Simplify llvm.dbg.declare for pointer arguments

[maleadt]

Thanks to @vtjnash, again. Fixes JuliaGPU/CUDAnative.jl#133

Before:

define void @julia_kernel_2({ i64 } addrspace(11)* nocapture nonnull readonly dereferenceable(8)) !dbg !4 {
top:
  %1 = call %jl_value_t*** @julia.ptls_states()
  %2 = bitcast %jl_value_t*** %1 to %jl_value_t addrspace(10)**
  %3 = getelementptr inbounds %jl_value_t addrspace(10)*, %jl_value_t addrspace(10)** %2, i64 2
  %4 = bitcast %jl_value_t addrspace(10)** %3 to i64**
  %5 = load i64*, i64** %4, !tbaa !14, !invariant.load !3
  %"#temp#" = alloca { i64 } addrspace(11)*
  store { i64 } addrspace(11)* %0, { i64 } addrspace(11)** %"#temp#"
  call void @llvm.dbg.declare(metadata { i64 } addrspace(11)** %"#temp#", metadata !13, metadata !DIExpression(DW_OP_deref)), !dbg !17
  ret void, !dbg !17
}

The store into a stack slot turned out problematic for the GC invariant verifier, but only in the case of CUDAnative kernels where we wrap functions and thus the following check failed to prevent that check:

julia/src/llvm-gc-invariant-verifier.cpp

Lines 86 to 88 in 1001087

	/* We currently don't obey this for arguments. That's ok - they're
	externally rooted. */
	if (!isa<Argument>(SI.getValueOperand())) {

This check can probably go now?

After:

define void @julia_kernel_2({ i64 } addrspace(11)* nocapture nonnull readonly dereferenceable(8)) !dbg !4 {
top:
  %1 = call %jl_value_t*** @julia.ptls_states()
  %2 = bitcast %jl_value_t*** %1 to %jl_value_t addrspace(10)**
  %3 = getelementptr inbounds %jl_value_t addrspace(10)*, %jl_value_t addrspace(10)** %2, i64 2
  %4 = bitcast %jl_value_t addrspace(10)** %3 to i64**
  %5 = load i64*, i64** %4, !tbaa !14, !invariant.load !3
  call void @llvm.dbg.declare(metadata { i64 } addrspace(11)* %0, metadata !13, metadata !DIExpression(DW_OP_deref)), !dbg !17
  ret void, !dbg !17
}

[maleadt]

Would be nice to have in 1.2, so I'll merge this.

[kryptine]

:(

…

On Mon, Apr 15, 2019, 19:39 Tim Besard ***@***.***> wrote: Merged #31694 <#31694> into master. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#31694 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AF44wyVsNtKi85D5ko2gPS55NLKoOBcxks5vhHLtgaJpZM4cqM54> .

[StefanKarpinski]

Why are you unhappy about this, @kryptine?