_b = _mm_load_ss(&igtbl_b[src[2]]);
src += 4;
/* use calc'd indices to output RGB values */
dest[0] = otdata_r[output[0]];
dest[1] = otdata_g[output[1]];
dest[2] = otdata_b[output[2]];
dest += 4;
}
/* handle final (maybe only) pixel */
vec_r = _mm_shuffle_ps(vec_r, vec_r, 0);
vec_g = _mm_shuffle_ps(vec_g, vec_g, 0);
vec_b = _mm_shuffle_ps(vec_b, vec_b, 0);
vec_r = _mm_mul_ps(vec_r, mat0);
vec_g = _mm_mul_ps(vec_g, mat1);
vec_b = _mm_mul_ps(vec_b, mat2);
dest[3] = alpha;
vec_r = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b));
vec_r = _mm_max_ps(min, vec_r);
vec_r = _mm_min_ps(max, vec_r);
result = _mm_mul_ps(vec_r, scale);
*((__m64 *)&output[0]) = _mm_cvtps_pi32(result);
result = _mm_movehl_ps(result, result);
*((__m64 *)&output[2]) = _mm_cvtps_pi32(result);
dest[0] = otdata_r[output[0]];
dest[1] = otdata_g[output[1]];
dest[2] = otdata_b[output[2]];
_mm_empty();
}
|
