RFR: 8342662: C2: Add new phase for backend-specific lowering

[Jatin Bhateja]

On Mon, 21 Oct 2024 04:11:03 GMT, Jasmine Karthikeyan <jkarthikeyan at openjdk.org> wrote:

> Hi all,
> This patch adds a new pass to consolidate lowering of complex backend-specific code patterns, such as `MacroLogicV` and the optimization proposed by #21244. Moving these optimizations to backend code can simplify shared code, while also making it easier to develop more in-depth optimizations. The linked bug has an example of a new optimization this could enable. The new phase does GVN to de-duplicate nodes and calls nodes' `Value()` method, but it does not call `Identity()` or `Ideal()` to avoid undoing any changes done during lowering. It also reuses the IGVN worklist to avoid needing to re-create the notification mechanism.
> 
> In this PR only the skeleton code for the pass is added, moving `MacroLogicV` to this system will be done separately in a future patch. Tier 1 tests pass on my linux x64 machine. Feedback on this patch would be greatly appreciated!

src/hotspot/cpu/arm/c2_lowering_arm.cpp line 29:

> 27: #include "opto/phaseX.hpp"
> 28: 
> 29: Node* PhaseLowering::lower_node(Node* in) {

Suggestion:

Node* PhaseLowering::lower_node(Node* n) {

src/hotspot/cpu/ppc/c2_lowering_ppc.cpp line 29:

> 27: #include "opto/phaseX.hpp"
> 28: 
> 29: Node* PhaseLowering::lower_node(Node* in) {

Suggestion:

Node* PhaseLowering::lower_node(Node* n) {

src/hotspot/cpu/riscv/c2_lowering_riscv.cpp line 29:

> 27: #include "opto/phaseX.hpp"
> 28: 
> 29: Node* PhaseLowering::lower_node(Node* in) {

Suggestion:

Node* PhaseLowering::lower_node(Node* n) {

src/hotspot/cpu/s390/c2_lowering_s390.cpp line 29:

> 27: #include "opto/phaseX.hpp"
> 28: 
> 29: Node* PhaseLowering::lower_node(Node* in) {

Suggestion:

Node* PhaseLowering::lower_node(Node* n) {

src/hotspot/cpu/x86/c2_lowering_x86.cpp line 29:

> 27: #include "opto/phaseX.hpp"
> 28: 
> 29: Node* PhaseLowering::lower_node(Node* in) {

Suggestion:

Node* PhaseLowering::lower_node(Node* n) {

src/hotspot/share/opto/compile.cpp line 2466:

> 2464:     print_method(PHASE_BEFORE_LOWERING, 3);
> 2465: 
> 2466:     PhaseLowering lower(&igvn);

Any specific reason to have lowering after loop optimizations ?
Lowered nodes may change the loop body size thereby impacting unrolling decisions.

src/hotspot/share/opto/phaseX.cpp line 2301:

> 2299:   while(_igvn->_worklist.size() != 0) {
> 2300:     Node* n = _igvn->_worklist.pop();
> 2301:     Node* new_node = lower_node(n);

_PhaseLowring::lower_node_ may do complex transformation where by replacing a graph pallet rooted at current node by another pallet. For each newly created node in new pallet, it should make sure to either directly run _igvn.transform, thereby triggering Ideal / Identity / Value sub-passed over it, OR insert the node into _igvn.worklist for lazy processing, in latter case you are consuming entire worklist after running over only Value transforms before existing the lowering phase.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1812113082
PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1812113953
PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1812114516
PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1812118894
PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1812119441
PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1812140992
PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1812110851