TY - GEN
T1 - Approximating warps with intra-warp operand value similarity
AU - Wong, Daniel
AU - Kim, Nam Sung
AU - Annavaram, Murali
N1 - This work was supported in part by NSF CCF-0954211, CCF-0953603, CNS-1217102, and DARPA (HR0011-12-2-0020). Nam Sung Kim has a financial interest in Samsung Semiconductor and AMD.
PY - 2016/4/1
Y1 - 2016/4/1
N2 - Value locality, the recurrence of a previously-seen value, has been the enabler of myriad optimization techniques in traditional processors. Value similarity relaxes the constraint of value locality by allowing values to differ in the lowest significant bits where values are micro-architecturally near. With the end of Dennard Scaling and the turn towards massively parallel accelerators, we revisit value similarity in the context of GPUs. We identify a form of value similarity called intra-warp operand value similarity, which is abundant in GPUs. We present Warp Approximation, which leverages intra-warp operand value similarity to trade off accuracy for energy. Warp Approximation dynamically identifies intra-warp operand value similarity in hardware, and executes a single representative thread on behalf of all the active threads in a warp, thereby producing a representative value with approximate value locality. This representative value can then be stored compactly in the register file as a value similar scalar, reducing the read and write energy when dealing with approximate data. With Warp Approximation, we can reduce execution unit energy by 37%, register file energy by 28%, and improve overall GPGPU energy efficiency by 26% with minimal quality degradation.
AB - Value locality, the recurrence of a previously-seen value, has been the enabler of myriad optimization techniques in traditional processors. Value similarity relaxes the constraint of value locality by allowing values to differ in the lowest significant bits where values are micro-architecturally near. With the end of Dennard Scaling and the turn towards massively parallel accelerators, we revisit value similarity in the context of GPUs. We identify a form of value similarity called intra-warp operand value similarity, which is abundant in GPUs. We present Warp Approximation, which leverages intra-warp operand value similarity to trade off accuracy for energy. Warp Approximation dynamically identifies intra-warp operand value similarity in hardware, and executes a single representative thread on behalf of all the active threads in a warp, thereby producing a representative value with approximate value locality. This representative value can then be stored compactly in the register file as a value similar scalar, reducing the read and write energy when dealing with approximate data. With Warp Approximation, we can reduce execution unit energy by 37%, register file energy by 28%, and improve overall GPGPU energy efficiency by 26% with minimal quality degradation.
UR - http://www.scopus.com/inward/record.url?scp=84965010104&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84965010104&partnerID=8YFLogxK
U2 - 10.1109/HPCA.2016.7446063
DO - 10.1109/HPCA.2016.7446063
M3 - Conference contribution
AN - SCOPUS:84965010104
T3 - Proceedings - International Symposium on High-Performance Computer Architecture
SP - 176
EP - 187
BT - Proceedings of the 2016 IEEE International Symposium on High-Performance Computer Architecture, HPCA 2016
PB - IEEE Computer Society
T2 - 22nd IEEE International Symposium on High Performance Computer Architecture, HPCA 2016
Y2 - 12 March 2016 through 16 March 2016
ER -