[libc-commits] [libc] [libc] Implement basic 'fenv.h' utilities on the AMD GPU (PR #83500)

[Joseph Huber via libc-commits]

================
@@ -0,0 +1,301 @@
+//===-- amdgpu floating point env manipulation functions --------*- C++ -*-===//
+//
+// 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 LLVM_LIBC_SRC___SUPPORT_FPUTIL_AMDGPU_FENVIMPL_H
+#define LLVM_LIBC_SRC___SUPPORT_FPUTIL_AMDGPU_FENVIMPL_H
+
+#include "src/__support/GPU/utils.h"
+#include "src/__support/macros/attributes.h"
+#include "src/__support/macros/properties/architectures.h"
+
+#if !defined(LIBC_TARGET_ARCH_IS_AMDGPU)
+#error "Invalid include"
+#endif
+
+#include <fenv.h>
+#include <stdint.h>
+
+namespace LIBC_NAMESPACE {
+namespace fputil {
+
+namespace internal {
+
+// Gets the immediate argument to access the AMDGPU hardware register. The
+// register access is encoded in a 16-bit immediate value according to the
+// following layout.
+//
+// ┌──────────────┬──────────────┬───────────────┐
+// │  SIZE[15:11] │ OFFSET[10:6] │    ID[5:0]    │
+// └──────────────┴──────────────┴───────────────┘
+//
+// This will read the size number of bits starting at the offset bit from the
+// corresponding hardware register ID.
+constexpr uint16_t get_register(uint8_t id, uint8_t offset, uint8_t size) {
+  return static_cast<uint16_t>(size << 11 | offset << 6 | id);
----------------
jhuber6 wrote:

I think that's just too noisy. The function is constexpr because the argument needs to be a compile time constant. I could change the arguments to `uint32_t` if you'd prefer.

https://github.com/llvm/llvm-project/pull/83500