Nikita Popov

Improve handling of multiple builds in .ninja_log

1

When using `-a` mode, if `.ninja_log` contains multiple builds, produce a trace that is still sequential (with one incremental build after each other), instead of pretending that they all happened...

Uninitialized memory treated as poison

2

It looks like alive2 is treating uninitialized memory as poison: https://alive2.llvm.org/ce/z/VKRwvf ```llvm define i8 @src() { %a = alloca i8, align 1 %v = load i8, ptr %a, align 1...

False positive with nocapture

3

https://alive2.llvm.org/ce/z/9jjAiv ```llvm define void @src(ptr align 4 %val) { %val1 = alloca i32, align 4 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %val1, ptr align 4 %val, i64 4, i1 false) call...

For example simde_vmul_laneq_s32: https://github.com/simd-everywhere/simde/blob/51743e7920b6e867678cb50e9c62effe28f70b33/simde/arm/neon/mul_lane.h#L389 The multiplication here is UB if signed overflow occurs. The ARM documentation does not specify such UB: https://developer.arm.com/architectures/instruction-sets/intrinsics/vmul_laneq_s32 The correct way to perform the multiplication would...

[SCCP] Improve worklist management

2

SCCP currently stores instructions whose lattice value has changed in a worklist, and then updates their users in the main loop. This may result in instructions unnecessarily being visited multiple...

[AArch64] Mark neon.stN intrinsics as writeonly

3

I found this peculiar comment in EarlyCSE: https://github.com/llvm/llvm-project/blob/1c78d8d9d7bcb4b20910047ad7db35f177a17c8c/llvm/lib/Transforms/Scalar/EarlyCSE.cpp#L1620-L1624 Looking back over history, this seems to be referring to the aarch64.neon.stN intrinsics, which are indeed not marked writeonly (though the ldN...

No user-directed error if model fails to load

5

### Issue Description When using `ramalama run`, if the model fails to load (for example, because it exceeds the GPU memory size), no user-directed error is printed. Instead the prompt...

https://alive2.llvm.org/ce/z/uYhgqi ```llvm target datalayout = "p:64:64:64:32" define i1 @src_ptrtoint(ptr %p, i32 %offset) { %p2 = getelementptr nuw i8, ptr %p, i32 %offset %p.addr = ptrtoint ptr %p to i32 %p2.addr...

[ConstantInt] Disable implicit truncation in ConstantInt::get()

3

Disable implicit truncation in the ConstantInt constructor by default. This means that it needs to be passed a signed/unsigned (depending on the IsSigned flag) value matching the bit width. The...