Revision as of 09:17, 15 February 2020

Exynos Cheetah (M4) is the successor to the M3, an 8 nm ARM microarchitecture designed by Samsung for their consumer electronics.

Process Technology

The M4 is fabricated on Samsung's 8 nm process (8LPP).

Compiler support

Architecture

The M4 is an incremental microarchitecture that brought a die shrink and minor enhancements.

Key changes from M3

• 8 nm process (from 10 nm)
• ARMv8.2 (from ARMv8)
  • Support for full FP16 scalar extension
  • Support for integer dot product extension
• Front end
  • Larger instruction queue (48 entries, up from 40)
• Back end
  • LSU executiion units reorganized
  • Floating-point execution units reorganized

This list is incomplete; you can help by expanding it.

Block Diagram

Individual Core

Memory Hierarchy

• Cache
  • L1I Caches
    • 64 KiB, 4-way set associative
      • 128 B line size
      • per core
    • Parity-protected
  • L1D Cache
    • 64 KiB, 8-way set associative
      • 64 B line size
      • per core
    • 4 cycles for fastest load-to-use
    • 32 B/cycle load bandwidth
    • 16 B/cycle store bandwidth
  • L2 Cache
    • 512 KiB, 8-way set associative
    • Inclusive of L1
    • 12 cycles latency
    • 32 B/cycle bandwidth
  • L3 Cache
    • 2 MiB, 16-way set associative
      • 1 MiB slice/core
    • Exlusive of L2
    • ~37-cycle typical (NUCA)
  • BIU
    • 80 outstanding transactions

The M3 TLB consists of dedicated L1 TLB for instruction cache (ITLB) and another one for data cache (DTLB). Additionally, there is a unified L2 TLB (STLB).

• TLBs
  • ITLB
    • 512-entry
  • DTLB
    • 32-entry
    • 512-entry Mid-level DTLB
  • STLB
    • 4,096-entry
    • Per core

• BPU
  • 4K-entry main BTB
  • 128-entry µBTB
  • 64-entry return stack
  • 16K-entry L2 BTB

Core

The core of the M4 is largely the same as M3. A number of buffers have been enlarged and some of the execution units have been reorganized.

Execution engine

Floating-point cluster

The execution units on the M4 have been reorganized. In total, three new units were also added - a second FP square root unit, a second vector multiplication unit, and a new horizontal vector arithmetic unit.

Floating-point pipe changes.

Memory subsystem

Samsung also made an enhancement to the M4 memory subsystem. In the M3, there were three AGUs - two dedicated Load AGUs and a single dedicated Store AGU. In the M4, Samsung changed one of the dedicated Load AGUs into a generic AGU capable of handling both loads and stores. In other words, the M4 can now schedule both load and store µOPs on two ports.

All M4 Processors

Bibliography

• LLVM: lib/Target/AArch64/AArch64SchedExynosM4.td