Agner`s CPU blog

Software optimization resources | E-mail subscription to this blog | www.agner.org

List Messageboards

Whole-function vectorization and conditionals

Author: Sylvain Collange

Date: 2016-08-16 10:30

Correction: I meant WAW dependencies in my previous messages, not WAR dependencies.

Agner wrote:

Could you have two separate sets of regiser read ports, one that is wired to the address calculation stage in the pipeline (general purpose registers only), and one that is wired to the execution stage?

If both sets of ports access the same physical array of registers, then the RF component still needs m+n ports, regardless of where its ports are connected to.
If you replicate the array itself to make two copies of the register file, then yes, it works: you get twice as many read ports for about twice the area. It does not work for write ports, though: you still need to perform all writebacks to both register files to keep their data coherent. Then another complication is the bypass network, which does not scale well with the number of ports either.

You may want to take advantage of the fact that mask registers are a subset of general-purpose registers to replicate only part of the array. It would work on an in-order processor, but as soon as you rename registers this property does not hold any more.

Interesting. I wonder where you have all this information from?

Unfortunately, academic papers seldom go down to that level of detail, so you have to microbenchmark, speculate, and read patents.
For register renaming, here are the classic ones from the P6/K7 era:
https://www.google.com/patents/US6047369
https://www.google.com/patents/US5632023
Even on these 20 year-old architectures, you can see the renaming techniques are quite involved. (Thanks to the infamous INC/DEC instructions that only update a subset of the Flags register...)

No. Without zeroing, the mask chooses between the result and the first input operand, which is not the destination operand.

Now I get it, thanks! Indeed, you need no extra input since you will be reading this register anyway.
I was expecting the semantics of masked instructions to be "pretend nothing happens on lanes with a 0 bit mask". In your ISA even 0-masked lanes have the side effect of a register move. This is unusual, but why not... Wouldn't that complicate register allocation, though? Do you have code examples showing how a compiler can perform if-conversion from some arbitrary code?

Also are there instruction encodings to chose between the result and the second operand (e.g. v0 = v5 ? v1 + v3 * v4 : v3; for a conditional Horner step) ?

Reply To This Message

Previous Message

Proposal for instruction set - now on Github new - Agner - 2016-06-26

Proposal for instruction set - now on Github new - Joe Duarte - 2016-07-04

Proposal for instruction set - now on Github new - Agner - 2016-07-04

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-07-06

Proposal for instruction set - now on Github new - Agner - 2016-07-06

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-07-07

Proposal for instruction set - now on Github new - Agner - 2016-07-07

Whole-function vectorization and conditionals new - Sylvain Collange - 2016-08-15

Whole-function vectorization and conditionals new - Agner - 2016-08-15

Whole-function vectorization and conditionals new - Sylvain Collange - 2016-08-15

Whole-function vectorization and conditionals new - Agner - 2016-08-15

Whole-function vectorization and conditionals new - Sylvain Collange - 2016-08-15

Whole-function vectorization and conditionals new - Agner - 2016-08-15

Number of input dependencies new - Agner - 2016-08-16

Whole-function vectorization and conditionals - Sylvain Collange - 2016-08-16

Whole-function vectorization and conditionals new - Agner - 2016-08-17

Merging with first operand new - Sylvain Collange - 2016-08-18

Merging with first operand new - Agner - 2016-08-19

SIMD exceptions are fine with masking new - Sylvain Collange - 2016-08-19

SIMD exceptions are fine with masking new - Agner - 2016-08-20

SIMD exceptions are fine with masking new - Hubert Lamontagne - 2016-08-20

SIMD exceptions are fine with masking new - Sylvain Collange - 2016-08-25

Merging with first operand new - Hubert Lamontagne - 2016-08-19

Proposal for instruction set - now on Github new - Joe Duarte - 2016-08-17

Proposal for instruction set - now on Github new - Agner - 2016-08-18

Proposal for instruction set - now on Github new - Joe Duarte - 2016-08-31

Proposal for instruction set - now on Github new - Agner - 2016-08-31

Proposal for instruction set - now on Github new - Jorcy Neto - 2016-09-01

Proposal for instruction set - now on Github new - Yuhong Bao - 2016-07-12

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-07-12

Things from MIPS (and novel things) new - Anonymous - 2016-07-28

Things from MIPS (and novel things) new - Agner - 2016-07-28

Things from MIPS (and novel things) new - Hubert Lamontagne - 2016-07-28

Matrix multiplication new - Agner - 2016-07-29

Matrix multiplication new - Hubert Lamontagne - 2016-07-29

Matrix multiplication new - John D. McCalpin - 2016-07-29

Matrix multiplication new - Agner - 2016-07-29

Introduction website new - Agner - 2016-08-01

Introduction website new - EricTL - 2017-07-17

Introduction website new - Agner - 2017-07-18

Introduction website new - EricTL - 2017-07-20

Introduction website new - Agner - 2017-07-20

Proposal for instruction set - now on Github new - Joe Duarte - 2016-08-04

Proposal for instruction set - now on Github new - Agner - 2016-08-04

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-08-05

Proposal for instruction set - now on Github new - Agner - 2016-08-06

Proposal for instruction set - now on Github new - fanoI - 2016-08-08

Proposal for instruction set - now on Github new - Agner - 2016-08-08

Proposal for instruction set - now on Github new - fanoI - 2016-08-09

Proposal for instruction set - now on Github new - Joe Duarte - 2016-08-08

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-08-09

Proposal for instruction set - now on Github new - Joe Duarte - 2016-08-11

Proposal for instruction set - now on Github new - Agner - 2016-08-12

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-08-12

Proposal for instruction set - now on Github new - grant galitz - 2016-08-22

Proposal for instruction set - now on Github new - Agner - 2016-08-22

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-08-24

ARM with scalable vector extensions new - Agner - 2016-08-23

ARM with scalable vector extensions new - Jorcy Neto - 2016-08-23

ARM with scalable vector extensions new - Hubert Lamontagne - 2016-08-26

ARM with scalable vector extensions new - Jorcy Neto - 2016-12-20

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-09-05

Proposal for instruction set - now on Github new - Agner - 2016-09-05

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-09-05

Proposal for instruction set - now on Github new - Agner - 2016-09-06

Proposal for instruction set - now on Github new - Bigos - 2016-09-06

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-09-06

Proposal for instruction set - now on Github new - Agner - 2016-09-07

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-09-07

Proposal for instruction set - now on Github new - Agner - 2016-09-08

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2016-09-08

Proposal for instruction set - now on Github new - Commenter - 2016-09-07

Proposal for instruction set - now on Github new - Bigos - 2016-09-08

Paging new - Kurt Baumgardner - 2016-09-09

Paging new - Agner - 2016-09-10

Paging new - Hubert Lamontagne - 2016-09-11

Paging new - Kurt Baumgardner - 2016-09-13

Paging new - Agner - 2016-09-13

Paging new - Kurt Baumgardner - 2016-09-13

Paging new - Hubert Lamontagne - 2016-09-13

Paging new - Kurt Baumgardner - 2016-09-14

Paging new - Hubert Lamontagne - 2016-09-11

Paging new - Kurt Baumgardner - 2016-09-13

Paging new - Agner - 2016-09-14

Paging new - Jorcy Neto - 2016-09-18

A null register? new - csdt - 2016-09-23

A null register? new - Agner - 2016-09-24

A null register? new - Hubert Lamontagne - 2016-09-24

A null register? new - csdt - 2016-09-26

A null register? new - Agner - 2016-09-27

Indexed registers new - Kurt Baumgardner - 2016-09-26

Indexed registers new - Agner - 2016-09-27

Indexed registers new - Kurt Baumgardner - 2016-09-27

Indexed registers new - Agner - 2016-09-28

Indexed registers new - Hubert Lamontagne - 2016-09-28

Indexed registers new - Kurt Baumgardner - 2016-10-03

Indexed registers new - Agner - 2016-10-03

Indexed registers new - Hubert Lamontagne - 2016-10-04

Bilinear Interpolation new - Hubert Lamontagne - 2016-10-28

Bilinear Interpolation new - Agner - 2016-10-29

Bilinear Interpolation new - Hubert Lamontagne - 2016-10-29

Bilinear Interpolation new - Agner - 2016-10-30

Bilinear Interpolation new - Hubert Lamontagne - 2016-10-30

ForwardCom version 1.04 new - Agner - 2016-12-08

ForwardCom version 1.04 new - Matthias Bentrup - 2016-12-12

ForwardCom version 1.04 new - Agner - 2016-12-12

ForwardCom version 1.04 new - Matthias Bentrup - 2016-12-14

Async system calls; horizontal packing instruction new - Joe Duarte - 2016-12-14

Async system calls; horizontal packing instruction new - Agner - 2016-12-15

Comparison of instruction sets new - Agner - 2016-12-17

Comparison of instruction sets new - Joe Duarte - 2016-12-28

Comparison of instruction sets new - Agner - 2016-12-29

Comparison of instruction sets new - Hubert Lamontagne - 2016-12-30

Comparison of instruction sets new - Hubert Lamontagne - 2017-01-05

ForwardCom version 1.05 new - Agner - 2017-01-22

Syscall/ISR acceleration new - Jonathan Brandmeyer - 2017-01-22

Syscall/ISR acceleration new - Agner - 2017-01-23

Syscall/ISR acceleration new - Jonathan Brandmeyer - 2017-01-25

Syscall/ISR acceleration new - Agner - 2017-01-25

ForwardCom version 1.05 new - JiÅ™Ã Moravec - 2017-01-23

ForwardCom version 1.05 new - Agner - 2017-01-24

Jump prefetch? new - csdt - 2017-01-27

Jump prefetch? new - Agner - 2017-01-27

Jump prefetch? new - csdt - 2017-01-30

Jump prefetch? new - Agner - 2017-01-30

Jump prefetch? new - csdt - 2017-01-30

Jump prefetch? new - Agner - 2017-01-31

Jump prefetch? new - csdt - 2017-01-31

Jump prefetch? new - Hubert Lamontagne - 2017-02-01

Jump prefetch? new - Agner - 2017-02-01

Jump prefetch? new - Hubert Lamontagne - 2017-02-01

Jump prefetch? new - Agner - 2017-02-02

Jump prefetch? new - Agner - 2017-02-14

Jump prefetch? new - Hubert Lamontagne - 2017-01-31

High precision arithmetic new - fanoI - 2017-03-21

High precision arithmetic new - Agner - 2017-03-21

Intel's Control-flow Enforcement Technology new - Joe Duarte - 2017-04-13

Intel's Control-flow Enforcement Technology new - Agner - 2017-04-14

Proposal for instruction set - now on Github new - Agner - 2017-04-27

Assembler with metaprogramming features new - Agner - 2017-07-27

Assembler with metaprogramming features new - Kai Rese - 2017-08-11

Assembler with metaprogramming features new - Agner - 2017-08-11

Assembler with metaprogramming features new - Kai Rese - 2017-08-14

Assembler with metaprogramming features new - Agner - 2017-08-14

Assembler with metaprogramming features new - Kai Rese - 2017-08-15

Number of register file ports in implementations new - Hubert Lamontagne - 2017-08-22

Number of register file ports in implementations new - Agner - 2017-08-23

Number of register file ports in implementations new - Hubert Lamontagne - 2017-08-27

Number of register file ports in implementations new - Agner - 2017-08-28

Number of register file ports in implementations new - Bigos - 2017-08-28

Number of register file ports in implementations new - Hubert Lamontagne - 2017-08-28

Proposal for instruction set - now on Github new - yeengief - 2017-09-20

Proposal for instruction set - now on Github new - Agner - 2017-09-20

Proposal for instruction set - now on Github new - yeengief - 2017-09-20

Proposal for instruction set - now on Github new - Agner - 2017-09-20

Proposal for instruction set - now on Github new - yeengief - 2017-09-21

Proposal for instruction set - now on Github new - Agner - 2017-09-21

Proposal for instruction set - now on Github new - yeengief - 2017-09-21

Proposal for instruction set - now on Github new - Agner - 2017-09-23

Proposal for instruction set - now on Github new - - - 2017-09-22

Proposal for instruction set - now on Github new - Agner - 2017-09-23

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2017-09-25

Proposal for instruction set - now on Github new - Agner - 2017-09-26

Proposal for instruction set - now on Github new - Hubert Lamontagne - 2017-09-26

New assembler, new version, new forum new - Agner - 2017-11-03

List Messageboards