diff --git a/simd/CMakeLists.txt b/simd/CMakeLists.txt
index 223251c..db1d966 100644
--- a/simd/CMakeLists.txt
+++ b/simd/CMakeLists.txt
@@ -272,7 +272,7 @@
 endif()
 if(NEON_INTRINSICS OR BITS EQUAL 32)
   set(SIMD_SOURCES ${SIMD_SOURCES} arm/aarch${BITS}/jchuff-neon.c
-    arm/jfdctint-neon.c)
+    arm/jdcolor-neon.c arm/jfdctint-neon.c)
 endif()
 if(BITS EQUAL 32)
   set_source_files_properties(${SIMD_SOURCES} COMPILE_FLAGS -mfpu=neon)
diff --git a/simd/arm/aarch32/jsimd_neon.S b/simd/arm/aarch32/jsimd_neon.S
index 4cd293d..cfebe1b 100644
--- a/simd/arm/aarch32/jsimd_neon.S
+++ b/simd/arm/aarch32/jsimd_neon.S
@@ -1270,14 +1270,9 @@
 /*****************************************************************************/
 
 /*
- * jsimd_ycc_extrgb_convert_neon
- * jsimd_ycc_extbgr_convert_neon
- * jsimd_ycc_extrgbx_convert_neon
- * jsimd_ycc_extbgrx_convert_neon
- * jsimd_ycc_extxbgr_convert_neon
- * jsimd_ycc_extxrgb_convert_neon
+ * jsimd_ycc_rgb565_convert_neon
  *
- * Colorspace conversion YCbCr -> RGB
+ * Colorspace conversion YCbCr -> RGB565
  */
 
 
@@ -1319,39 +1314,7 @@
 .endm
 
 .macro do_store bpp, size
-  .if \bpp == 24
-    .if \size == 8
-      vst3.8        {d10, d11, d12}, [RGB]!
-    .elseif \size == 4
-      vst3.8        {d10[0], d11[0], d12[0]}, [RGB]!
-      vst3.8        {d10[1], d11[1], d12[1]}, [RGB]!
-      vst3.8        {d10[2], d11[2], d12[2]}, [RGB]!
-      vst3.8        {d10[3], d11[3], d12[3]}, [RGB]!
-    .elseif \size == 2
-      vst3.8        {d10[4], d11[4], d12[4]}, [RGB]!
-      vst3.8        {d10[5], d11[5], d12[5]}, [RGB]!
-    .elseif \size == 1
-      vst3.8        {d10[6], d11[6], d12[6]}, [RGB]!
-    .else
-      .error unsupported macroblock size
-    .endif
-  .elseif \bpp == 32
-    .if \size == 8
-      vst4.8        {d10, d11, d12, d13}, [RGB]!
-    .elseif \size == 4
-      vst4.8        {d10[0], d11[0], d12[0], d13[0]}, [RGB]!
-      vst4.8        {d10[1], d11[1], d12[1], d13[1]}, [RGB]!
-      vst4.8        {d10[2], d11[2], d12[2], d13[2]}, [RGB]!
-      vst4.8        {d10[3], d11[3], d12[3], d13[3]}, [RGB]!
-    .elseif \size == 2
-      vst4.8        {d10[4], d11[4], d12[4], d13[4]}, [RGB]!
-      vst4.8        {d10[5], d11[5], d12[5], d13[5]}, [RGB]!
-    .elseif \size == 1
-      vst4.8        {d10[6], d11[6], d12[6], d13[6]}, [RGB]!
-    .else
-      .error unsupported macroblock size
-    .endif
-  .elseif \bpp == 16
+  .if \bpp == 16
     .if \size == 8
       vst1.16       {q15}, [RGB]!
     .elseif \size == 4
@@ -1398,17 +1361,11 @@
     vaddw.u8        q11, q10, d0
     vaddw.u8        q12, q12, d0
     vaddw.u8        q14, q14, d0
-  .if \bpp != 16
-    vqmovun.s16     d1\g_offs, q11
-    vqmovun.s16     d1\r_offs, q12
-    vqmovun.s16     d1\b_offs, q14
-  .else  /* rgb565 */
     vqshlu.s16      q13, q11, #8
     vqshlu.s16      q15, q12, #8
     vqshlu.s16      q14, q14, #8
     vsri.u16        q15, q13, #5
     vsri.u16        q15, q14, #11
-  .endif
 .endm
 
 .macro do_yuv_to_rgb_stage2_store_load_stage1
@@ -1431,20 +1388,6 @@
     vmlal.s16       q10, d8, d1[2]  /* multiply by -23401 */
                                        vaddw.u8        q12, q12, d0
                                        vaddw.u8        q14, q14, d0
-  .if \bpp != 16  /**************** rgb24/rgb32 ******************************/
-                                       vqmovun.s16     d1\g_offs, q11
-    pld             [Y, #64]
-                                       vqmovun.s16     d1\r_offs, q12
-    vld1.8          {d0}, [Y, :64]!
-                                       vqmovun.s16     d1\b_offs, q14
-    vmull.s16       q11, d7, d1[1]  /* multiply by -11277 */
-    vmlal.s16       q11, d9, d1[2]  /* multiply by -23401 */
-                                       do_store        \bpp, 8
-    vmull.s16       q12, d8, d1[0]  /* multiply by 22971 */
-    vmull.s16       q13, d9, d1[0]  /* multiply by 22971 */
-    vmull.s16       q14, d6, d1[3]  /* multiply by 29033 */
-    vmull.s16       q15, d7, d1[3]  /* multiply by 29033 */
-  .else  /**************************** rgb565 ********************************/
                                        vqshlu.s16      q13, q11, #8
     pld             [Y, #64]
                                        vqshlu.s16      q15, q12, #8
@@ -1459,7 +1402,6 @@
     vmull.s16       q14, d6, d1[3]
                                        do_store        \bpp, 8
     vmull.s16       q15, d7, d1[3]
-  .endif
 .endm
 
 .macro do_yuv_to_rgb
@@ -1595,12 +1537,6 @@
 .endm
 
 /*--------------------------------- id ----- bpp R  G  B */
-generate_jsimd_ycc_rgb_convert_neon extrgb,  24, 0, 1, 2
-generate_jsimd_ycc_rgb_convert_neon extbgr,  24, 2, 1, 0
-generate_jsimd_ycc_rgb_convert_neon extrgbx, 32, 0, 1, 2
-generate_jsimd_ycc_rgb_convert_neon extbgrx, 32, 2, 1, 0
-generate_jsimd_ycc_rgb_convert_neon extxbgr, 32, 3, 2, 1
-generate_jsimd_ycc_rgb_convert_neon extxrgb, 32, 1, 2, 3
 generate_jsimd_ycc_rgb_convert_neon rgb565,  16, 0, 0, 0
 
 .purgem do_load
diff --git a/simd/arm/aarch64/jsimd.c b/simd/arm/aarch64/jsimd.c
index 9a1196b..b588d8b 100644
--- a/simd/arm/aarch64/jsimd.c
+++ b/simd/arm/aarch64/jsimd.c
@@ -349,20 +349,28 @@
 
   switch (cinfo->out_color_space) {
   case JCS_EXT_RGB:
+#ifndef NEON_INTRINSICS
     if (simd_features & JSIMD_FASTST3)
+#endif
       neonfct = jsimd_ycc_extrgb_convert_neon;
+#ifndef NEON_INTRINSICS
     else
       neonfct = jsimd_ycc_extrgb_convert_neon_slowst3;
+#endif
     break;
   case JCS_EXT_RGBX:
   case JCS_EXT_RGBA:
     neonfct = jsimd_ycc_extrgbx_convert_neon;
     break;
   case JCS_EXT_BGR:
+#ifndef NEON_INTRINSICS
     if (simd_features & JSIMD_FASTST3)
+#endif
       neonfct = jsimd_ycc_extbgr_convert_neon;
+#ifndef NEON_INTRINSICS
     else
       neonfct = jsimd_ycc_extbgr_convert_neon_slowst3;
+#endif
     break;
   case JCS_EXT_BGRX:
   case JCS_EXT_BGRA:
@@ -377,10 +385,14 @@
     neonfct = jsimd_ycc_extxrgb_convert_neon;
     break;
   default:
+#ifndef NEON_INTRINSICS
     if (simd_features & JSIMD_FASTST3)
+#endif
       neonfct = jsimd_ycc_extrgb_convert_neon;
+#ifndef NEON_INTRINSICS
     else
       neonfct = jsimd_ycc_extrgb_convert_neon_slowst3;
+#endif
     break;
   }
 
diff --git a/simd/arm/aarch64/jsimd_neon.S b/simd/arm/aarch64/jsimd_neon.S
index 9af7062..dd97aae 100644
--- a/simd/arm/aarch64/jsimd_neon.S
+++ b/simd/arm/aarch64/jsimd_neon.S
@@ -1929,16 +1929,20 @@
 .endm
 
 /*--------------------------------- id ----- bpp R  rsize G  gsize B  bsize defsize fast_st3*/
+#ifndef NEON_INTRINSICS
 generate_jsimd_ycc_rgb_convert_neon extrgb,  24, 0, .4h,  1, .4h,  2, .4h,  .8b,    1
 generate_jsimd_ycc_rgb_convert_neon extbgr,  24, 2, .4h,  1, .4h,  0, .4h,  .8b,    1
 generate_jsimd_ycc_rgb_convert_neon extrgbx, 32, 0, .4h,  1, .4h,  2, .4h,  .8b,    1
 generate_jsimd_ycc_rgb_convert_neon extbgrx, 32, 2, .4h,  1, .4h,  0, .4h,  .8b,    1
 generate_jsimd_ycc_rgb_convert_neon extxbgr, 32, 3, .4h,  2, .4h,  1, .4h,  .8b,    1
 generate_jsimd_ycc_rgb_convert_neon extxrgb, 32, 1, .4h,  2, .4h,  3, .4h,  .8b,    1
+#endif
 generate_jsimd_ycc_rgb_convert_neon rgb565,  16, 0, .4h,  0, .4h,  0, .4h,  .8b,    1
 
+#ifndef NEON_INTRINSICS
 generate_jsimd_ycc_rgb_convert_neon extrgb,  24, 0, .4h,  1, .4h,  2, .4h,  .8b,    0
 generate_jsimd_ycc_rgb_convert_neon extbgr,  24, 2, .4h,  1, .4h,  0, .4h,  .8b,    0
+#endif
 
 .purgem do_load
 .purgem do_store
diff --git a/simd/arm/jdcolext-neon.c b/simd/arm/jdcolext-neon.c
new file mode 100644
index 0000000..5e4baa9
--- /dev/null
+++ b/simd/arm/jdcolext-neon.c
@@ -0,0 +1,309 @@
+/*
+ * jdcolext-neon.c - colorspace conversion (Arm Neon)
+ *
+ * Copyright (C) 2020, Arm Limited.  All Rights Reserved.
+ * Copyright (C) 2020, D. R. Commander.  All Rights Reserved.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* This file is included by jdcolor-neon.c. */
+
+
+/* YCbCr -> RGB conversion is defined by the following equations:
+ *    R = Y                        + 1.40200 * (Cr - 128)
+ *    G = Y - 0.34414 * (Cb - 128) - 0.71414 * (Cr - 128)
+ *    B = Y + 1.77200 * (Cb - 128)
+ *
+ * Scaled integer constants are used to avoid floating-point arithmetic:
+ *    0.3441467 = 11277 * 2^-15
+ *    0.7141418 = 23401 * 2^-15
+ *    1.4020386 = 22971 * 2^-14
+ *    1.7720337 = 29033 * 2^-14
+ * These constants are defined in jdcolor-neon.c.
+ *
+ * To ensure correct results, rounding is used when descaling.
+ */
+
+/* Notes on safe memory access for YCbCr -> RGB conversion routines:
+ *
+ * Input memory buffers can be safely overread up to the next multiple of
+ * ALIGN_SIZE bytes, since they are always allocated by alloc_sarray() in
+ * jmemmgr.c.
+ *
+ * The output buffer cannot safely be written beyond output_width, since
+ * output_buf points to a possibly unpadded row in the decompressed image
+ * buffer allocated by the calling program.
+ */
+
+void jsimd_ycc_rgb_convert_neon(JDIMENSION output_width, JSAMPIMAGE input_buf,
+                                JDIMENSION input_row, JSAMPARRAY output_buf,
+                                int num_rows)
+{
+  JSAMPROW outptr;
+  /* Pointers to Y, Cb, and Cr data */
+  JSAMPROW inptr0, inptr1, inptr2;
+
+  const int16x4_t consts = vld1_s16(jsimd_ycc_rgb_convert_neon_consts);
+  const int16x8_t neg_128 = vdupq_n_s16(-128);
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    input_row++;
+    outptr = *output_buf++;
+    int cols_remaining = output_width;
+    for (; cols_remaining >= 16; cols_remaining -= 16) {
+      uint8x16_t y  = vld1q_u8(inptr0);
+      uint8x16_t cb = vld1q_u8(inptr1);
+      uint8x16_t cr = vld1q_u8(inptr2);
+      /* Subtract 128 from Cb and Cr. */
+      int16x8_t cr_128_l =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128),
+                                       vget_low_u8(cr)));
+      int16x8_t cr_128_h =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128),
+                                       vget_high_u8(cr)));
+      int16x8_t cb_128_l =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128),
+                                       vget_low_u8(cb)));
+      int16x8_t cb_128_h =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128),
+                                       vget_high_u8(cb)));
+      /* Compute G-Y: - 0.34414 * (Cb - 128) - 0.71414 * (Cr - 128) */
+      int32x4_t g_sub_y_ll = vmull_lane_s16(vget_low_s16(cb_128_l), consts, 0);
+      int32x4_t g_sub_y_lh = vmull_lane_s16(vget_high_s16(cb_128_l),
+                                            consts, 0);
+      int32x4_t g_sub_y_hl = vmull_lane_s16(vget_low_s16(cb_128_h), consts, 0);
+      int32x4_t g_sub_y_hh = vmull_lane_s16(vget_high_s16(cb_128_h),
+                                            consts, 0);
+      g_sub_y_ll = vmlsl_lane_s16(g_sub_y_ll, vget_low_s16(cr_128_l),
+                                  consts, 1);
+      g_sub_y_lh = vmlsl_lane_s16(g_sub_y_lh, vget_high_s16(cr_128_l),
+                                  consts, 1);
+      g_sub_y_hl = vmlsl_lane_s16(g_sub_y_hl, vget_low_s16(cr_128_h),
+                                  consts, 1);
+      g_sub_y_hh = vmlsl_lane_s16(g_sub_y_hh, vget_high_s16(cr_128_h),
+                                  consts, 1);
+      /* Descale G components: shift right 15, round, and narrow to 16-bit. */
+      int16x8_t g_sub_y_l = vcombine_s16(vrshrn_n_s32(g_sub_y_ll, 15),
+                                         vrshrn_n_s32(g_sub_y_lh, 15));
+      int16x8_t g_sub_y_h = vcombine_s16(vrshrn_n_s32(g_sub_y_hl, 15),
+                                         vrshrn_n_s32(g_sub_y_hh, 15));
+      /* Compute R-Y: 1.40200 * (Cr - 128) */
+      int16x8_t r_sub_y_l = vqrdmulhq_lane_s16(vshlq_n_s16(cr_128_l, 1),
+                                               consts, 2);
+      int16x8_t r_sub_y_h = vqrdmulhq_lane_s16(vshlq_n_s16(cr_128_h, 1),
+                                               consts, 2);
+      /* Compute B-Y: 1.77200 * (Cb - 128) */
+      int16x8_t b_sub_y_l = vqrdmulhq_lane_s16(vshlq_n_s16(cb_128_l, 1),
+                                               consts, 3);
+      int16x8_t b_sub_y_h = vqrdmulhq_lane_s16(vshlq_n_s16(cb_128_h, 1),
+                                               consts, 3);
+      /* Add Y. */
+      int16x8_t r_l =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(r_sub_y_l),
+                                       vget_low_u8(y)));
+      int16x8_t r_h =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(r_sub_y_h),
+                                       vget_high_u8(y)));
+      int16x8_t b_l =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(b_sub_y_l),
+                                       vget_low_u8(y)));
+      int16x8_t b_h =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(b_sub_y_h),
+                                       vget_high_u8(y)));
+      int16x8_t g_l =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(g_sub_y_l),
+                                       vget_low_u8(y)));
+      int16x8_t g_h =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(g_sub_y_h),
+                                       vget_high_u8(y)));
+
+#ifdef RGB_ALPHA
+      uint8x16x4_t rgba;
+      /* Convert each component to unsigned and narrow, clamping to [0-255]. */
+      rgba.val[RGB_RED] = vcombine_u8(vqmovun_s16(r_l), vqmovun_s16(r_h));
+      rgba.val[RGB_GREEN] = vcombine_u8(vqmovun_s16(g_l), vqmovun_s16(g_h));
+      rgba.val[RGB_BLUE] = vcombine_u8(vqmovun_s16(b_l), vqmovun_s16(b_h));
+      /* Set alpha channel to opaque (0xFF). */
+      rgba.val[RGB_ALPHA] = vdupq_n_u8(0xFF);
+      /* Store RGBA pixel data to memory. */
+      vst4q_u8(outptr, rgba);
+#else
+      uint8x16x3_t rgb;
+      /* Convert each component to unsigned and narrow, clamping to [0-255]. */
+      rgb.val[RGB_RED] = vcombine_u8(vqmovun_s16(r_l), vqmovun_s16(r_h));
+      rgb.val[RGB_GREEN] = vcombine_u8(vqmovun_s16(g_l), vqmovun_s16(g_h));
+      rgb.val[RGB_BLUE] = vcombine_u8(vqmovun_s16(b_l), vqmovun_s16(b_h));
+      /* Store RGB pixel data to memory. */
+      vst3q_u8(outptr, rgb);
+#endif
+      /* Increment pointers. */
+      inptr0 += 16;
+      inptr1 += 16;
+      inptr2 += 16;
+      outptr += (RGB_PIXELSIZE * 16);
+    }
+
+    if (cols_remaining >= 8) {
+      uint8x8_t y  = vld1_u8(inptr0);
+      uint8x8_t cb = vld1_u8(inptr1);
+      uint8x8_t cr = vld1_u8(inptr2);
+      /* Subtract 128 from Cb and Cr. */
+      int16x8_t cr_128 =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128), cr));
+      int16x8_t cb_128 =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128), cb));
+      /* Compute G-Y: - 0.34414 * (Cb - 128) - 0.71414 * (Cr - 128) */
+      int32x4_t g_sub_y_l = vmull_lane_s16(vget_low_s16(cb_128), consts, 0);
+      int32x4_t g_sub_y_h = vmull_lane_s16(vget_high_s16(cb_128), consts, 0);
+      g_sub_y_l = vmlsl_lane_s16(g_sub_y_l, vget_low_s16(cr_128), consts, 1);
+      g_sub_y_h = vmlsl_lane_s16(g_sub_y_h, vget_high_s16(cr_128), consts, 1);
+      /* Descale G components: shift right 15, round, and narrow to 16-bit. */
+      int16x8_t g_sub_y = vcombine_s16(vrshrn_n_s32(g_sub_y_l, 15),
+                                       vrshrn_n_s32(g_sub_y_h, 15));
+      /* Compute R-Y: 1.40200 * (Cr - 128) */
+      int16x8_t r_sub_y = vqrdmulhq_lane_s16(vshlq_n_s16(cr_128, 1),
+                                             consts, 2);
+      /* Compute B-Y: 1.77200 * (Cb - 128) */
+      int16x8_t b_sub_y = vqrdmulhq_lane_s16(vshlq_n_s16(cb_128, 1),
+                                             consts, 3);
+      /* Add Y. */
+      int16x8_t r =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(r_sub_y), y));
+      int16x8_t b =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(b_sub_y), y));
+      int16x8_t g =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(g_sub_y), y));
+
+#ifdef RGB_ALPHA
+      uint8x8x4_t rgba;
+      /* Convert each component to unsigned and narrow, clamping to [0-255]. */
+      rgba.val[RGB_RED] = vqmovun_s16(r);
+      rgba.val[RGB_GREEN] = vqmovun_s16(g);
+      rgba.val[RGB_BLUE] = vqmovun_s16(b);
+      /* Set alpha channel to opaque (0xFF). */
+      rgba.val[RGB_ALPHA] = vdup_n_u8(0xFF);
+      /* Store RGBA pixel data to memory. */
+      vst4_u8(outptr, rgba);
+#else
+      uint8x8x3_t rgb;
+      /* Convert each component to unsigned and narrow, clamping to [0-255]. */
+      rgb.val[RGB_RED] = vqmovun_s16(r);
+      rgb.val[RGB_GREEN] = vqmovun_s16(g);
+      rgb.val[RGB_BLUE] = vqmovun_s16(b);
+      /* Store RGB pixel data to memory. */
+      vst3_u8(outptr, rgb);
+#endif
+      /* Increment pointers. */
+      inptr0 += 8;
+      inptr1 += 8;
+      inptr2 += 8;
+      outptr += (RGB_PIXELSIZE * 8);
+      cols_remaining -= 8;
+    }
+
+    /* Handle the tail elements. */
+    if (cols_remaining > 0) {
+      uint8x8_t y  = vld1_u8(inptr0);
+      uint8x8_t cb = vld1_u8(inptr1);
+      uint8x8_t cr = vld1_u8(inptr2);
+      /* Subtract 128 from Cb and Cr. */
+      int16x8_t cr_128 =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128), cr));
+      int16x8_t cb_128 =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(neg_128), cb));
+      /* Compute G-Y: - 0.34414 * (Cb - 128) - 0.71414 * (Cr - 128) */
+      int32x4_t g_sub_y_l = vmull_lane_s16(vget_low_s16(cb_128), consts, 0);
+      int32x4_t g_sub_y_h = vmull_lane_s16(vget_high_s16(cb_128), consts, 0);
+      g_sub_y_l = vmlsl_lane_s16(g_sub_y_l, vget_low_s16(cr_128), consts, 1);
+      g_sub_y_h = vmlsl_lane_s16(g_sub_y_h, vget_high_s16(cr_128), consts, 1);
+      /* Descale G components: shift right 15, round, and narrow to 16-bit. */
+      int16x8_t g_sub_y = vcombine_s16(vrshrn_n_s32(g_sub_y_l, 15),
+                                       vrshrn_n_s32(g_sub_y_h, 15));
+      /* Compute R-Y: 1.40200 * (Cr - 128) */
+      int16x8_t r_sub_y = vqrdmulhq_lane_s16(vshlq_n_s16(cr_128, 1),
+                                             consts, 2);
+      /* Compute B-Y: 1.77200 * (Cb - 128) */
+      int16x8_t b_sub_y = vqrdmulhq_lane_s16(vshlq_n_s16(cb_128, 1),
+                                             consts, 3);
+      /* Add Y. */
+      int16x8_t r =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(r_sub_y), y));
+      int16x8_t b =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(b_sub_y), y));
+      int16x8_t g =
+        vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(g_sub_y), y));
+
+#ifdef RGB_ALPHA
+      uint8x8x4_t rgba;
+      /* Convert each component to unsigned and narrow, clamping to [0-255]. */
+      rgba.val[RGB_RED] = vqmovun_s16(r);
+      rgba.val[RGB_GREEN] = vqmovun_s16(g);
+      rgba.val[RGB_BLUE] = vqmovun_s16(b);
+      /* Set alpha channel to opaque (0xFF). */
+      rgba.val[RGB_ALPHA] = vdup_n_u8(0xFF);
+      /* Store RGBA pixel data to memory. */
+      switch (cols_remaining) {
+      case 7:
+        vst4_lane_u8(outptr + 6 * RGB_PIXELSIZE, rgba, 6);
+      case 6:
+        vst4_lane_u8(outptr + 5 * RGB_PIXELSIZE, rgba, 5);
+      case 5:
+        vst4_lane_u8(outptr + 4 * RGB_PIXELSIZE, rgba, 4);
+      case 4:
+        vst4_lane_u8(outptr + 3 * RGB_PIXELSIZE, rgba, 3);
+      case 3:
+        vst4_lane_u8(outptr + 2 * RGB_PIXELSIZE, rgba, 2);
+      case 2:
+        vst4_lane_u8(outptr + RGB_PIXELSIZE, rgba, 1);
+      case 1:
+        vst4_lane_u8(outptr, rgba, 0);
+      default:
+        break;
+      }
+#else
+      uint8x8x3_t rgb;
+      /* Convert each component to unsigned and narrow, clamping to [0-255]. */
+      rgb.val[RGB_RED] = vqmovun_s16(r);
+      rgb.val[RGB_GREEN] = vqmovun_s16(g);
+      rgb.val[RGB_BLUE] = vqmovun_s16(b);
+      /* Store RGB pixel data to memory. */
+      switch (cols_remaining) {
+      case 7:
+        vst3_lane_u8(outptr + 6 * RGB_PIXELSIZE, rgb, 6);
+      case 6:
+        vst3_lane_u8(outptr + 5 * RGB_PIXELSIZE, rgb, 5);
+      case 5:
+        vst3_lane_u8(outptr + 4 * RGB_PIXELSIZE, rgb, 4);
+      case 4:
+        vst3_lane_u8(outptr + 3 * RGB_PIXELSIZE, rgb, 3);
+      case 3:
+        vst3_lane_u8(outptr + 2 * RGB_PIXELSIZE, rgb, 2);
+      case 2:
+        vst3_lane_u8(outptr + RGB_PIXELSIZE, rgb, 1);
+      case 1:
+        vst3_lane_u8(outptr, rgb, 0);
+      default:
+        break;
+      }
+#endif
+    }
+  }
+}
diff --git a/simd/arm/jdcolor-neon.c b/simd/arm/jdcolor-neon.c
new file mode 100644
index 0000000..abbef8c
--- /dev/null
+++ b/simd/arm/jdcolor-neon.c
@@ -0,0 +1,133 @@
+/*
+ * jdcolor-neon.c - colorspace conversion (Arm Neon)
+ *
+ * Copyright (C) 2020, Arm Limited.  All Rights Reserved.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+#define JPEG_INTERNALS
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
+#include "../jsimd.h"
+#include "align.h"
+
+#include <arm_neon.h>
+
+
+/* YCbCr -> RGB conversion constants */
+
+#define F_0_344  11277  /* 0.3441467 = 11277 * 2^-15 */
+#define F_0_714  23401  /* 0.7141418 = 23401 * 2^-15 */
+#define F_1_402  22971  /* 1.4020386 = 22971 * 2^-14 */
+#define F_1_772  29033  /* 1.7720337 = 29033 * 2^-14 */
+
+ALIGN(16) static const int16_t jsimd_ycc_rgb_convert_neon_consts[] = {
+  -F_0_344, F_0_714, F_1_402, F_1_772
+};
+
+
+/* Include inline routines for colorspace extensions. */
+
+#include "jdcolext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+
+#define RGB_RED  EXT_RGB_RED
+#define RGB_GREEN  EXT_RGB_GREEN
+#define RGB_BLUE  EXT_RGB_BLUE
+#define RGB_PIXELSIZE  EXT_RGB_PIXELSIZE
+#define jsimd_ycc_rgb_convert_neon  jsimd_ycc_extrgb_convert_neon
+#include "jdcolext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_neon
+
+#define RGB_RED  EXT_RGBX_RED
+#define RGB_GREEN  EXT_RGBX_GREEN
+#define RGB_BLUE  EXT_RGBX_BLUE
+#define RGB_ALPHA  3
+#define RGB_PIXELSIZE  EXT_RGBX_PIXELSIZE
+#define jsimd_ycc_rgb_convert_neon  jsimd_ycc_extrgbx_convert_neon
+#include "jdcolext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_ALPHA
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_neon
+
+#define RGB_RED  EXT_BGR_RED
+#define RGB_GREEN  EXT_BGR_GREEN
+#define RGB_BLUE  EXT_BGR_BLUE
+#define RGB_PIXELSIZE  EXT_BGR_PIXELSIZE
+#define jsimd_ycc_rgb_convert_neon  jsimd_ycc_extbgr_convert_neon
+#include "jdcolext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_neon
+
+#define RGB_RED  EXT_BGRX_RED
+#define RGB_GREEN  EXT_BGRX_GREEN
+#define RGB_BLUE  EXT_BGRX_BLUE
+#define RGB_ALPHA  3
+#define RGB_PIXELSIZE  EXT_BGRX_PIXELSIZE
+#define jsimd_ycc_rgb_convert_neon  jsimd_ycc_extbgrx_convert_neon
+#include "jdcolext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_ALPHA
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_neon
+
+#define RGB_RED  EXT_XBGR_RED
+#define RGB_GREEN  EXT_XBGR_GREEN
+#define RGB_BLUE  EXT_XBGR_BLUE
+#define RGB_ALPHA  0
+#define RGB_PIXELSIZE  EXT_XBGR_PIXELSIZE
+#define jsimd_ycc_rgb_convert_neon  jsimd_ycc_extxbgr_convert_neon
+#include "jdcolext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_ALPHA
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_neon
+
+#define RGB_RED  EXT_XRGB_RED
+#define RGB_GREEN  EXT_XRGB_GREEN
+#define RGB_BLUE  EXT_XRGB_BLUE
+#define RGB_ALPHA  0
+#define RGB_PIXELSIZE  EXT_XRGB_PIXELSIZE
+#define jsimd_ycc_rgb_convert_neon  jsimd_ycc_extxrgb_convert_neon
+#include "jdcolext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_ALPHA
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_neon
diff --git a/simd/jsimd.h b/simd/jsimd.h
index 40d4721..a2758b4 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -450,6 +450,8 @@
   (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
    JSAMPARRAY output_buf, int num_rows);
 
+#ifndef NEON_INTRINSICS
+
 EXTERN(void) jsimd_ycc_extrgb_convert_neon_slowst3
   (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
    JSAMPARRAY output_buf, int num_rows);
@@ -457,6 +459,8 @@
   (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
    JSAMPARRAY output_buf, int num_rows);
 
+#endif
+
 EXTERN(void) jsimd_ycc_rgb_convert_dspr2
   (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
    JSAMPARRAY output_buf, int num_rows);