[RISCV] Support overloaded version ntlh intrinsic function

[BeMg]

Summary

Here is the proposal https://github.com/riscv-non-isa/riscv-c-api-doc/pull/47.

The version that omit the domain argument imply domain=__RISCV_NTLH_ALL.

type __riscv_ntl_load (type *ptr);
void __riscv_ntl_store (type *ptr, type val);

[asb]

This seems functionally correct to me, but I'd welcome opinions from others who work more with the C intrinsics on if this is the best way to implement the overloading.

[kito-cheng]

clang/lib/Headers/riscv_ntlh.h

28

__SELECT_NTL_LOAD

38

__SELECT_NTL_STORE

[BeMg]

Add prefix __riscv_ for all marco

[craig.topper]

I think another option could be to do this in SemaChecking.cpp where we implement builtin_riscv_ntl_load and builtin_riscv_ntl_store. We already do custom type checking there. We could detect the missing argument and give it a default.

[BeMg]

Use semacheck and CGBuiltin to support overload version functionY

[kito-cheng]

LGTM, and the change seems like is fewer than my exception :)

[eopXD]

clang/test/CodeGen/RISCV/ntlh-intrinsics/riscv32-zihintntl.c

218

Nit: new line here?

[BeMg]

Add new line

[craig.topper]

LGTM

clang/lib/Headers/riscv_ntlh.h

 
 enum {
   __RISCV_NTLH_INNERMOST_PRIVATE = 2,
   __RISCV_NTLH_ALL_PRIVATE,
   __RISCV_NTLH_INNERMOST_SHARED,
   __RISCV_NTLH_ALL
 };
 
-#define __riscv_ntl_load(PTR, DOMAIN) __builtin_riscv_ntl_load((PTR), (DOMAIN))
-#define __riscv_ntl_store(PTR, VAL, DOMAIN)                                    \
+#define __riscv_ntl_load_with_domain(PTR, DOMAIN)                              \
+  __builtin_riscv_ntl_load((PTR), (DOMAIN))
+#define __riscv_ntl_load_without_domain(PTR)                                   \
+  __builtin_riscv_ntl_load((PTR), __RISCV_NTLH_ALL)
+#define __riscv_SELECT_NTL_LOAD(_1, _2, NAME, ...) NAME
+
+#define __riscv_ntl_load(...)                                                  \
+  __riscv_SELECT_NTL_LOAD(__VA_ARGS__, __riscv_ntl_load_with_domain,           \
+                          __riscv_ntl_load_without_domain)(__VA_ARGS__)
+
+#define __riscv_ntl_store_with_domain(PTR, VAL, DOMAIN)                        \
   __builtin_riscv_ntl_store((PTR), (VAL), (DOMAIN))
+#define __riscv_ntl_store_without_domain(PTR, VAL)                             \
+  __builtin_riscv_ntl_store((PTR), (VAL), __RISCV_NTLH_ALL)
+#define __riscv_SELECT_NTL_STORE(_1, _2, _3, NAME, ...) NAME
 
+#define __riscv_ntl_store(...)                                                 \
+  __riscv_SELECT_NTL_STORE(__VA_ARGS__, __riscv_ntl_store_with_domain,         \
+                           __riscv_ntl_store_without_domain)(__VA_ARGS__)
 #endif
 No newline at end of file

clang/test/CodeGen/RISCV/ntlh-intrinsics/riscv32-zihintntl.c

void ntl_all_sizes() {                                       // CHECK-LABEL: ntl_all_sizes
   d1 = __riscv_ntl_load(&d2, __RISCV_NTLH_ALL);   // CHECK: load double{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
   v4si1 = __riscv_ntl_load(&v4si2, __RISCV_NTLH_ALL);   // CHECK: load <4 x i32>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
   v8ss1 = __riscv_ntl_load(&v8ss2, __RISCV_NTLH_ALL);   // CHECK: load <8 x i16>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
   v16sc1 = __riscv_ntl_load(&v16sc2, __RISCV_NTLH_ALL);   // CHECK: load <16 x i8>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
   *scvi1 = __riscv_ntl_load(scvi2, __RISCV_NTLH_ALL);   // CHECK: load <vscale x 2 x i32>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
   *scvs1 = __riscv_ntl_load(scvs2, __RISCV_NTLH_ALL);   // CHECK: load <vscale x 4 x i16>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
   *scvc1 = __riscv_ntl_load(scvc2, __RISCV_NTLH_ALL);   // CHECK: load <vscale x 8 x i8>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
 
+  uc = __riscv_ntl_load(&sc);   // CHECK: load i8{{.*}}align 1, !nontemporal !4, !riscv-nontemporal-domain !8
+  sc = __riscv_ntl_load(&uc);   // CHECK: load i8{{.*}}align 1, !nontemporal !4, !riscv-nontemporal-domain !8
+  us = __riscv_ntl_load(&ss);   // CHECK: load i16{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !8
+  ss = __riscv_ntl_load(&us);   // CHECK: load i16{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !8
+  ui = __riscv_ntl_load(&si);   // CHECK: load i32{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !8
+  si = __riscv_ntl_load(&ui);   // CHECK: load i32{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !8
+  ull = __riscv_ntl_load(&sll); // CHECK: load i64{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  sll = __riscv_ntl_load(&ull); // CHECK: load i64{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  h1 = __riscv_ntl_load(&h2);   // CHECK: load half{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !8
+  f1 = __riscv_ntl_load(&f2);   // CHECK: load float{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !8
+  d1 = __riscv_ntl_load(&d2);   // CHECK: load double{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  v4si1 = __riscv_ntl_load(&v4si2);   // CHECK: load <4 x i32>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
+  v8ss1 = __riscv_ntl_load(&v8ss2);   // CHECK: load <8 x i16>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
+  v16sc1 = __riscv_ntl_load(&v16sc2);   // CHECK: load <16 x i8>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
+  *scvi1 = __riscv_ntl_load(scvi2);   // CHECK: load <vscale x 2 x i32>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  *scvs1 = __riscv_ntl_load(scvs2);   // CHECK: load <vscale x 4 x i16>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  *scvc1 = __riscv_ntl_load(scvc2);   // CHECK: load <vscale x 8 x i8>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+
   __riscv_ntl_store(&uc, 1, __RISCV_NTLH_INNERMOST_PRIVATE);    // CHECK: store i8{{.*}}align 1, !nontemporal !4, !riscv-nontemporal-domain !5
   __riscv_ntl_store(&sc, 1, __RISCV_NTLH_INNERMOST_PRIVATE);    // CHECK: store i8{{.*}}align 1, !nontemporal !4, !riscv-nontemporal-domain !5
   __riscv_ntl_store(&us, 1, __RISCV_NTLH_INNERMOST_PRIVATE);    // CHECK: store i16{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !5
   __riscv_ntl_store(&ss, 1, __RISCV_NTLH_INNERMOST_PRIVATE);    // CHECK: store i16{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !5
   __riscv_ntl_store(&ui, 1, __RISCV_NTLH_INNERMOST_PRIVATE);    // CHECK: store i32{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !5
   __riscv_ntl_store(&si, 1, __RISCV_NTLH_INNERMOST_PRIVATE);    // CHECK: store i32{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !5
   __riscv_ntl_store(&ull, 1, __RISCV_NTLH_INNERMOST_PRIVATE);   // CHECK: store i64{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !5
   __riscv_ntl_store(&sll, 1, __RISCV_NTLH_INNERMOST_PRIVATE);   // CHECK: store i64{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !5
void ntl_all_sizes() {                                       // CHECK-LABEL: ntl_all_sizes
   __riscv_ntl_store(&d1, 1.0, __RISCV_NTLH_ALL);  // CHECK: store double{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
   __riscv_ntl_store(&v4si1, v4si2, __RISCV_NTLH_ALL);  // CHECK: store <4 x i32>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
   __riscv_ntl_store(&v8ss1, v8ss2, __RISCV_NTLH_ALL);  // CHECK: store <8 x i16>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
   __riscv_ntl_store(&v16sc1, v16sc2, __RISCV_NTLH_ALL);  // CHECK: store <16 x i8>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
   __riscv_ntl_store(scvi2, *scvi1, __RISCV_NTLH_ALL);  // CHECK: store <vscale x 2 x i32>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
   __riscv_ntl_store(scvs2, *scvs1, __RISCV_NTLH_ALL);  // CHECK: store <vscale x 4 x i16>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
   __riscv_ntl_store(scvc2, *scvc1, __RISCV_NTLH_ALL);  // CHECK: store <vscale x 8 x i8>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
 
+  __riscv_ntl_store(&uc, 1);    // CHECK: store i8{{.*}}align 1, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&sc, 1);    // CHECK: store i8{{.*}}align 1, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&us, 1);    // CHECK: store i16{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&ss, 1);    // CHECK: store i16{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&ui, 1);    // CHECK: store i32{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&si, 1);    // CHECK: store i32{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&ull, 1);   // CHECK: store i64{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&sll, 1);   // CHECK: store i64{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&h1, 1.0);  // CHECK: store half{{.*}}align 2, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&f1, 1.0);  // CHECK: store float{{.*}}align 4, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&d1, 1.0);  // CHECK: store double{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&v4si1, v4si2);  // CHECK: store <4 x i32>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&v8ss1, v8ss2);  // CHECK: store <8 x i16>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(&v16sc1, v16sc2);  // CHECK: store <16 x i8>{{.*}}align 16, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(scvi2, *scvi1);  // CHECK: store <vscale x 2 x i32>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(scvs2, *scvs1);  // CHECK: store <vscale x 4 x i16>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
+  __riscv_ntl_store(scvc2, *scvc1);  // CHECK: store <vscale x 8 x i8>{{.*}}align 8, !nontemporal !4, !riscv-nontemporal-domain !8
 }
 // clang-format on
 
 // CHECK: !4 = !{i32 1}
 // CHECK: !5 = !{i32 2}
 // CHECK: !6 = !{i32 3}
 // CHECK: !7 = !{i32 4}
 // CHECK: !8 = !{i32 5}
 No newline at end of file