working on neon

module/neon/a8r8g8b8_to_yuvalp_box_neon.c

@@ -0,0 +1,124 @@
+/*
+Copyright 2024 Jay Sorg
+
+Permission to use, copy, modify, distribute, and sell this software and its
+documentation for any purpose is hereby granted without fee, provided that
+the above copyright notice appear in all copies and that both that
+copyright notice and this permission notice appear in supporting
+documentation.
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+OPEN GROUP BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+NEON SIMD a8r8g8b8_to_yuvalp
+
+*/
+
+#if defined(HAVE_CONFIG_H)
+#include <config_ac.h>
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <arm_neon.h>
+
+const uint32x4_t g_cd255 = { 255, 255, 255, 255 };
+const int16x8_t g_cw128 = { 128, 128, 128, 128, 128, 128, 128, 128 };
+const int16x8_t g_cw77  = {  77,  77,  77,  77,  77,  77,  77,  77 };
+const int16x8_t g_cw150 = { 150, 150, 150, 150, 150, 150, 150, 150 };
+const int16x8_t g_cw29  = {  29,  29,  29,  29,  29,  29,  29,  29 };
+const int16x8_t g_cw43  = {  43,  43,  43,  43,  43,  43,  43,  43 };
+const int16x8_t g_cw85  = {  85,  85,  85,  85,  85,  85,  85,  85 };
+const int16x8_t g_cw107 = { 107, 107, 107, 107, 107, 107, 107, 107 };
+const int16x8_t g_cw21  = {  21,  21,  21,  21,  21,  21,  21,  21 };
+
+/******************************************************************************/
+/* notes
+   address s8 should be aligned on 16 bytes, will be slower if not
+   width must be multiple of 8 and > 0
+   height must be > 0 */
+void
+a8r8g8b8_to_yuvalp_box_neon(const uint8_t *s8, int src_stride,
+                            uint8_t *d8, int dst_stride,
+                            int width, int height)
+{
+    int jndex;
+    int kndex;
+    uint32x4_t v1;
+    uint32x4_t blue1;
+    uint32x4_t blue2;
+    uint32x4_t green1;
+    uint32x4_t green2;
+    uint32x4_t red1;
+    uint32x4_t red2;
+    uint32x4_t alpha1;
+    uint32x4_t alpha2;
+    int16x8_t blue;
+    int16x8_t green;
+    int16x8_t red;
+    int16x8_t v2;
+    uint8x8_t out;
+
+    for (jndex = 0; jndex < height; jndex++)
+    {
+        for (kndex = 0; kndex < width; kndex += 8)
+        {
+            /* 4 pixels, 16 bytes */
+            v1 = vld1q_u32((const uint32_t *)(s8 + kndex * 4));
+            blue1 = vandq_u32(v1, g_cd255);
+            green1 = vandq_u32(vshrq_n_u32(v1, 8), g_cd255);
+            red1 = vandq_u32(vshrq_n_u32(v1, 16), g_cd255);
+            alpha1 = vandq_u32(vshrq_n_u32(v1, 24), g_cd255);
+            /* 4 pixels, 16 bytes */
+            v1 = vld1q_u32((const uint32_t *)(s8 + kndex * 4 + 16));
+            /* take care of alpha first */
+            alpha2 = vandq_u32(vshrq_n_u32(v1, 24), g_cd255);
+            out = vqmovn_u16(vcombine_u16(vqmovn_u32(alpha1),
+                                          vqmovn_u32(alpha2)));
+            vst1_u8(d8 + kndex * 8 + 3 * 64 * 64, out);
+            /* blue */
+            blue2 = vandq_u32(v1, g_cd255);
+            blue = (int16x8_t)vcombine_u16(vqmovn_u32(blue1),
+                                           vqmovn_u32(blue2));
+            /* green */
+            green2 = vandq_u32(vshrq_n_u32(v1, 8), g_cd255);
+            green = (int16x8_t)vcombine_u16(vqmovn_u32(green1),
+                                            vqmovn_u32(green2));
+            /* red */
+            red2 = vandq_u32(vshrq_n_u32(v1, 16), g_cd255);
+            red = (int16x8_t)vcombine_u16(vqmovn_u32(red1),
+                                          vqmovn_u32(red2));
+            /* _Y = (77 * _R + 150 * _G + 29 * _B) >> 8 */
+            v2 = vaddq_s16(vmulq_s16(blue, g_cw29),
+                           vmulq_s16(green, g_cw150));
+            v2 = vaddq_s16(v2, vmulq_s16(red, g_cw77));
+            out = vqmovn_u16((uint16x8_t)vshrq_n_s16(v2, 8));
+            vst1_u8(d8 + kndex * 8 + 0 * 64 * 64, out);
+            /* _U = ((-43 * _R - 85 * _G + 128 * _B) >> 8) + 128 */
+            v2 = vsubq_s16(vmulq_s16(blue, g_cw128),
+                           vmulq_s16(green, g_cw85));
+            v2 = vsubq_s16(v2, vmulq_s16(red, g_cw43));
+            v2 = vaddq_s16(vshrq_n_s16(v2, 8), g_cw128);
+            out = vqmovn_u16((uint16x8_t)v2);
+            vst1_u8(d8 + kndex * 8 + 1 * 64 * 64, out);
+            /* _V = ((128 * _R - 107 * _G -  21 * _B) >> 8) + 128 */
+            v2 = vsubq_s16(vmulq_s16(red, g_cw128),
+                           vmulq_s16(green, g_cw107));
+            v2 = vsubq_s16(v2, vmulq_s16(blue, g_cw21));
+            v2 = vaddq_s16(vshrq_n_s16(v2, 8), g_cw128);
+            out = vqmovn_u16((uint16x8_t)v2);
+            vst1_u8(d8 + kndex * 8 + 2 * 64 * 64, out);
+        }
+        s8 += src_stride;
+        d8 += dst_stride;
+    }
+}