Transparent hardware-assisted memory decompression
Provided are methods and systems for memory decompression using a hardware decompressor that minimizes or eliminates the involvement of software. Custom decompression hardware is added to the memory subsystem, where the decompression hardware handles read accesses caused by, for example, cache misses or requests from devices to compressed memory blocks, by reading a compressed block, decompressing it into an internal buffer, and returning the requested portion of the block. The custom hardware is designed to determine if the block is compressed, and determine the parameters of compression, by checking unused high bits of the physical address of the access. This allows compression to be implemented without additional metadata, because the necessary metadata can be stored in unused bits in the existing page table structures.
1. A hardware decompressor disposed between a processing unit of a computing device and a physical computer memory controller of the computing device, the hardware decompressor comprising:
a dedicated storage medium comprising an internal buffer, the internal buffer local to the dedicated storage medium and not a physical computer memory controlled by the physical computer memory controller; and
a hardware device, wherein the hardware device is configured to:
receive a memory access read request, the received request causing a cache miss or page fault;
determine whether data requested by the memory access read request is in a compressed memory block;
responsive to a determination that the data requested by the memory access read request is not in a compressed memory block, forward the memory access read request to the physical computer memory controller;
responsive to a determination that the data requested by the memory access read request is in a compressed memory block, determine whether the data from the compressed memory block has been previously decompressed and stored in the dedicated storage medium local to the hardware device;
responsive to a determination that the data from the compressed memory block is stored in the dedicated storage medium, return the data from the dedicated storage medium in response to the memory access read request; and
responsive to a determination that the data from the compressed memory block is not stored in the dedicated storage medium,
allocate storage space in the dedicated storage medium;
read the data from the compressed memory block;
decompress the read data from the compressed memory block into the dedicated storage medium; and
return the decompressed data from the dedicated storage medium in response to the memory access read request.
2. The hardware decompressor of claim 1 , wherein the hardware device is configured to:
read the data from the compressed memory block until the data requested by the memory access read request is located; and
decompress the read data from the compressed memory block into the dedicated storage medium only to the point where the requested data is fully retrieved.
3. The hardware decompressor of claim 1 , wherein the hardware device is configured to:
determine whether the data requested by the memory access read request is in the compressed memory block by checking unused bits of page table structures that store information about virtual memory pages.
4. The hardware decompressor of claim 1 , wherein the hardware device is configured to:
selectively alternate between a full memory decompression mode and a partial memory decompression mode.
5. The hardware decompressor of claim 4 , wherein the hardware device is configured to:
alternate between the full memory decompression mode and the partial memory decompression mode based on one or more performance parameters of a device or program associated with the received memory access read request.
6. The hardware decompressor of claim 4 , wherein the hardware device is configured to:
alternate between the full memory decompression mode and the partial memory decompression mode based on a signal received from a device or program associated with the received memory access read request.
7. A method for memory decompression, the method comprising:
receiving a memory access read request, the received request causing a cache miss or page fault to occur;
determining whether data requested by the memory access read request is in a compressed memory block;
in response to determining that the data requested by the memory access read request is not in a compressed memory block, sending the memory access read request to a physical computer memory controller;
in response to determining that the data requested by the memory access read request is in a compressed memory block, determining whether the data from the compressed memory block has previously been decompressed by a hardware decompression device and stored in an internal buffer memory that is local to the hardware decompression device, the internal buffer memory being separate and distinct from a physical computer memory controlled by the physical computer memory controller;
in response to determining that the data from the compressed memory block is stored in the internal buffer memory, returning the data from the internal buffer memory in response to the received memory access read request; and
in response to determining that the data from the compressed memory block is not stored in the internal buffer memory,
allocating storage space in the internal buffer memory;
reading the data from the compressed memory block;
decompressing the read data from the compressed memory block into the internal buffer memory; and
returning the decompressed data from the internal buffer memory in response to the memory access read request.
8. The method of claim 7 , wherein the hardware decompression device is disposed between a processing unit and the physical computer memory.
9. The method of claim 7 , further comprising:
reading the data from the compressed memory block until the data requested by the memory access read request is located; and
decompressing the read data from the compressed memory block into the internal buffer memory only to the point where the requested data is fully retrieved.
10. The method of claim 7 , further comprising:
determining whether the data requested by the memory access read request is in the compressed memory block by checking unused bits of page table structures that store information about virtual memory pages.
11. The method of claim 7 , further comprising:
selectively alternating between a full memory decompression mode and a partial memory decompression mode.
12. The method of claim 11 , wherein the alternation between the full memory decompression mode and the partial memory decompression mode is based on one or more performance parameters of a device or program associated with the received memory access read request.
13. The method of claim 11 , wherein the alternation between the full memory decompression mode and the partial memory decompression mode is based on a signal received from a device or program associated with the received memory access read request.
14. A system for memory decompression, the system comprising:
system memory;
a system memory controller;
at least one processor; and
a hardware decompression device configured to:
receive a memory access read request, the received request causing a cache miss or page fault to occur;
determine whether data requested by the memory access read request is in a compressed memory block within the system memory;
in response to determining that the data requested by the memory access read request is not in a compressed memory block within the system memory, send the memory access read request to the system memory controller to answer the read request;
in response to determining that the data requested by the memory access read request is in a compressed memory block within the system memory, determine whether the data from the compressed memory block has previously been decompressed and stored in an internal buffer within a dedicated storage medium contained within the hardware decompression device that is not the system memory;
in response to determining that the data from the compressed memory block is stored in the internal buffer, return the data from the internal buffer in response to the received memory access read request; and
in response to determining that the data from the compressed memory block is not stored in the internal buffer,
allocate storage space in the internal buffer;
read the data from the compressed memory block;
decompress the read data from the compressed memory block into the internal buffer; and
return the decompressed data from the internal buffer in response to the memory access read request.
15. The system of claim 14 , wherein the at least one processor is further caused to:
read the data from the compressed memory block until the data requested by the memory access read request is located; and
decompress the read data from the compressed memory block into the internal buffer only to the point where the requested data is fully retrieved.
16. The system of claim 14 , wherein the at least one processor is further caused to:
determine whether the data requested by the memory access read request is in the compressed memory block by checking unused bits of page table structures that store information about virtual memory pages.
17. The system of claim 14 , wherein the at least one processor is further caused to:
alternate between a full memory decompression mode and a partial memory decompression mode.
18. The system of claim 17 , wherein the at least one processor is further caused to:
alternate between the full memory decompression mode and the partial memory decompression mode based on one or more performance parameters of a device or program associated with the received memory access read request.
19. The system of claim 17 , wherein the at least one processor is further caused to:
alternate between the full memory decompression mode and the partial memory decompression mode based on a signal received from a device or program associated with the received memory access read request.
20. The system of claim 14 , wherein the hardware decompression device is disposed within a cache controller of the processor.