optimize base64 for riscv64 with RVV vrgather and vectorized decode#3388
Merged
Conversation
- Add RVV-accelerated base64 encode using vrgather vector indexed load for parallel table lookup (16 bytes per iteration, LMUL=4) - Add RVV-accelerated base64 decode using vectorized range comparisons and vmerge chain for classification (16 input bytes per iteration) - Integrate RVV dispatch into modp_b64.cc with compile-time selection when __riscv_vector is defined - Add modp_b64_rvv.cc to CMakeLists.txt build sources - Performance (SG2044 rv64gcv): decode 3-5x faster, up to 764 MB/s - Tested with RFC 4648 vectors and 100-round random data roundtrip Signed-off-by: Xiaofei Gong <gongxiaofei24@iscas.ac.cn> Signed-off-by: YuanSheng <yuansheng@isrc.iscas.ac.cn>
Felix-Gong
force-pushed
the
riscv-base64-rvv
branch
2 times, most recently
from
July 15, 2026 00:25
0b651dc to
0b98453
Compare
Contributor
There was a problem hiding this comment.
Pull request overview
This PR introduces RISC-V Vector (RVV) accelerated base64 encode/decode paths in the existing modp_b64 implementation under src/butil/third_party/, aiming to improve performance on rv64gcv while keeping non-RVV builds unchanged via compile-time dispatch.
Changes:
- Add RVV-specific base64 encode/decode implementation (
modp_b64_rvv.{h,cc}) and dispatch to it frommodp_b64.ccwhen__riscv_vectoris available. - Include the RVV header from
modp_b64.hfor RVV builds. - Add the new RVV source file to the CMake build.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 5 comments.
Show a summary per file
| File | Description |
|---|---|
| src/butil/third_party/modp_b64/modp_b64.h | Includes RVV header when building with RVV enabled. |
| src/butil/third_party/modp_b64/modp_b64.cc | Adds compile-time RVV dispatch for encode/decode. |
| src/butil/third_party/modp_b64/modp_b64_rvv.h | Declares RVV-accelerated encode/decode entry points. |
| src/butil/third_party/modp_b64/modp_b64_rvv.cc | Implements RVV-accelerated encode/decode routines. |
| CMakeLists.txt | Adds the RVV implementation file to BUTIL_SOURCES. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| @@ -0,0 +1,23 @@ | |||
| /* ... Apache 2.0 license header ... */ | |||
Contributor
Author
There was a problem hiding this comment.
Fixed. Replaced the placeholder with the full Apache-2.0 license header matching the style used in other modp_b64 files.
| break; | ||
| } | ||
|
|
||
| return (size_t)(d - (uint8_t*)dest); |
Contributor
Author
There was a problem hiding this comment.
Fixed. Added *d = '\0' before return in both the RVV fast path and scalar tail, matching modp_b64_encode() behavior.
Comment on lines
+150
to
+154
| if (len == 0) return 0; | ||
|
|
||
| size_t i = 0; | ||
| uint8_t* p = (uint8_t*)dest; | ||
|
|
Contributor
Author
There was a problem hiding this comment.
Fixed. Added trailing '=' stripping loop before decode, matching modp_b64_decode() validation logic.
Comment on lines
+215
to
+237
| /* Scalar fallback for remaining <16 bytes or chunks with errors */ | ||
| { | ||
| size_t remaining = len - i; | ||
| const uint8_t* s = (const uint8_t*)(src + i); | ||
| uint8_t* d = p; | ||
|
|
||
| for (size_t si = 0; si + 4 <= remaining; si += 4) { | ||
| uint8_t c0 = s[si], c1 = s[si+1], c2 = s[si+2], c3 = s[si+3]; | ||
| int d0, d1, d2, d3; | ||
|
|
||
| #define DECODE_CHAR(c, val) do { if (c >= 'A' && c <= 'Z') val = c - 'A'; else if (c >= 'a' && c <= 'z') val = c - 'a' + 26; else if (c >= '0' && c <= '9') val = c - '0' + 52; else if (c == '+') val = 62; else if (c == '/') val = 63; else return MODP_B64_RVV_ERROR; } while (0) | ||
|
|
||
| DECODE_CHAR(c0, d0); DECODE_CHAR(c1, d1); | ||
| DECODE_CHAR(c2, d2); DECODE_CHAR(c3, d3); | ||
|
|
||
| #undef DECODE_CHAR | ||
|
|
||
| *d++ = (uint8_t)((d0 << 2) | (d1 >> 4)); | ||
| *d++ = (uint8_t)(((d1 & 0x0F) << 4) | (d2 >> 2)); | ||
| *d++ = (uint8_t)(((d2 & 0x03) << 6) | d3); | ||
| } | ||
| return (size_t)(d - (uint8_t*)dest); | ||
| } |
Contributor
Author
There was a problem hiding this comment.
Fixed. Added explicit handling for leftover 2 and 3 chars after the main decode loop, producing the correct partial output (1 or 2 bytes respectively).
Comment on lines
419
to
+420
| ${PROJECT_SOURCE_DIR}/src/butil/third_party/modp_b64/modp_b64.cc | ||
| ${PROJECT_SOURCE_DIR}/src/butil/third_party/modp_b64/modp_b64_rvv.cc |
Contributor
Author
There was a problem hiding this comment.
Fixed. Added modp_b64_rvv.cc to the BUTIL_SRCS list in BUILD.bazel.
- Replace placeholder license in modp_b64_rvv.h with Apache-2.0 header - Add NUL terminator in rvv_modp_b64_encode scalar tail - Strip trailing '=' padding at entry of rvv_modp_b64_decode - Handle leftover 2/3 chars in decode scalar fallback - Add modp_b64_rvv.cc to BUILD.bazel BUTIL_SRCS All 5 fixes verified on real rv64gcv hardware (8/8 gtest pass). Signed-off-by: Xiaofei Gong <gongxiaofei24@iscas.ac.cn> Signed-off-by: YuanSheng <yuansheng@isrc.iscas.ac.cn>
Contributor
|
LGTM |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
RVV-accelerated base64 encode/decode for RISC-V 64-bit (rv64gcv).
Changes
modp_b64_rvv.cc: RVV vrgather encode + vectorized decode (80% RVV)modp_b64_rvv.h: Apache-2.0 license headermodp_b64.cc: Runtime dispatch viaUSE_RVV_BASE64BUILD.bazel: Addmodp_b64_rvv.cctoBUTIL_SRCSEncode
Scalar 6-bit index extraction + vectorized vrgather lookup (16 bytes/iter, LMUL=4).
Processes 12 input bytes → 16 output bytes per vector loop iteration.
Decode
Vectorized range classification (A-Z, a-z, 0-9, +, /) + vmerge chain.
Final 4×6→3 byte packing remains scalar (data dependencies prevent vectorization).
Performance (SG2044 rv64gcv, 10-round avg, -O2)
Tested
brpc_base64_unittestall passed