Request for reviews(M): 7193318: C2: remove number of inputs requirement from Node's new operator (original) (raw)

Christian Thalinger christian.thalinger at oracle.com
Mon Sep 24 18:12:49 PDT 2012

On Sep 24, 2012, at 12:04 PM, Christian Thalinger <christian.thalinger at oracle.com> wrote:

On Sep 24, 2012, at 11:57 AM, Christian Thalinger <christian.thalinger at oracle.com> wrote:

On Sep 24, 2012, at 11:09 AM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote: I regenerated webrev using Tom's webrev script so that udiffs are more clear:

http://cr.openjdk.java.net/~kvn/7193318/webrev/ Looks great. Thanks for doing this cleanup! One minor question/suggestion: do we want to replace phase->C and friends with a local C variable? - mulhi = phase->transform(new (phase->C) ConvL2INode(mulhi)); + mulhi = phase->transform(new (C) ConvL2INode(mulhi)); Would make the code more compact and more readable.

...but it would be a too big change to integrate it into this one. Let's file another RFE for that.

-- Chris

-- Chris

-- Chris

Typo "necassary": + // Allocate memory for the necassary number of edges. We do such alignment for sparc only. See comment in allocation.hpp. Please, change question to statement: + // Do we really need to allocate space for in array to have + // double alignment? May be we do since we are casting it to be + // Node**. Otherwise looks good. thanks, Vladimir Bharadwaj Yadavalli wrote: De-couple memory allocation for edges from memory allocation for Node object. This allows use of placement new operator of Node, viz., new(sizet, Compile *) instead of node(sizet, Compile *, int) thereby eliminating the need to specify the number of edges in the new operator. Deleted placement new operator of Node - node(sizet, Compile *, int). Testing done: jtreg, and JPRT Benchmarking: refworkload. Thanks, Bharadwaj

More information about the hotspot-compiler-dev mailing list