Unrolled build for #150426

Rollup merge of #150426 - ZuseZ4:offload-register-lib, r=davidtwco

Update offload test and verify that tgt_(un)register_lib have the right type

Apparently, we weren't running offload tests when Enzyme wasn't built. Time to fix that.
Also adds a test mode which generates the host IR, but does not expect device IR/artifacts. This way, we don't have to handle artifacts and paths in our tests.
Also removes some outdated documentation.

cc `@Kevinsala,` `@Sa4dUs`

closes: #150415

~~blocked on `needs-offload` infrastructure landing in #150427

Author: rust-timer

compiler/rustc_codegen_llvm/src/base.rs

@@ -93,8 +93,13 @@ pub(crate) fn compile_codegen_unit(
             // They are necessary for correct offload execution. We do this here to simplify the
             // `offload` intrinsic, avoiding the need for tracking whether it's the first
             // intrinsic call or not.
-            let has_host_offload =
-                cx.sess().opts.unstable_opts.offload.iter().any(|o| matches!(o, Offload::Host(_)));
+            let has_host_offload = cx
+                .sess()
+                .opts
+                .unstable_opts
+                .offload
+                .iter()
+                .any(|o| matches!(o, Offload::Host(_) | Offload::Test));
             if has_host_offload && !cx.sess().target.is_like_gpu {
                 cx.offload_globals.replace(Some(OffloadGlobals::declare(&cx)));
             }

compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs

@@ -49,8 +49,9 @@ impl<'ll> OffloadGlobals<'ll> {
         let bin_desc = cx.type_named_struct("struct.__tgt_bin_desc");
         cx.set_struct_body(bin_desc, &tgt_bin_desc_ty, false);
 
-        let register_lib = declare_offload_fn(&cx, "__tgt_register_lib", mapper_fn_ty);
-        let unregister_lib = declare_offload_fn(&cx, "__tgt_unregister_lib", mapper_fn_ty);
+        let reg_lib_decl = cx.type_func(&[cx.type_ptr()], cx.type_void());
+        let register_lib = declare_offload_fn(&cx, "__tgt_register_lib", reg_lib_decl);
+        let unregister_lib = declare_offload_fn(&cx, "__tgt_unregister_lib", reg_lib_decl);
         let init_ty = cx.type_func(&[], cx.type_void());
         let init_rtls = declare_offload_fn(cx, "__tgt_init_all_rtls", init_ty);
 

compiler/rustc_session/src/config.rs

@@ -196,6 +196,8 @@ pub enum Offload {
     Device,
     /// Second step in the offload pipeline, generates the host code to call kernels.
     Host(String),
+    /// Test is similar to Host, but allows testing without a device artifact.
+    Test,
 }
 
 /// The different settings that the `-Z autodiff` flag can have.

compiler/rustc_session/src/options.rs

@@ -794,7 +794,8 @@ mod desc {
     pub(crate) const parse_list_with_polarity: &str =
         "a comma-separated list of strings, with elements beginning with + or -";
     pub(crate) const parse_autodiff: &str = "a comma separated list of settings: `Enable`, `PrintSteps`, `PrintTA`, `PrintTAFn`, `PrintAA`, `PrintPerf`, `PrintModBefore`, `PrintModAfter`, `PrintModFinal`, `PrintPasses`, `NoPostopt`, `LooseTypes`, `Inline`, `NoTT`";
-    pub(crate) const parse_offload: &str = "a comma separated list of settings: `Enable`";
+    pub(crate) const parse_offload: &str =
+        "a comma separated list of settings: `Host=<Absolute-Path>`, `Device`, `Test`";
     pub(crate) const parse_comma_list: &str = "a comma-separated list of strings";
     pub(crate) const parse_opt_comma_list: &str = parse_comma_list;
     pub(crate) const parse_number: &str = "a number";
@@ -1471,6 +1472,13 @@ pub mod parse {
                     }
                     Offload::Device
                 }
+                "Test" => {
+                    if let Some(_) = arg {
+                        // Test does not accept a value
+                        return false;
+                    }
+                    Offload::Test
+                }
                 _ => {
                     // FIXME(ZuseZ4): print an error saying which value is not recognized
                     return false;

tests/codegen-llvm/gpu_offload/gpu_host.rs

@@ -1,15 +1,10 @@
-//@ compile-flags: -Zoffload=Enable -Zunstable-options -C opt-level=3  -Clto=fat
+//@ compile-flags: -Zoffload=Test -Zunstable-options -C opt-level=3  -Clto=fat
 //@ no-prefer-dynamic
-//@ needs-enzyme
+//@ needs-offload
 
 // This test is verifying that we generate __tgt_target_data_*_mapper before and after a call to the
 // kernel_1. Better documentation to what each global or variable means is available in the gpu
-// offlaod code, or the LLVM offload documentation. This code does not launch any GPU kernels yet,
-// and will be rewritten once a proper offload frontend has landed.
-//
-// We currently only handle memory transfer for specific calls to functions named `kernel_{num}`,
-// when inside of a function called main. This, too, is a temporary workaround for not having a
-// frontend.
+// offload code, or the LLVM offload documentation.
 
 #![feature(rustc_attrs)]
 #![feature(core_intrinsics)]
@@ -22,6 +17,20 @@ fn main() {
     core::hint::black_box(&x);
 }
 
+#[unsafe(no_mangle)]
+#[inline(never)]
+pub fn kernel_1(x: &mut [f32; 256]) {
+    core::intrinsics::offload(_kernel_1, [256, 1, 1], [32, 1, 1], (x,))
+}
+
+#[unsafe(no_mangle)]
+#[inline(never)]
+pub fn _kernel_1(x: &mut [f32; 256]) {
+    for i in 0..256 {
+        x[i] = 21.0;
+    }
+}
+
 // CHECK: %struct.ident_t = type { i32, i32, i32, i32, ptr }
 // CHECK: %struct.__tgt_offload_entry = type { i64, i16, i16, i32, ptr, ptr, i64, i64, ptr }
 // CHECK: %struct.__tgt_bin_desc = type { i32, ptr, ptr, ptr }
@@ -36,8 +45,9 @@ fn main() {
 // CHECK: @.offloading.entry_name._kernel_1 = internal unnamed_addr constant [10 x i8] c"_kernel_1\00", section ".llvm.rodata.offloading", align 1
 // CHECK: @.offloading.entry._kernel_1 = internal constant %struct.__tgt_offload_entry { i64 0, i16 1, i16 1, i32 0, ptr @._kernel_1.region_id, ptr @.offloading.entry_name._kernel_1, i64 0, i64 0, ptr null }, section "llvm_offload_entries", align 8
 
-// CHECK: Function Attrs: nounwind
 // CHECK: declare i32 @__tgt_target_kernel(ptr, i64, i32, i32, ptr, ptr)
+// CHECK: declare void @__tgt_register_lib(ptr) local_unnamed_addr
+// CHECK: declare void @__tgt_unregister_lib(ptr) local_unnamed_addr
 
 // CHECK: define{{( dso_local)?}} void @main()
 // CHECK-NEXT: start:
@@ -94,17 +104,3 @@ fn main() {
 // CHECK-NEXT:   call void @__tgt_unregister_lib(ptr nonnull %EmptyDesc)
 // CHECK-NEXT:   ret void
 // CHECK-NEXT: }
-
-#[unsafe(no_mangle)]
-#[inline(never)]
-pub fn kernel_1(x: &mut [f32; 256]) {
-    core::intrinsics::offload(_kernel_1, [256, 1, 1], [32, 1, 1], (x,))
-}
-
-#[unsafe(no_mangle)]
-#[inline(never)]
-pub fn _kernel_1(x: &mut [f32; 256]) {
-    for i in 0..256 {
-        x[i] = 21.0;
-    }
-}