forked from pool/nodejs-electron
56 lines
2.9 KiB
Diff
56 lines
2.9 KiB
Diff
|
--- src/v8/src/objects/simd.cc.orig 2025-08-15 17:36:40.113358284 +0200
|
||
|
|
+++ src/v8/src/objects/simd.cc 2025-08-17 16:29:26.826424105 +0200
|
||
|
|
@@ -953,7 +953,7 @@ bool Uint8ArrayFromHexWithNeon(const bas
|
||
|
|
uint8x16_t second_part_first_batch =
|
||
|
|
vld1q_u8(reinterpret_cast<const uint8_t*>(&input_vector[i + 8]));
|
||
|
|
first_batch =
|
||
|
|
- vmovn_high_u16(vmovn_u16(first_batch), second_part_first_batch);
|
||
|
|
+ vmovn_high_u16(vmovn_u16(vreinterpretq_u16_u8(first_batch)), vreinterpretq_u16_u8(second_part_first_batch));
|
||
|
|
}
|
||
|
|
|
||
|
|
// Load second batch of 16 hex characters into a Neon register
|
||
|
|
@@ -966,20 +966,20 @@ bool Uint8ArrayFromHexWithNeon(const bas
|
||
|
|
uint8x16_t second_part_second_batch =
|
||
|
|
vld1q_u8(reinterpret_cast<const uint8_t*>(&input_vector[i + 24]));
|
||
|
|
second_batch =
|
||
|
|
- vmovn_high_u16(vmovn_u16(second_batch), second_part_second_batch);
|
||
|
|
+ vmovn_high_u16(vmovn_u16(vreinterpretq_u16_u8(second_batch)), vreinterpretq_u16_u8(second_part_second_batch));
|
||
|
|
}
|
||
|
|
|
||
|
|
// low nibbles are values with even indexes in fist_batch.
|
||
|
|
// (0x36, 0x00, 0x66, 0x00, 0x46, 0x00, 0x32, 0x00, 0x31, 0x00, 0x32, 0x00,
|
||
|
|
// 0x31, 0x00, 0x32, 0x00)
|
||
|
|
- uint8x16_t first_batch_lo_nibbles =
|
||
|
|
- vreinterpretq_u8_u16(vshrq_n_u16(vreinterpretq_u16_u8(first_batch), 8));
|
||
|
|
+ uint16x8_t first_batch_lo_nibbles =
|
||
|
|
+ vshrq_n_u16(vreinterpretq_u16_u8(first_batch), 8);
|
||
|
|
|
||
|
|
// low nibbles are values with even indexes in second_batch.
|
||
|
|
// (0x36, 0x00, 0x66, 0x00, 0x66, 0x00, 0x32, 0x00, 0x31, 0x00, 0x32, 0x00,
|
||
|
|
// 0x31, 0x00, 0x66, 0x00)
|
||
|
|
- uint8x16_t second_batch_lo_nibbles = vreinterpretq_u8_u16(
|
||
|
|
- vshrq_n_u16(vreinterpretq_u16_u8(second_batch), 8));
|
||
|
|
+ uint16x8_t second_batch_lo_nibbles =
|
||
|
|
+ vshrq_n_u16(vreinterpretq_u16_u8(second_batch), 8);
|
||
|
|
|
||
|
|
// Append low nibbles of first batch and second batch and remove 0x00s.
|
||
|
|
// (0x36, 0x66, 0x46, 0x32, 0x31, 0x32, 0x31, 0x32, 0x36, 0x66, 0x66, 0x32,
|
||
|
|
@@ -992,7 +992,7 @@ bool Uint8ArrayFromHexWithNeon(const bas
|
||
|
|
// significant byte. (0x36, 0x36, 0x36, 0x36, 0x36, 0x37, 0x36, 0x37, 0x36,
|
||
|
|
// 0x36, 0x36, 0x36, 0x36, 0x37, 0x36, 0x66)
|
||
|
|
uint8x16_t hi_nibbles =
|
||
|
|
- vmovn_high_u16(vmovn_u16(first_batch), second_batch);
|
||
|
|
+ vmovn_high_u16(vmovn_u16(vreinterpretq_u16_u8(first_batch)), vreinterpretq_u16_u8(second_batch));
|
||
|
|
|
||
|
|
// mapping low nibbles to uint8_t values.
|
||
|
|
// (0x06, 0x0f, 0x0f, 0x02, 0x01, 0x02, 0x01, 0x02, 0x06, 0x0f, 0x0f, 0x02,
|
||
|
|
@@ -1023,7 +1023,7 @@ bool Uint8ArrayFromHexWithNeon(const bas
|
||
|
|
// (0x60, 0x60, 0x60, 0x60, 0x60, 0x70, 0x60, 0x70, 0x60, 0x60, 0x60, 0x60,
|
||
|
|
// 0x60, 0x70, 0x60, 0xf0)
|
||
|
|
uint8x16_t uint8_shifted_high_nibbles =
|
||
|
|
- vshlq_n_u64(vreinterpretq_u64_u8(uint8_high_nibbles), 4);
|
||
|
|
+ vreinterpretq_u8_u64(vshlq_n_u64(vreinterpretq_u64_u8(uint8_high_nibbles), 4));
|
||
|
|
|
||
|
|
// final result of combining pairs of uint8_t values of low and high
|
||
|
|
// nibbles.
|