MSVC Build Tools v14.51 delivers significant compiler optimization improvements measured against SPEC CPU 2017 and Unreal Engine's CitySample benchmark. Key changes include: a new SSA-based loop optimizer replacing a decades-old monolithic implementation (2.5% compile throughput gain, 23 bugs fixed); an expanded SLP vectorizer with oversized vector support for Arm64 and sub-vector-width vectorization on x64; multiple SROA (Scalar Replacement of Aggregates) improvements covering indirect assignments, larger structs up to 64 bytes, unions, address-taken relaxation, casted fields, and repacking; hoisting of vectorizer pointer overlap checks out of inner loops; logical-to-bitwise OR conversion; shift-CMP folding; improved ARM NEON codegen using SHLL2; unconditional store execution; AVX stack round-trip elimination; single and limited call site inlining via whole-program analysis; enhanced branch elimination for heapification patterns; loop unswitching for iterator loops; memset/memcpy improvements using overlapping copies and direct YMM expansion; Arm64 bitwise ops with shifted registers; ternary operator simplifications; cross-control-flow copy propagation; and relaxed loop unrolling constraints. Integer benchmark gains reach up to 6.5% on Arm64 with PGO. The release is available in Visual Studio 2026 Insiders preview.
Table of contents
New SSA Loop Optimizer Copy linkNew SLP Vectorizer Copy linkSROA Improvements Copy linkHoisting Vectorizer Pointer Overlap Checks Copy linkLogical to Bitwise OR Copy linkShift-CMP folding Copy linkNeon Codegen Improvement Copy linkUnconditional Store Execution Copy linkImproved AVX Optimization Copy linkSingle and Limited Call Site Inlining Copy linkBranch Elimination Copy linkLoop Unswitching Copy linkMemset and Memcpy improvements Copy linkArm64 Bitwise Ops with Shifted Registers Copy linkTernary Operator Optimization Copy linkImproved Copy Prop Copy linkLoop Unrolling Copy linkConclusion Copy linkSort: