[RISCV] Add C intrinsics for scalar bitmanip and crypto

This adds riscv_bitmanip and riscv_crypto.h

This is based on the proposed spec here riscv-non-isa/riscv-c-api-doc#44

Tests that previously used builtins directly now use the intrinsics.

Reviewed By: wangpc

Differential Revision: https://reviews.llvm.org/D155647

clang/lib/Headers/CMakeLists.txt

@@ -116,6 +116,8 @@ set(ppc_htm_files
   )
 
 set(riscv_files
+  riscv_bitmanip.h
+  riscv_crypto.h
   riscv_ntlh.h
   sifive_vector.h
   )

clang/lib/Headers/riscv_bitmanip.h

@@ -0,0 +1,179 @@
+/*===---- riscv_bitmanip.h - RISC-V Zb* intrinsics --------------------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __RISCV_BITMANIP_H
+#define __RISCV_BITMANIP_H
+
+#include <stdint.h>
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+#if defined(__riscv_zbb)
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_orc_b_32(uint32_t __x) {
+  return __builtin_riscv_orc_b_32(__x);
+}
+
+static __inline__ unsigned __attribute__((__always_inline__, __nodebug__))
+__riscv_clz_32(uint32_t __x) {
+  return __builtin_riscv_clz_32(__x);
+}
+
+static __inline__ unsigned __attribute__((__always_inline__, __nodebug__))
+__riscv_ctz_32(uint32_t __x) {
+  return __builtin_riscv_ctz_32(__x);
+}
+
+static __inline__ unsigned __attribute__((__always_inline__, __nodebug__))
+__riscv_cpop_32(uint32_t __x) {
+  return __builtin_riscv_cpop_32(__x);
+}
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_orc_b_64(uint64_t __x) {
+  return __builtin_riscv_orc_b_64(__x);
+}
+
+static __inline__ unsigned __attribute__((__always_inline__, __nodebug__))
+__riscv_clz_64(uint64_t __x) {
+  return __builtin_riscv_clz_64(__x);
+}
+
+static __inline__ unsigned __attribute__((__always_inline__, __nodebug__))
+__riscv_ctz_64(uint64_t __x) {
+  return __builtin_riscv_ctz_64(__x);
+}
+
+static __inline__ unsigned __attribute__((__always_inline__, __nodebug__))
+__riscv_cpop_64(uint64_t __x) {
+  return __builtin_riscv_cpop_64(__x);
+}
+#endif
+#endif // defined(__riscv_zbb)
+
+#if defined(__riscv_zbb) || defined(__riscv_zbkb)
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_rev8_32(uint32_t __x) {
+  return __builtin_bswap32(__x);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_rol_32(uint32_t __x, uint32_t __y) {
+  return __builtin_rotateleft32(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_ror_32(uint32_t __x, uint32_t __y) {
+  return __builtin_rotateright32(__x, __y);
+}
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_rev8_64(uint64_t __x) {
+  return __builtin_bswap64(__x);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_rol_64(uint64_t __x, uint32_t __y) {
+  return __builtin_rotateleft64(__x, __y);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_ror_64(uint64_t __x, uint32_t __y) {
+  return __builtin_rotateright64(__x, __y);
+}
+#endif
+#endif // defined(__riscv_zbb) || defined(__riscv_zbkb)
+
+#if defined(__riscv_zbkb)
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_brev8_32(uint32_t __x) {
+  return __builtin_riscv_brev8_32(__x);
+}
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_brev8_64(uint64_t __x) {
+  return __builtin_riscv_brev8_64(__x);
+}
+#endif
+
+#if __riscv_xlen == 32
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_unzip_32(uint32_t __x) {
+  return __builtin_riscv_unzip_32(__x);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_zip_32(uint32_t __x) {
+  return __builtin_riscv_zip_32(__x);
+}
+#endif
+#endif // defined(__riscv_zbkb)
+
+#if defined(__riscv_zbkc)
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_clmul_32(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_clmul_32(__x, __y);
+}
+
+#if __riscv_xlen == 32
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_clmulh_32(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_clmulh_32(__x, __y);
+}
+#endif
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_clmul_64(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_clmul_64(__x, __y);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_clmulh_64(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_clmulh_64(__x, __y);
+}
+#endif
+#endif // defined(__riscv_zbkc)
+
+#if defined(__riscv_zbkx)
+#if __riscv_xlen == 32
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_xperm4_32(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_xperm4_32(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_xperm8_32(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_xperm8_32(__x, __y);
+}
+#endif
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_xperm4_64(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_xperm4_64(__x, __y);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_xperm8_64(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_xperm8_64(__x, __y);
+}
+#endif
+#endif // defined(__riscv_zbkx)
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif

clang/lib/Headers/riscv_crypto.h

@@ -0,0 +1,180 @@
+/*===---- riscv_crypto.h - RISC-V Zk* intrinsics ---------------------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __RISCV_CRYPTO_H
+#define __RISCV_CRYPTO_H
+
+#include <stdint.h>
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+#if defined(__riscv_zknd)
+#if __riscv_xlen == 32
+#define __riscv_aes32dsi(x, y, bs) __builtin_riscv_aes32dsi(x, y, bs)
+#define __riscv_aes32dsmi(x, y, bs) __builtin_riscv_aes32dsmi(x, y, bs)
+#endif
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_aes64ds(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_aes64ds(__x, __y);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_aes64dsm(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_aes64dsm(__x, __y);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_aes64im(uint64_t __x) {
+  return __builtin_riscv_aes64im(__x);
+}
+#endif
+#endif // defined(__riscv_zknd)
+
+#if defined(__riscv_zkne)
+#if __riscv_xlen == 32
+#define __riscv_aes32esi(x, y, bs) __builtin_riscv_aes32esi(x, y, bs)
+#define __riscv_aes32esmi(x, y, bs) __builtin_riscv_aes32esmi(x, y, bs)
+#endif
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_aes64es(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_aes64es(__x, __y);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_aes64esm(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_aes64esm(__x, __y);
+}
+#endif
+#endif // defined(__riscv_zknd)
+
+#if defined(__riscv_zknd) || defined(__riscv_zkne)
+#if __riscv_xlen == 64
+#define __riscv_aes64ks1i(x, rnum) __builtin_riscv_aes64ks1i(x, rnum)
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_aes64ks2(uint64_t __x, uint64_t __y) {
+  return __builtin_riscv_aes64ks2(__x, __y);
+}
+#endif
+#endif // defined(__riscv_zknd) || defined(__riscv_zkne)
+
+#if defined(__riscv_zknh)
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha256sig0(uint32_t __x) {
+  return __builtin_riscv_sha256sig0(__x);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha256sig1(uint32_t __x) {
+  return __builtin_riscv_sha256sig1(__x);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha256sum0(uint32_t __x) {
+  return __builtin_riscv_sha256sum0(__x);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha256sum1(uint32_t __x) {
+  return __builtin_riscv_sha256sum1(__x);
+}
+
+#if __riscv_xlen == 32
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sig0h(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sig0h(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sig0l(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sig0l(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sig1h(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sig1h(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sig1l(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sig1l(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sum0l(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sum0l(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sum0r(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sum0r(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sum1l(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sum1l(__x, __y);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sum1r(uint32_t __x, uint32_t __y) {
+  return __builtin_riscv_sha512sum1r(__x, __y);
+}
+#endif
+
+#if __riscv_xlen == 64
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sig0(uint64_t __x) {
+  return __builtin_riscv_sha512sig0(__x);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sig1(uint64_t __x) {
+  return __builtin_riscv_sha512sig1(__x);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sum0(uint64_t __x) {
+  return __builtin_riscv_sha512sum0(__x);
+}
+
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sha512sum1(uint64_t __x) {
+  return __builtin_riscv_sha512sum1(__x);
+}
+#endif
+#endif // defined(__riscv_zknh)
+
+#if defined(__riscv_zksh)
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sm3p0(uint32_t __x) {
+  return __builtin_riscv_sm3p0(__x);
+}
+
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
+__riscv_sm3p1(uint32_t __x) {
+  return __builtin_riscv_sm3p1(__x);
+}
+#endif // defined(__riscv_zksh)
+
+#if defined(__riscv_zksed)
+#define __riscv_sm4ed(x, y, bs) __builtin_riscv_sm4ed(x, y, bs);
+#define __riscv_sm4ks(x, y, bs) __builtin_riscv_sm4ks(x, y, bs);
+#endif // defined(__riscv_zksh)
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif