Thông tin chung
  English
  Đề tài NC khoa học
  Bài báo, báo cáo khoa học
  Hướng dẫn Sau đại học
  Sách và giáo trình
  Các học phần và môn giảng dạy
  Giải thưởng khoa học, Phát minh, sáng chế
  Khen thưởng
  Thông tin khác
  Tài liệu tham khảo
  Hiệu chỉnh
 
Số người truy cập: 106,842,188

 On the rounding error of fused multiply-accumulate based floating-point dot-product architectures
Tác giả hoặc Nhóm tác giả: Huynh Viet Thang
Nơi đăng: Tạp chí khoa học công nghệ ĐHĐN
cvs weekly sale cvs print prescription savings cards
; Số: NA;Từ->đến trang: NA;Năm: 2012
Lĩnh vực: Kỹ thuật; Loại: Bài báo khoa học; Thể loại: Trong nước
TÓM TẮT
This paper focuses on the rounding error analysis using affine arithmetic (AA) and bit width allocation of floating-point (FP) dot-product implementations on reconfigurable hardware for advanced digital signal processing applications. An AA-based rounding error model for the fused multiply-accumulate (FMA) operation is, for the first time, suggested in this paper. Using the proposed error model, we then both experimentally and analytically investigate the rounding error of the FMA-based FP dot-product implementation. It is surprisingly shown that the overall rounding error of a floating-point dot-product implementation can significantly be improved by employing a parallel architecture, rather than by employing an FMA. Additionally, the analytical rounding error models for all floating-point dot-product architectures are derived, allowing for an efficient design space exploration and which are the keys to specialised code generators.
walgreens prints coupons rx coupons printable free printable coupons
ABSTRACT
This paper focuses on the rounding error analysis using affine arithmetic (AA) and bit width allocation of floating-point (FP) dot-product implementations on reconfigurable hardware for advanced digital signal processing applications. An AA-based rounding error model for the fused multiply-accumulate (FMA) operation is, for the first time, suggested in this paper. Using the proposed error model, we then both experimentally and analytically investigate the rounding error of the FMA-based FP dot-product implementation. It is surprisingly shown that the overall rounding error of a floating-point dot-product implementation can significantly be improved by employing a parallel architecture, rather than by employing an FMA. Additionally, the analytical rounding error models for all floating-point dot-product architectures are derived, allowing for an efficient design space exploration and which are the keys to specialised code generators.
© Đại học Đà Nẵng
 
 
Địa chỉ: 41 Lê Duẩn Thành phố Đà Nẵng
Điện thoại: (84) 0236 3822 041 ; Email: dhdn@ac.udn.vn