Fix fast getptls ccall lowering. (#55507)

(cherry picked from commit 5a633b7)

Author: gbaraldi

src/ccall.cpp

@@ -1584,9 +1584,8 @@ static jl_cgval_t emit_ccall(jl_codectx_t &ctx, jl_value_t **args, size_t nargs)
         emit_gc_safepoint(ctx.builder, ctx.types().T_size, get_current_ptls(ctx), ctx.tbaa().tbaa_const);
         return ghostValue(ctx, jl_nothing_type);
     }
-    else if (is_libjulia_func("jl_get_ptls_states")) {
+    else if (is_libjulia_func(jl_get_ptls_states)) {
         ++CCALL_STAT(jl_get_ptls_states);
-        assert(lrt == ctx.types().T_size);
         assert(!isVa && !llvmcall && nccallargs == 0);
         JL_GC_POP();
         return mark_or_box_ccall_result(ctx,

src/julia_threads.h

@@ -287,10 +287,7 @@ typedef struct _jl_tls_states_t {
 #endif
 } jl_tls_states_t;
 
-#ifndef JL_LIBRARY_EXPORTS
-// deprecated (only for external consumers)
 JL_DLLEXPORT void *jl_get_ptls_states(void);
-#endif
 
 // Update codegen version in `ccall.cpp` after changing either `pause` or `wake`
 #ifdef __MIC__

test/compiler/codegen.jl

@@ -870,3 +870,6 @@ function foonopreds()
     pkgid.uuid !== nothing ? pkgid.uuid : false
 end
 @test foonopreds() !== nothing
+
+# Core.getptls() special handling
+@test !occursin("call ptr @jlplt", get_llvm(Core.getptls, Tuple{})) #It should lower to a direct load of the ptls and not a ccall