2 * This file is part of FFmpeg.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
19 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
20 * DEALINGS IN THE SOFTWARE.
23 #include "cuda/vector_helpers.cuh"
25 typedef float4 (*coeffs_function_t)(float);
27 __device__ inline float4 lanczos_coeffs(float x)
29 const float pi = 3.141592654f;
31 float4 res = make_float4(
37 res.x = res.x == 0.0f ? 1.0f :
38 __sinf(res.x) * __sinf(res.x / 2.0f) / (res.x * res.x / 2.0f);
39 res.y = res.y == 0.0f ? 1.0f :
40 __sinf(res.y) * __sinf(res.y / 2.0f) / (res.y * res.y / 2.0f);
41 res.z = res.z == 0.0f ? 1.0f :
42 __sinf(res.z) * __sinf(res.z / 2.0f) / (res.z * res.z / 2.0f);
43 res.w = res.w == 0.0f ? 1.0f :
44 __sinf(res.w) * __sinf(res.w / 2.0f) / (res.w * res.w / 2.0f);
46 return res / (res.x + res.y + res.z + res.w);
49 __device__ inline float4 bicubic_coeffs(float x)
51 const float A = -0.75f;
54 res.x = ((A * (x + 1) - 5 * A) * (x + 1) + 8 * A) * (x + 1) - 4 * A;
55 res.y = ((A + 2) * x - (A + 3)) * x * x + 1;
56 res.z = ((A + 2) * (1 - x) - (A + 3)) * (1 - x) * (1 - x) + 1;
57 res.w = 1.0f - res.x - res.y - res.z;
62 __device__ inline void derived_fast_coeffs(float4 coeffs, float x, float *h0, float *h1, float *s)
64 float g0 = coeffs.x + coeffs.y;
65 float g1 = coeffs.z + coeffs.w;
67 *h0 = coeffs.y / g0 - 0.5f;
68 *h1 = coeffs.w / g1 + 1.5f;
73 __device__ inline V apply_coeffs(float4 coeffs, V c0, V c1, V c2, V c3)
75 V res = c0 * coeffs.x;
84 __device__ inline void Subsample_Bicubic(coeffs_function_t coeffs_function,
85 cudaTextureObject_t src_tex,
87 int dst_width, int dst_height, int dst_pitch,
88 int src_width, int src_height,
91 int xo = blockIdx.x * blockDim.x + threadIdx.x;
92 int yo = blockIdx.y * blockDim.y + threadIdx.y;
94 if (yo < dst_height && xo < dst_width)
96 float hscale = (float)src_width / (float)dst_width;
97 float vscale = (float)src_height / (float)dst_height;
98 float xi = (xo + 0.5f) * hscale - 0.5f;
99 float yi = (yo + 0.5f) * vscale - 0.5f;
100 float px = floor(xi);
101 float py = floor(yi);
105 float factor = bit_depth > 8 ? 0xFFFF : 0xFF;
107 float4 coeffsX = coeffs_function(fx);
108 float4 coeffsY = coeffs_function(fy);
110 #define PIX(x, y) tex2D<floatT>(src_tex, (x), (y))
112 dst[yo * dst_pitch + xo] = from_floatN<T, floatT>(
113 apply_coeffs<floatT>(coeffsY,
114 apply_coeffs<floatT>(coeffsX, PIX(px - 1, py - 1), PIX(px, py - 1), PIX(px + 1, py - 1), PIX(px + 2, py - 1)),
115 apply_coeffs<floatT>(coeffsX, PIX(px - 1, py ), PIX(px, py ), PIX(px + 1, py ), PIX(px + 2, py )),
116 apply_coeffs<floatT>(coeffsX, PIX(px - 1, py + 1), PIX(px, py + 1), PIX(px + 1, py + 1), PIX(px + 2, py + 1)),
117 apply_coeffs<floatT>(coeffsX, PIX(px - 1, py + 2), PIX(px, py + 2), PIX(px + 1, py + 2), PIX(px + 2, py + 2))
125 /* This does not yield correct results. Most likely because of low internal precision in tex2D linear interpolation */
127 __device__ inline void Subsample_FastBicubic(coeffs_function_t coeffs_function,
128 cudaTextureObject_t src_tex,
130 int dst_width, int dst_height, int dst_pitch,
131 int src_width, int src_height,
134 int xo = blockIdx.x * blockDim.x + threadIdx.x;
135 int yo = blockIdx.y * blockDim.y + threadIdx.y;
137 if (yo < dst_height && xo < dst_width)
139 float hscale = (float)src_width / (float)dst_width;
140 float vscale = (float)src_height / (float)dst_height;
141 float xi = (xo + 0.5f) * hscale - 0.5f;
142 float yi = (yo + 0.5f) * vscale - 0.5f;
143 float px = floor(xi);
144 float py = floor(yi);
148 float factor = bit_depth > 8 ? 0xFFFF : 0xFF;
150 float4 coeffsX = coeffs_function(fx);
151 float4 coeffsY = coeffs_function(fy);
155 derived_fast_coeffs(coeffsX, fx, &h0x, &h1x, &sx);
156 derived_fast_coeffs(coeffsY, fy, &h0y, &h1y, &sy);
158 #define PIX(x, y) tex2D<floatT>(src_tex, (x), (y))
161 PIX(px + h0x, py + h0y),
162 PIX(px + h1x, py + h0y),
163 PIX(px + h0x, py + h1y),
164 PIX(px + h1x, py + h1y)
169 dst[yo * dst_pitch + xo] = from_floatN<T, floatT>(
171 lerp_scalar(pix[3], pix[2], sx),
172 lerp_scalar(pix[1], pix[0], sx),
180 #define BICUBIC_KERNEL(T) \
181 __global__ void Subsample_Bicubic_ ## T(cudaTextureObject_t src_tex, \
183 int dst_width, int dst_height, int dst_pitch, \
184 int src_width, int src_height, \
187 Subsample_Bicubic<T>(&bicubic_coeffs, src_tex, dst, \
188 dst_width, dst_height, dst_pitch, \
189 src_width, src_height, \
193 BICUBIC_KERNEL(uchar)
194 BICUBIC_KERNEL(uchar2)
195 BICUBIC_KERNEL(uchar4)
197 BICUBIC_KERNEL(ushort)
198 BICUBIC_KERNEL(ushort2)
199 BICUBIC_KERNEL(ushort4)
202 #define LANCZOS_KERNEL(T) \
203 __global__ void Subsample_Lanczos_ ## T(cudaTextureObject_t src_tex, \
205 int dst_width, int dst_height, int dst_pitch, \
206 int src_width, int src_height, \
209 Subsample_Bicubic<T>(&lanczos_coeffs, src_tex, dst, \
210 dst_width, dst_height, dst_pitch, \
211 src_width, src_height, \
215 LANCZOS_KERNEL(uchar)
216 LANCZOS_KERNEL(uchar2)
217 LANCZOS_KERNEL(uchar4)
219 LANCZOS_KERNEL(ushort)
220 LANCZOS_KERNEL(ushort2)
221 LANCZOS_KERNEL(ushort4)