~ Added pixman-arm-intrisics.cpp

gindemit · Jul 10, 2023 · 7c11dc1 · 7c11dc1
1 parent cd1cdfa
commit 7c11dc1
Show file tree

Hide file tree

Showing 2 changed files with 48 additions and 2 deletions.
diff --git a/dependency/pixman/pixman-arm-intrisics.cpp b/dependency/pixman/pixman-arm-intrisics.cpp
@@ -0,0 +1,46 @@
+#include <arm_neon.h>
+
+extern "C" void pixman_composite_src_n_8888_asm_neon(int32_t w, int32_t h,
+                                                     uint32_t *dst,
+                                                     int32_t   dst_stride,
+                                                     uint32_t  src)
+{
+    uint8x8_t v_src = vdup_n_u8(src);
+
+    for (int32_t y = 0; y < h; y++)
+    {
+        for (int32_t x = 0; x < w; x += 8)
+        {
+            vst1_u8((uint8_t *)(dst + x), v_src);
+        }
+        dst += dst_stride;
+    }
+}
+
+extern "C" void pixman_composite_over_n_8888_asm_neon(int32_t w, int32_t h,
+                                                      uint32_t *dst,
+                                                      int32_t   dst_stride,
+                                                      uint32_t  src)
+{
+    // Extract the source alpha and replicate it to all 8 lanes of a NEON vector
+    uint8x8_t v_src_alpha = vdup_n_u8(src >> 24);
+    // Extract the source color and replicate it to all 8 lanes of a NEON vector
+    uint8x8_t v_src_color = vdup_n_u8(src);
+
+    for (int32_t y = 0; y < h; y++)
+    {
+        for (int32_t x = 0; x < w; x += 8)
+        {
+            // Load 8 destination pixels
+            uint8x8_t v_dst_color = vld1_u8((uint8_t *)(dst + x));
+
+            // Calculate the result color = source color * source alpha + destination color * (1 - source alpha)
+            // Note that we need to shift right by 8 because the alpha blending operation can result in values greater than 255
+            uint8x8_t v_res_color = vshrq_n_u8(vmlaq_u8(vmlsq_u8(v_dst_color, v_dst_color, v_src_alpha), v_src_color, v_src_alpha), 8);
+
+            // Store the result to memory
+            vst1_u8((uint8_t *)(dst + x), v_res_color);
+        }
+        dst += dst_stride;
+    }
+}
diff --git a/projects/CMake/CMakeLists.txt b/projects/CMake/CMakeLists.txt
@@ -48,12 +48,12 @@ if (ANDROID)
       message("Compile asm source for armv7-a")
       target_compile_options(rlottie PUBLIC -fno-integrated-as)
       target_compile_definitions(rlottie PUBLIC USE_ARM_NEON)
-      target_sources(rlottie PRIVATE ${PIXMAN_ROOT}/pixman-arm-neon-asm.S)
+      target_sources(rlottie PRIVATE ${PIXMAN_ROOT}/pixman-arm-intrisics.cpp)
   elseif("${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "aarch64")
       message("Compile asm source for aarch64")
       target_compile_options(rlottie PUBLIC -fno-integrated-as)
       target_compile_definitions(rlottie PUBLIC USE_ARM_NEON __ARM64_NEON__)
-      target_sources(rlottie PRIVATE ${PIXMAN_ROOT}/pixman-arma64-neon-asm.S)
+      target_sources(rlottie PRIVATE ${PIXMAN_ROOT}/pixman-arm-intrisics.cpp)
   endif()
 endif()