Use fewer instructions when unpacking uint6s.

Differential Revision: D64548639

Pull Request resolved: pytorch#1109

torchao/experimental/kernels/cpu/aarch64/bitpacking/uint6.h

@@ -114,15 +114,20 @@ TORCHAO_ALWAYS_INLINE inline void vec_unpack_32_uint6_values(
   uint8x8_t packed1 = vld1_u8(packed + 8);
   uint8x8_t packed2 = vld1_u8(packed + 16);
 
-  // unpacked[3] = ((packed[0] & 0b1100'0000u) >> 6) |
-  //               ((packed[1] & 0b1100'0000u) >> 4) |
-  //               ((packed[2] & 0b1100'0000u) >> 2);
-  const uint8x8_t high = vdup_n_u8(0b1100'0000u);
   uint8x8_t unpacked3;
-  unpacked3 = vorr_u8(
-      vshr_n_u8(vand_u8(packed0, high), 6),
-      vshr_n_u8(vand_u8(packed1, high), 4));
-  unpacked3 = vorr_u8(unpacked3, vshr_n_u8(vand_u8(packed2, high), 2));
+  // We want to extract bits 123456 and place them in unpacked3.
+  // Packed structure is:
+  //
+  // packed0: 56 | abcdef
+  // packed1: 34 | ghijkl
+  // packed2: 12 | mnopqr
+  //
+  // unpacked3 = 1234 ghij
+  unpacked3 = vsri_n_u8(packed2, packed1, 2);
+  // unpacked3 = 1234 56ab
+  unpacked3 = vsri_n_u8(unpacked3, packed0, 4);
+  // unpacked3 = 0012 3456
+  unpacked3 = vshr_n_u8(unpacked3, 2);
 
   // unpacked[i] = packed[i] & 0b11'1111u;
   const uint8x8_t mask = vdup_n_u8(0b11'1111u);
@@ -183,14 +188,19 @@ TORCHAO_ALWAYS_INLINE inline void vec_unpack_64_uint6_values(
   unpacked1 = vld1q_u8(packed + 16);
   unpacked2 = vld1q_u8(packed + 32);
 
-  // unpacked[3] = ((packed[0] & 0b1100'0000u) >> 6) |
-  //               ((packed[1] & 0b1100'0000u) >> 4) |
-  //               ((packed[2] & 0b1100'0000u) >> 2);
-  const uint8x16_t high = vdupq_n_u8(0b1100'0000u);
-  unpacked3 = vorrq_u8(
-      vshrq_n_u8(vandq_u8(unpacked0, high), 6),
-      vshrq_n_u8(vandq_u8(unpacked1, high), 4));
-  unpacked3 = vorrq_u8(unpacked3, vshrq_n_u8(vandq_u8(unpacked2, high), 2));
+  // We want to extract bits 123456 and place them in unpacked3.
+  // Packed structure is:
+  //
+  // packed0: 56 | abcdef
+  // packed1: 34 | ghijkl
+  // packed2: 12 | mnopqr
+  //
+  // unpacked3 = 1234 ghij
+  unpacked3 = vsriq_n_u8(unpacked2, unpacked1, 2);
+  // unpacked3 = 1234 56ab
+  unpacked3 = vsriq_n_u8(unpacked3, unpacked0, 4);
+  // unpacked3 = 0012 3456
+  unpacked3 = vshrq_n_u8(unpacked3, 2);
 
   // unpacked[i] = packed[i] & 0b11'1111u;
   const uint8x16_t mask = vdupq_n_u8(0b11'1111u);