Optimized and correct version of movemask128 for ARM

Closes #99

https://gcc.godbolt.org/z/cTjKqzcvn

Previous version was not correct because movemask thought of having bytes 0xFF. We can fully match the semantics + do it faster with USRA instructions.

Re-submission to a develop branch

diff --git a/src/util/arch/arm/simd_utils.h b/src/util/arch/arm/simd_utils.h
index 902d36249eba96f468304a63224ffefb6f55bf78..e6836b2526a078b9313c1cc7edf2221943ecad0c 100644 (file)
--- a/src/util/arch/arm/simd_utils.h
+++ b/src/util/arch/arm/simd_utils.h
@@ -379,17 +379,15 @@ static really_inline m128 eq64_m128(m128 a, m128 b) {
 }
 
 static really_inline u32 movemask128(m128 a) {
-    static const uint8x16_t powers = { 1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128 };
-
-    // Compute the mask from the input
-    uint8x16_t mask  = (uint8x16_t) vpaddlq_u32(vpaddlq_u16(vpaddlq_u8(vandq_u8((uint8x16_t)a, powers))));
-    uint8x16_t mask1 = vextq_u8(mask, (uint8x16_t)zeroes128(), 7);
-    mask = vorrq_u8(mask, mask1);
-
-    // Get the resulting bytes
-    uint16_t output;
-    vst1q_lane_u16((uint16_t*)&output, (uint16x8_t)mask, 0);
-    return output;
+    uint8x16_t input = vreinterpretq_u8_s32(a);
+    uint16x8_t high_bits = vreinterpretq_u16_u8(vshrq_n_u8(input, 7));
+    uint32x4_t paired16 =
+        vreinterpretq_u32_u16(vsraq_n_u16(high_bits, high_bits, 7));
+    uint64x2_t paired32 =
+        vreinterpretq_u64_u32(vsraq_n_u32(paired16, paired16, 14));
+    uint8x16_t paired64 =
+        vreinterpretq_u8_u64(vsraq_n_u64(paired32, paired32, 28));
+    return vgetq_lane_u8(paired64, 0) | ((int) vgetq_lane_u8(paired64, 8) << 8);
 }
 
 static really_inline m128 set1_16x8(u8 c) {