fix alignment of emit_unbox_store copy (#52505)

The dest alignment might be determined to be greater than the source.
For example, observed with:

    code_llvm(dump_module=true, optimize=false, (Int,)) do x
        Pair(ntuple(i -> 0x00, 8), x)
    end

Where the alignment of the first field of the Pair is at least 4, but
the alignment of the ntuple data is 1.

(discovered while analyzing an ARM build failure @staticfloat)

Author: vtjnash

doc/src/devdocs/debuggingtips.md

@@ -41,11 +41,16 @@ useful.
 
 ## Useful Julia functions for Inspecting those variables
 
-  * `jl_gdblookup($rip)` :: For looking up the current function and line. (use `$eip` on i686 platforms)
+  * `jl_print_task_backtraces(0)` :: Similar to gdb's `thread apply all bt` or lldb's `thread backtrace
+    all`. Runs all threads while printing backtraces for all existing tasks.
+  * `jl_gdblookup($pc)` :: For looking up the current function and line.
+  * `jl_gdblookupinfo($pc)` :: For looking up the current method instance object.
+  * `jl_gdbdumpcode(mi)` :: For dumping all of `code_typed/code_llvm/code_asm` when the REPL is not working right.
   * `jlbacktrace()` :: For dumping the current Julia backtrace stack to stderr. Only usable after
     `record_backtrace()` has been called.
   * `jl_dump_llvm_value(Value*)` :: For invoking `Value->dump()` in gdb, where it doesn't work natively.
     For example, `f->linfo->functionObject`, `f->linfo->specFunctionObject`, and `to_function(f->linfo)`.
+  * `jl_dump_llvm_module(Module*)` :: For invoking `Module->dump()` in gdb, where it doesn't work natively.
   * `Type->dump()` :: only works in lldb. Note: add something like `;1` to prevent lldb from printing
     its prompt over the output
   * `jl_eval_string("expr")` :: for invoking side-effects to modify the current state or to lookup

src/aotcompile.cpp

@@ -1820,6 +1820,58 @@ void addTargetPasses(legacy::PassManagerBase *PM, const Triple &triple, TargetIR
     PM->add(createTargetTransformInfoWrapperPass(std::move(analysis)));
 }
 
+// sometimes in GDB you want to find out what code was created from a mi
+extern "C" JL_DLLEXPORT_CODEGEN jl_code_info_t *jl_gdbdumpcode(jl_method_instance_t *mi)
+{
+    jl_llvmf_dump_t llvmf_dump;
+    size_t world = jl_current_task->world_age;
+    JL_STREAM *stream = (JL_STREAM*)STDERR_FILENO;
+
+    jl_printf(stream, "---- dumping IR for ----\n");
+    jl_static_show(stream, (jl_value_t*)mi);
+    jl_printf(stream, "\n----\n");
+
+    jl_printf(stream, "\n---- unoptimized IR ----");
+    jl_get_llvmf_defn(&llvmf_dump, mi, world, 0, false, jl_default_cgparams);
+    if (llvmf_dump.F) {
+        jl_value_t *ir = jl_dump_function_ir(&llvmf_dump, 0, 1, "source");
+        jl_static_show(stream, ir);
+    }
+    jl_printf(stream, "----\n");
+
+    jl_printf(stream, "\n---- optimized IR ----");
+    jl_get_llvmf_defn(&llvmf_dump, mi, world, 0, true, jl_default_cgparams);
+    if (llvmf_dump.F) {
+        jl_value_t *ir = jl_dump_function_ir(&llvmf_dump, 0, 1, "source");
+        jl_static_show(stream, ir);
+    }
+    jl_printf(stream, "----\n");
+
+    jl_printf(stream, "\n---- assembly ----");
+    jl_get_llvmf_defn(&llvmf_dump, mi, world, 0, true, jl_default_cgparams);
+    if (llvmf_dump.F) {
+        jl_value_t *ir = jl_dump_function_asm(&llvmf_dump, 0, "", "source", 0, true);
+        jl_static_show(stream, ir);
+    }
+    jl_printf(stream, "----\n");
+
+    jl_code_info_t *src = NULL;
+    jl_value_t *ci = jl_default_cgparams.lookup(mi, world, world);
+    if (ci != jl_nothing) {
+        jl_code_instance_t *codeinst = (jl_code_instance_t*)ci;
+        src = (jl_code_info_t*)jl_atomic_load_relaxed(&codeinst->inferred);
+        if ((jl_value_t*)src != jl_nothing && !jl_is_code_info(src) && jl_is_method(mi->def.method)) {
+            JL_GC_PUSH2(&codeinst, &src);
+            src = jl_uncompress_ir(mi->def.method, codeinst, (jl_value_t*)src);
+            JL_GC_POP();
+        }
+    }
+    if (!src || (jl_value_t*)src == jl_nothing) {
+        src = jl_type_infer(mi, world, 0);
+    }
+    return src;
+}
+
 // --- native code info, and dump function to IR and ASM ---
 // Get pointer to llvm::Function instance, compiling if necessary
 // for use in reflection from Julia.

src/cgutils.cpp

@@ -2911,7 +2911,8 @@ static void init_bits_cgval(jl_codectx_t &ctx, Value *newv, const jl_cgval_t& v,
 {
     // newv should already be tagged
     if (v.ispointer()) {
-        emit_memcpy(ctx, newv, jl_aliasinfo_t::fromTBAA(ctx, tbaa), v, jl_datatype_size(v.typ), sizeof(void*), julia_alignment(v.typ));
+        unsigned align = std::max(julia_alignment(v.typ), (unsigned)sizeof(void*));
+        emit_memcpy(ctx, newv, jl_aliasinfo_t::fromTBAA(ctx, tbaa), v, jl_datatype_size(v.typ), align, julia_alignment(v.typ));
     }
     else {
         init_bits_value(ctx, newv, v.V, tbaa);

src/intrinsics.cpp

@@ -520,7 +520,7 @@ static void emit_unbox_store(jl_codectx_t &ctx, const jl_cgval_t &x, Value *dest
     }
 
     Value *src = data_pointer(ctx, x);
-    emit_memcpy(ctx, dest, jl_aliasinfo_t::fromTBAA(ctx, tbaa_dest), src, jl_aliasinfo_t::fromTBAA(ctx, x.tbaa), jl_datatype_size(x.typ), alignment, alignment, isVolatile);
+    emit_memcpy(ctx, dest, jl_aliasinfo_t::fromTBAA(ctx, tbaa_dest), src, jl_aliasinfo_t::fromTBAA(ctx, x.tbaa), jl_datatype_size(x.typ), alignment, julia_alignment(x.typ), isVolatile);
 }
 
 static jl_datatype_t *staticeval_bitstype(const jl_cgval_t &targ)