Coalescing adjacent gather/scatter operations

Patent metadata

Field	Value
Publication number	US-11599362-B2
Application number	US-202117316680-A
Country	US
Kind code	B2
Filing date	May 10, 2021
Priority date	Dec 26, 2012
Publication date	Mar 7, 2023
Grant date	Mar 7, 2023

Abstract

According to one embodiment, a processor includes an instruction decoder to decode a first instruction to gather data elements from memory, the first instruction having a first operand specifying a first storage location and a second operand specifying a first memory address storing a plurality of data elements. The processor further includes an execution unit coupled to the instruction decoder, in response to the first instruction, to read contiguous a first and a second of the data elements from a memory location based on the first memory address indicated by the second operand, and to store the first data element in a first entry of the first storage location and a second data element in a second entry of a second storage location corresponding to the first entry of the first storage location.

First claim

What is claimed is:

1. A processor comprising:
a data cache to cache data;
an instruction cache to cache instructions;
a decode unit coupled to the instruction cache, the decode unit to decode the instructions, including a first instruction, the first instruction having a first field to specify a first destination vector register, the first instruction having a second field to specify a 64-bit general-purpose register as a source of a base address, the first instruction having a third field to specify a mask register as a source of a mask, and the first instruction to indicate a data element width of 64-bits; and
an execution unit coupled to the decode unit, coupled to the first destination vector register, and coupled to the 64-bit general-purpose register, the execution unit based on the first instruction to:
for mask bits of the mask that are one:
load a corresponding structure from a memory based on the base address, the structure to include, as contiguous elements in the memory, a first 64-bit data element, a second 64-bit data element, and a third 64-bit data element; and
store the first 64-bit data element in a 64-bit data element position of the first destination vector register corresponding to the mask bit, the second 64-bit data element in a 64-bit data element position of a second destination vector register corresponding to the mask bit, and the third 64-bit data element in a 64-bit data element position of a third destination vector register corresponding to the mask bit; and

for mask bits of the mask that are zero, not load the corresponding structure from the memory.

2. The processor of claim 1 , wherein the first, second, and third destination vector registers are a sequence of registers.

3. The processor of claim 1 , wherein the first instruction has a data element width field to indicate the data element width of 64-bits.

4. The processor of claim 1 , further comprising a plurality of mask registers including the mask register.

5. The processor of claim 1 , wherein the mask register is a 16-bit register, and wherein the first destination register comprises at least 128-bits.

6. The processor of claim 1 , wherein the third field is a three-bit field to specify the mask register.

7. The processor of claim 1 , wherein the mask register is one of a set of registers, the set of registers including a plurality of mask registers and a given register, the processor not supporting use of the given register for masking.

8. The processor of claim 1 , wherein the instruction indicates that the mask register is to be used for zeroing-masking in which values in the 64-bit data element positions of the first, second, and third destination vector registers corresponding to the mask bits that are zero are to be zeroed.

9. The processor of claim 1 , wherein the decode unit is to decode at least one instruction of the instructions having a field to specify the mask register and indicating that the mask register is to be used for merging-masking in which values in a destination vector register corresponding to mask bits that are zero are to be preserved.

10. The processor of claim 1 , wherein not all bits of the mask register are utilized by the instruction.

11. The processor of claim 1 , wherein for the mask bits of the mask that are zero, the processor is to store values of zero to the 64-bit data element position of the first destination vector register, the 64-bit data element position of the second destination vector register, and the 64-bit data element position of the third destination vector register.

12. The processor of claim 1 , further comprising an instruction fetch unit coupled to the instruction cache to fetch the instructions, and wherein the processor has a reduced instruction set computing (RISC) architecture.

13. A method comprising:
caching data in a data cache;
caching instructions in an instruction cache;
decoding the instructions, including a first instruction, the first instruction having a first field specifying a first destination vector register, the first instruction having a second field specifying a 64-bit general-purpose register as a source of a base address, the first instruction having a third field specifying a mask register as a source of a mask, and the first instruction indicating a data element width of 64-bits; and
executing the first instruction, including:
for mask bits of the mask that are one:
loading a corresponding structure from a memory based on the base address, the structure including, as contiguous elements in the memory, a first 64-bit data element, a second 64-bit data element, and a third 64-bit data element; and
storing the first 64-bit data element in a 64-bit data element position of the first destination vector register corresponding to the mask bit, the second 64-bit data element in a 64-bit data element position of a second destination vector register corresponding to the mask bit, and the third 64-bit data element in a 64-bit data element position of a third destination vector register corresponding to the mask bit; and

for mask bits of the mask that are zero, not loading the corresponding structure from the memory.

14. The method of claim 13 , wherein the first, second, and third destination vector registers are a sequence of registers, and wherein the first instruction has a data element width field to indicate the data element width of 64-bits, wherein the mask register is a 16-bit register, and wherein the first destination register comprises at least 128-bits, and wherein the third field is a three-bit field to specify the mask register.

15. The method of claim 13 , wherein the mask register is one of a set of registers, the set of registers including a plurality of mask registers and a given register, the given register not configured to perform masking.

16. The method of claim 13 , wherein not all bits of the mask register are utilized by the instruction, and wherein for the mask bits of the mask that are zero, the processor is to store values of zero to the 64-bit data element position of the first destination vector register, the 64-bit data element position of the second destination vector register, and the 64-bit data element position of the third destination vector register.

17. A system comprising:
a system memory; and
a processor coupled with the system memory, the processor comprising:
a data cache to cache data;
an instruction cache to cache instructions;
a decode unit coupled to the instruction cache, the decode unit to decode the instructions, including a first instruction, the first instruction having a first field to specify a first destination vector register, the first instruction having a second field to specify a 64-bit general-purpose register as a source of a base address, the first instruction having a third field to specify a mask register as a source of a mask, and the first instruction to indicate a data element width of 64-bits; and
an execution unit coupled to the decode unit, coupled to the first destination vector register, and coupled to the 64-bit general-purpose register, the execution unit based on the first instruction to:
for mask bits of the mask that are one:
load a corresponding structure from a memory based on the base address, the structure to include, as contiguous elements in the memory, a first 64-bit data element, a second 64-bit data element, and a third 64-bit data element; and
store the first 64-bit data element in a 64-bit data element position of the first destination vector register corresponding to the mask bit, the second 64-bit data element

Assignees

• Intel Corp

Inventors

• Forsyth Andrew T
• Hickmann Brian J
• Hall Jonathan C
• Hughes Christopher J

Classifications

• G06F9/3887

  controlled by a single instruction for multiple data lanes [SIMD] · CPC title

• G06F9/30038

  using a mask · CPC title

• G06F9/3888

  controlled by a single instruction for multiple threads [SIMT] in parallel · CPC title

• G06F9/30018 Primary

  Bit or string instructions · CPC title

• G06F9/30036

  Instructions to perform operations on packed data, e.g. vector, tile or matrix operations · CPC title