diff options
| author | android-build-team Robot <android-build-team-robot@google.com> | 2019-06-29 03:11:56 +0000 |
|---|---|---|
| committer | android-build-team Robot <android-build-team-robot@google.com> | 2019-06-29 03:11:56 +0000 |
| commit | 96399584ccc0b735f43543f5599d246558c871cf (patch) | |
| tree | 61f45a95ef42e952ec121583ed73b8649e058bc1 /files/source/compare_neon64.cc | |
| parent | 073a65c5dbada4de5b11a353bc3ab38952c3155d (diff) | |
| parent | ab123ac62c872f89e20f10c96e651ead21414ffc (diff) | |
| download | libyuv-96399584ccc0b735f43543f5599d246558c871cf.tar.gz | |
Snap for 5697374 from ab123ac62c872f89e20f10c96e651ead21414ffc to rvc-release
Change-Id: I2b203f3622defc19e5a6afab23de12c4a62f16e5
Diffstat (limited to 'files/source/compare_neon64.cc')
| -rw-r--r-- | files/source/compare_neon64.cc | 88 |
1 files changed, 57 insertions, 31 deletions
diff --git a/files/source/compare_neon64.cc b/files/source/compare_neon64.cc index f9c7df98..6e8f672a 100644 --- a/files/source/compare_neon64.cc +++ b/files/source/compare_neon64.cc @@ -20,39 +20,65 @@ extern "C" { #if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__) -uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) { - volatile uint32 sse; - asm volatile ( - "eor v16.16b, v16.16b, v16.16b \n" - "eor v18.16b, v18.16b, v18.16b \n" - "eor v17.16b, v17.16b, v17.16b \n" - "eor v19.16b, v19.16b, v19.16b \n" +// 256 bits at a time +// uses short accumulator which restricts count to 131 KB +uint32_t HammingDistance_NEON(const uint8_t* src_a, + const uint8_t* src_b, + int count) { + uint32_t diff; + asm volatile( + "movi v4.8h, #0 \n" - "1: \n" - MEMACCESS(0) - "ld1 {v0.16b}, [%0], #16 \n" - MEMACCESS(1) - "ld1 {v1.16b}, [%1], #16 \n" - "subs %w2, %w2, #16 \n" - "usubl v2.8h, v0.8b, v1.8b \n" - "usubl2 v3.8h, v0.16b, v1.16b \n" - "smlal v16.4s, v2.4h, v2.4h \n" - "smlal v17.4s, v3.4h, v3.4h \n" - "smlal2 v18.4s, v2.8h, v2.8h \n" - "smlal2 v19.4s, v3.8h, v3.8h \n" - "b.gt 1b \n" + "1: \n" + "ld1 {v0.16b, v1.16b}, [%0], #32 \n" + "ld1 {v2.16b, v3.16b}, [%1], #32 \n" + "eor v0.16b, v0.16b, v2.16b \n" + "eor v1.16b, v1.16b, v3.16b \n" + "cnt v0.16b, v0.16b \n" + "cnt v1.16b, v1.16b \n" + "subs %w2, %w2, #32 \n" + "add v0.16b, v0.16b, v1.16b \n" + "uadalp v4.8h, v0.16b \n" + "b.gt 1b \n" - "add v16.4s, v16.4s, v17.4s \n" - "add v18.4s, v18.4s, v19.4s \n" - "add v19.4s, v16.4s, v18.4s \n" - "addv s0, v19.4s \n" - "fmov %w3, s0 \n" - : "+r"(src_a), - "+r"(src_b), - "+r"(count), - "=r"(sse) - : - : "cc", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19"); + "uaddlv s4, v4.8h \n" + "fmov %w3, s4 \n" + : "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(diff) + : + : "cc", "v0", "v1", "v2", "v3", "v4"); + return diff; +} + +uint32_t SumSquareError_NEON(const uint8_t* src_a, + const uint8_t* src_b, + int count) { + uint32_t sse; + asm volatile( + "eor v16.16b, v16.16b, v16.16b \n" + "eor v18.16b, v18.16b, v18.16b \n" + "eor v17.16b, v17.16b, v17.16b \n" + "eor v19.16b, v19.16b, v19.16b \n" + + "1: \n" + "ld1 {v0.16b}, [%0], #16 \n" + "ld1 {v1.16b}, [%1], #16 \n" + "subs %w2, %w2, #16 \n" + "usubl v2.8h, v0.8b, v1.8b \n" + "usubl2 v3.8h, v0.16b, v1.16b \n" + "smlal v16.4s, v2.4h, v2.4h \n" + "smlal v17.4s, v3.4h, v3.4h \n" + "smlal2 v18.4s, v2.8h, v2.8h \n" + "smlal2 v19.4s, v3.8h, v3.8h \n" + "b.gt 1b \n" + + "add v16.4s, v16.4s, v17.4s \n" + "add v18.4s, v18.4s, v19.4s \n" + "add v19.4s, v16.4s, v18.4s \n" + "addv s0, v19.4s \n" + "fmov %w3, s0 \n" + : "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse) + : + : "cc", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19"); return sse; } |