diff --git a/lib/src/dsp/arm/gainmapmath_neon.cpp b/lib/src/dsp/arm/gainmapmath_neon.cpp
index 68a84940..dc419165 100644
--- a/lib/src/dsp/arm/gainmapmath_neon.cpp
+++ b/lib/src/dsp/arm/gainmapmath_neon.cpp
@@ -324,12 +324,12 @@ uhdr_error_info_t convertYuv_neon(uhdr_raw_image_t* image, uhdr_color_gamut_t sr
 ALIGNED(16)
 const uint16_t kRgb709ToYuv_coeffs_neon[8] = {3484, 11717, 1183, 1877, 6315, 8192, 7441, 751};
 
-// RGB Display P3 -> Yuv Display P3
-// Y = 0.2289746 * R + 0.6917385 * G + 0.0792869 * B
-// U = -0.124346335 * R + -0.375653665 * G + 0.5 * B
-// V = 0.5 * R + -0.448583471 * G + -0.051416529 * B
+// RGB Display P3 -> YUV using BT.601 luma coefficients, matching p3RgbToYuv().
+// Y = 0.299 * R + 0.587 * G + 0.114 * B
+// U = -0.168735892 * R + -0.331264108 * G + 0.5 * B
+// V = 0.5 * R + -0.418687589 * G + -0.081312411 * B
 ALIGNED(16)
-const uint16_t kRgbDispP3ToYuv_coeffs_neon[8] = {3752, 11333, 1299, 2037, 6155, 8192, 7350, 842};
+const uint16_t kRgb601ToYuv_coeffs_neon[8] = {4899, 9617, 1868, 2765, 5427, 8192, 6860, 1332};
 
 // RGB Bt2100 -> Yuv Bt2100
 // Y = 0.2627 * R + 0.677998 * G + 0.059302 * B
@@ -456,9 +456,9 @@ std::unique_ptr<uhdr_raw_image_ext_t> convert_raw_input_to_ycbcr_neon(uhdr_raw_i
     if (src->cg == UHDR_CG_BT_709) {
       coeffs_ptr = kRgb709ToYuv_coeffs_neon;
     } else if (src->cg == UHDR_CG_BT_2100) {
-      coeffs_ptr = kRgbDispP3ToYuv_coeffs_neon;
-    } else if (src->cg == UHDR_CG_DISPLAY_P3) {
       coeffs_ptr = kRgb2100ToYuv_coeffs_neon;
+    } else if (src->cg == UHDR_CG_DISPLAY_P3) {
+      coeffs_ptr = kRgb601ToYuv_coeffs_neon;
     } else {
       return dst;
     }
diff --git a/tests/gainmapmath_test.cpp b/tests/gainmapmath_test.cpp
index 570742db..4e1ef5a8 100644
--- a/tests/gainmapmath_test.cpp
+++ b/tests/gainmapmath_test.cpp
@@ -11,6 +11,8 @@
 #include <gtest/gtest.h>
 #include <gmock/gmock.h>
 
+#include <vector>
+
 #include "ultrahdr/gainmapmath.h"
 #ifdef UHDR_ENABLE_SMPTE2094_50
 #include "smpte2094_50/smpte2094_50.h"
@@ -875,6 +877,49 @@ TEST_F(GainMapMathTest, YuvConversionNeon) {
     EXPECT_NEAR(result7.v, expected_values.v.at(7), 1);
   }
 }
+
+TEST_F(GainMapMathTest, Rgba8888ToYuv444NeonMatchesScalarForWideGamuts) {
+  constexpr size_t kWidth = 16;
+  constexpr size_t kHeight = 1;
+  const std::array<std::array<uint8_t, 4>, 4> colors{{
+      {{255, 0, 0, 255}},
+      {{0, 255, 0, 255}},
+      {{0, 0, 255, 255}},
+      {{37, 149, 233, 255}},
+  }};
+
+  std::vector<uint8_t> rgba(kWidth * kHeight * 4);
+  for (size_t x = 0; x < kWidth; ++x) {
+    std::copy(colors[x % colors.size()].begin(), colors[x % colors.size()].end(),
+              rgba.begin() + x * 4);
+  }
+
+  for (uhdr_color_gamut_t gamut : {UHDR_CG_DISPLAY_P3, UHDR_CG_BT_2100}) {
+    uhdr_raw_image_t source{};
+    source.fmt = UHDR_IMG_FMT_32bppRGBA8888;
+    source.cg = gamut;
+    source.ct = UHDR_CT_SRGB;
+    source.range = UHDR_CR_FULL_RANGE;
+    source.w = kWidth;
+    source.h = kHeight;
+    source.planes[UHDR_PLANE_PACKED] = rgba.data();
+    source.stride[UHDR_PLANE_PACKED] = kWidth;
+
+    auto scalar = convert_raw_input_to_ycbcr(&source);
+    auto neon = convert_raw_input_to_ycbcr_neon(&source);
+    ASSERT_NE(scalar, nullptr);
+    ASSERT_NE(neon, nullptr);
+
+    for (auto plane : {UHDR_PLANE_Y, UHDR_PLANE_U, UHDR_PLANE_V}) {
+      const auto* scalar_data = static_cast<const uint8_t*>(scalar->planes[plane]);
+      const auto* neon_data = static_cast<const uint8_t*>(neon->planes[plane]);
+      for (size_t x = 0; x < kWidth; ++x) {
+        EXPECT_NEAR(scalar_data[x], neon_data[x], 1)
+            << "gamut=" << gamut << ", plane=" << plane << ", x=" << x;
+      }
+    }
+  }
+}
 #endif
 
 TEST_F(GainMapMathTest, TransformYuv420) {