Fixed off-by-one blending errors in the NEON drawhelper code.

For example blending alpha 0xff with alpha 0xff and a 0.5 opacity gave
a result of 0xfe instead of the correct 0xff. This caused some
autotests to fail on ARM/NEON.

Reviewed-by: TrustMe

1 files changed, 65 insertions, 65 deletions

diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp
index 25860a0..77c5202 100644
--- a/src/gui/painting/qdrawhelper_neon.cpp
+++ b/src/gui/painting/qdrawhelper_neon.cpp
@@ -48,43 +48,43 @@
 
 QT_BEGIN_NAMESPACE
 
-static inline int16x8_t qvdiv_255_s16(int16x8_t x, int16x8_t half)
+static inline uint16x8_t qvdiv_255_u16(uint16x8_t x, uint16x8_t half)
 {
     // result = (x + (x >> 8) + 0x80) >> 8
 
-    const int16x8_t temp = vshrq_n_s16(x, 8); // x >> 8
-    const int16x8_t sum_part = vaddq_s16(x, half); // x + 0x80
-    const int16x8_t sum = vaddq_s16(temp, sum_part);
+    const uint16x8_t temp = vshrq_n_u16(x, 8); // x >> 8
+    const uint16x8_t sum_part = vaddq_u16(x, half); // x + 0x80
+    const uint16x8_t sum = vaddq_u16(temp, sum_part);
 
-    return vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(sum), 8));
+    return vshrq_n_u16(sum, 8);
 }
 
-static inline int16x8_t qvbyte_mul_s16(int16x8_t x, int16x8_t alpha, int16x8_t half)
+static inline uint16x8_t qvbyte_mul_u16(uint16x8_t x, uint16x8_t alpha, uint16x8_t half)
 {
     // t = qRound(x * alpha / 255.0)
 
-    const int16x8_t t = vmulq_s16(x, alpha); // t
-    return qvdiv_255_s16(t, half);
+    const uint16x8_t t = vmulq_u16(x, alpha); // t
+    return qvdiv_255_u16(t, half);
 }
 
-static inline int16x8_t qvinterpolate_pixel_255(int16x8_t x, int16x8_t a, int16x8_t y, int16x8_t b, int16x8_t half)
+static inline uint16x8_t qvinterpolate_pixel_255(uint16x8_t x, uint16x8_t a, uint16x8_t y, uint16x8_t b, uint16x8_t half)
 {
     // t = x * a + y * b
 
-    const int16x8_t ta = vmulq_s16(x, a);
-    const int16x8_t tb = vmulq_s16(y, b);
+    const uint16x8_t ta = vmulq_u16(x, a);
+    const uint16x8_t tb = vmulq_u16(y, b);
 
-    return qvdiv_255_s16(vaddq_s16(ta, tb), half);
+    return qvdiv_255_u16(vaddq_u16(ta, tb), half);
 }
 
-static inline int16x8_t qvsource_over_s16(int16x8_t src16, int16x8_t dst16, int16x8_t half, int16x8_t full)
+static inline uint16x8_t qvsource_over_u16(uint16x8_t src16, uint16x8_t dst16, uint16x8_t half, uint16x8_t full)
 {
-    const int16x4_t alpha16_high = vdup_lane_s16(vget_high_s16(src16), 3);
-    const int16x4_t alpha16_low = vdup_lane_s16(vget_low_s16(src16), 3);
+    const uint16x4_t alpha16_high = vdup_lane_u16(vget_high_u16(src16), 3);
+    const uint16x4_t alpha16_low = vdup_lane_u16(vget_low_u16(src16), 3);
 
-    const int16x8_t alpha16 = vsubq_s16(full, vcombine_s16(alpha16_low, alpha16_high));
+    const uint16x8_t alpha16 = vsubq_u16(full, vcombine_u16(alpha16_low, alpha16_high));
 
-    return vaddq_s16(src16, qvbyte_mul_s16(dst16, alpha16, half));
+    return vaddq_u16(src16, qvbyte_mul_u16(dst16, alpha16, half));
 }
 
 void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
@@ -94,21 +94,21 @@ void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
 {
     const uint *src = (const uint *) srcPixels;
     uint *dst = (uint *) destPixels;
-    int16x8_t half = vdupq_n_s16(0x80);
-    int16x8_t full = vdupq_n_s16(0xff);
+    uint16x8_t half = vdupq_n_u16(0x80);
+    uint16x8_t full = vdupq_n_u16(0xff);
     if (const_alpha == 256) {
         for (int y = 0; y < h; ++y) {
             int x = 0;
             for (; x < w-3; x += 4) {
-                int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
+                uint32x4_t src32 = vld1q_u32((uint32_t *)&src[x]);
                 if ((src[x] & src[x+1] & src[x+2] & src[x+3]) >= 0xff000000) {
                     // all opaque
-                    vst1q_s32((int32_t *)&dst[x], src32);
+                    vst1q_u32((uint32_t *)&dst[x], src32);
                 } else if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
-                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
+                    uint32x4_t dst32 = vld1q_u32((uint32_t *)&dst[x]);
 
-                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
-                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
+                    const uint8x16_t src8 = vreinterpretq_u8_u32(src32);
+                    const uint8x16_t dst8 = vreinterpretq_u8_u32(dst32);
 
                     const uint8x8_t src8_low = vget_low_u8(src8);
                     const uint8x8_t dst8_low = vget_low_u8(dst8);
@@ -116,19 +116,19 @@ void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
                     const uint8x8_t src8_high = vget_high_u8(src8);
                     const uint8x8_t dst8_high = vget_high_u8(dst8);
 
-                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
-                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
+                    const uint16x8_t src16_low = vmovl_u8(src8_low);
+                    const uint16x8_t dst16_low = vmovl_u8(dst8_low);
 
-                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
-                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
+                    const uint16x8_t src16_high = vmovl_u8(src8_high);
+                    const uint16x8_t dst16_high = vmovl_u8(dst8_high);
 
-                    const int16x8_t result16_low = qvsource_over_s16(src16_low, dst16_low, half, full);
-                    const int16x8_t result16_high = qvsource_over_s16(src16_high, dst16_high, half, full);
+                    const uint16x8_t result16_low = qvsource_over_u16(src16_low, dst16_low, half, full);
+                    const uint16x8_t result16_high = qvsource_over_u16(src16_high, dst16_high, half, full);
 
-                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
-                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
+                    const uint32x2_t result32_low = vreinterpret_u32_u8(vmovn_u16(result16_low));
+                    const uint32x2_t result32_high = vreinterpret_u32_u8(vmovn_u16(result16_high));
 
-                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
+                    vst1q_u32((uint32_t *)&dst[x], vcombine_u32(result32_low, result32_high));
                 }
             }
             for (; x<w; ++x) {
@@ -143,16 +143,16 @@ void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
         }
     } else if (const_alpha != 0) {
         const_alpha = (const_alpha * 255) >> 8;
-        int16x8_t const_alpha16 = vdupq_n_s16(const_alpha);
+        uint16x8_t const_alpha16 = vdupq_n_u16(const_alpha);
         for (int y = 0; y < h; ++y) {
             int x = 0;
             for (; x < w-3; x += 4) {
                 if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
-                    int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
-                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
+                    uint32x4_t src32 = vld1q_u32((uint32_t *)&src[x]);
+                    uint32x4_t dst32 = vld1q_u32((uint32_t *)&dst[x]);
 
-                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
-                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
+                    const uint8x16_t src8 = vreinterpretq_u8_u32(src32);
+                    const uint8x16_t dst8 = vreinterpretq_u8_u32(dst32);
 
                     const uint8x8_t src8_low = vget_low_u8(src8);
                     const uint8x8_t dst8_low = vget_low_u8(dst8);
@@ -160,22 +160,22 @@ void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
                     const uint8x8_t src8_high = vget_high_u8(src8);
                     const uint8x8_t dst8_high = vget_high_u8(dst8);
 
-                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
-                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
+                    const uint16x8_t src16_low = vmovl_u8(src8_low);
+                    const uint16x8_t dst16_low = vmovl_u8(dst8_low);
 
-                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
-                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
+                    const uint16x8_t src16_high = vmovl_u8(src8_high);
+                    const uint16x8_t dst16_high = vmovl_u8(dst8_high);
 
-                    const int16x8_t srcalpha16_low = qvbyte_mul_s16(src16_low, const_alpha16, half);
-                    const int16x8_t srcalpha16_high = qvbyte_mul_s16(src16_high, const_alpha16, half);
+                    const uint16x8_t srcalpha16_low = qvbyte_mul_u16(src16_low, const_alpha16, half);
+                    const uint16x8_t srcalpha16_high = qvbyte_mul_u16(src16_high, const_alpha16, half);
 
-                    const int16x8_t result16_low = qvsource_over_s16(srcalpha16_low, dst16_low, half, full);
-                    const int16x8_t result16_high = qvsource_over_s16(srcalpha16_high, dst16_high, half, full);
+                    const uint16x8_t result16_low = qvsource_over_u16(srcalpha16_low, dst16_low, half, full);
+                    const uint16x8_t result16_high = qvsource_over_u16(srcalpha16_high, dst16_high, half, full);
 
-                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
-                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
+                    const uint32x2_t result32_low = vreinterpret_u32_u8(vmovn_u16(result16_low));
+                    const uint32x2_t result32_high = vreinterpret_u32_u8(vmovn_u16(result16_high));
 
-                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
+                    vst1q_u32((uint32_t *)&dst[x], vcombine_u32(result32_low, result32_high));
                 }
             }
             for (; x<w; ++x) {
@@ -206,19 +206,19 @@ void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl,
         if (const_alpha != 0) {
             const uint *src = (const uint *) srcPixels;
             uint *dst = (uint *) destPixels;
-            int16x8_t half = vdupq_n_s16(0x80);
+            uint16x8_t half = vdupq_n_u16(0x80);
             const_alpha = (const_alpha * 255) >> 8;
             int one_minus_const_alpha = 255 - const_alpha;
-            int16x8_t const_alpha16 = vdupq_n_s16(const_alpha);
-            int16x8_t one_minus_const_alpha16 = vdupq_n_s16(255 - const_alpha);
+            uint16x8_t const_alpha16 = vdupq_n_u16(const_alpha);
+            uint16x8_t one_minus_const_alpha16 = vdupq_n_u16(255 - const_alpha);
             for (int y = 0; y < h; ++y) {
                 int x = 0;
                 for (; x < w-3; x += 4) {
-                    int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
-                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
+                    uint32x4_t src32 = vld1q_u32((uint32_t *)&src[x]);
+                    uint32x4_t dst32 = vld1q_u32((uint32_t *)&dst[x]);
 
-                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
-                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
+                    const uint8x16_t src8 = vreinterpretq_u8_u32(src32);
+                    const uint8x16_t dst8 = vreinterpretq_u8_u32(dst32);
 
                     const uint8x8_t src8_low = vget_low_u8(src8);
                     const uint8x8_t dst8_low = vget_low_u8(dst8);
@@ -226,19 +226,19 @@ void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl,
                     const uint8x8_t src8_high = vget_high_u8(src8);
                     const uint8x8_t dst8_high = vget_high_u8(dst8);
 
-                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
-                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
+                    const uint16x8_t src16_low = vmovl_u8(src8_low);
+                    const uint16x8_t dst16_low = vmovl_u8(dst8_low);
 
-                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
-                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
+                    const uint16x8_t src16_high = vmovl_u8(src8_high);
+                    const uint16x8_t dst16_high = vmovl_u8(dst8_high);
 
-                    const int16x8_t result16_low = qvinterpolate_pixel_255(src16_low, const_alpha16, dst16_low, one_minus_const_alpha16, half);
-                    const int16x8_t result16_high = qvinterpolate_pixel_255(src16_high, const_alpha16, dst16_high, one_minus_const_alpha16, half);
+                    const uint16x8_t result16_low = qvinterpolate_pixel_255(src16_low, const_alpha16, dst16_low, one_minus_const_alpha16, half);
+                    const uint16x8_t result16_high = qvinterpolate_pixel_255(src16_high, const_alpha16, dst16_high, one_minus_const_alpha16, half);
 
-                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
-                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
+                    const uint32x2_t result32_low = vreinterpret_u32_u8(vmovn_u16(result16_low));
+                    const uint32x2_t result32_high = vreinterpret_u32_u8(vmovn_u16(result16_high));
 
-                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
+                    vst1q_u32((uint32_t *)&dst[x], vcombine_u32(result32_low, result32_high));
                 }
                 for (; x<w; ++x) {
                     uint s = src[x];