This is a CL that aims to do a general cleanup of DecompressionElimination
to make it easier for devs to look at it, and to create new test cases.
Combined direct decompression & compression tests since they can be
summarized with a for loop in just one test that tries out
all the combinations.
Also created 'global' accesses to stop repeating them in every test.
Same for compression and decompression ops.
Added EXPECT in test cases that had none.
Added dots after comments.
Variables now use underscore instead of camelCase.
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703, v8:9183
Change-Id: I38a5c6549e0b4ff89c3271ead23b626e8b6b4843
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1628788
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Commit-Queue: Ross McIlroy <rmcilroy@chromium.org>
Auto-Submit: Santiago Aboy Solanes <solanes@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62345}
We previously only optimized cases like
Parent <- Decompression <- Compression <- Child
to
Parent <- Child
This CL also adds the complementary optimization, namely, it reduces
Parent <- Compression <- Decompression <- Child
as above.
Such a cases became apparent after a recent extension of CSA load elimination (see https://chromium-review.googlesource.com/c/v8/v8/+/1660626), breaking a load elimination test case and thus the pointer compression build.
R=jarin@chromium.org, solanes@chromium.org
Change-Id: Ic730d05175f214e7055f94704141744ca44fefe5
Bug: v8:9353
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1664070
Commit-Queue: Georg Schmid <gsps@google.com>
Reviewed-by: Michael Stanton <mvstanton@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62246}
Currently, decompression elimination may reduce phis by pushing decompressions in the value inputs of the phi "down" and replacing it by a single decompression following the phi node. Because of the way that the replacement is currently done, other reducers in the same phase will not generally get a chance to revisit the modified phi.
In the specific case of v8:9335 this blocked an additional optimization in CommonOperatorReducer from being applied, causing the overall load elimination test to fail.
This CL fixes the replacement behavior in decompression elimination to also allow for revisitations of the modified phi node.
Bug: v8:9335 v8:9336
Change-Id: I3ca5686dacb41a525160b08456905ba77cf28b39
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Change-Id: I3ca5686dacb41a525160b08456905ba77cf28b39
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1648238
Reviewed-by: Santiago Aboy Solanes <solanes@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Commit-Queue: Georg Schmid <gsps@google.com>
Cr-Commit-Position: refs/heads/master@{#62054}
CompressedHeapConstant is used in the DecompressionElimination Reducer to
create compressed HeapConstant values. It won't appear in the graph
up until that point.
This CL enables back the disabled tests in DecompressionElimination, as
well as generating the CompressedHeapConstant in that reducer.
The RelocInfo has already been added for x64 but not for arm64. Therefore,
the x64 version is now doing the mov on 32 bits. The support for ARM will
come in a following CL, and for now it is doing the mov in 64 bits.
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703, v8:9298
Change-Id: If0ca4f937cfa60501679e66f6fd5ded2df38f605
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1632236
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61950}
There were some cases that were not yet contemplated on machine graph
verifier.
Also, there is some work to be done to create a Compressed HeapConstant.
Until that happens, we have to ignore HeapConstants for
DecompressionElimination's reductions.
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703, v8:9298
Change-Id: I9de8dd4272866830807a8d88e625e863fb5f1d0d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1632209
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61892}
Reduces compressions of constants, going from
Constant <- Compress <- Child
to
Compressed_Constant <- Child
This pattern commonly appeared when the Constant was being used as a
Store value (e.g StoreElement's value).
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703
Change-Id: I2e71f79ddd2a6fba42cdfe782cc89fff3a8d5ac1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1627988
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61887}
This reduction replaces the Phi's input decompressions with their parent
node, if and only if all of the Phi's inputs are Decompress nodes.
Also, if we have different Decompress nodes as inputs, we need to use
a conservative decompression after the Phi.
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703
Change-Id: I8cc0264f9d08fe5ad25364f18c9f305afc54529c
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1624785
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61820}
Adding the case of having a decompress and a constant (both HeapConstant
and Int64Constant).
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703
Change-Id: Ic9475e9762575e7f6ca2937d832638f7c9897e1d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1613253
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61736}
This is a border case that appeared in one of our tests. It seems a good
idea to test for it. It does not seem to be common enough to handle it
specifically (i.e replacing the Word64Equal with a true bool constant)
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703
Change-Id: I27e85a2493ab03b92591e788a185d6109904eef1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1613252
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61726}
Added the comparison of two decompresses to the DecompressionElimination
AdvancedReducer. Note that it works in the case that lhs is equal to rhs.
Also added tests for its implementation.
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703
Change-Id: I21676a78b592859692768c3499ea11117d3bb5a5
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1609793
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61647}
After introducing explicit compress and decompress nodes for each load
and store we are left with many nodes that are redundant.
This CL aims to eliminate redundant decompressions and compressions in the
"direct decompression & compression" category.
Also added tests to test the new reducer.
Cq-Include-Trybots: luci.v8.try:v8_linux64_pointer_compression_rel_ng
Cq-Include-Trybots: luci.v8.try:v8_linux64_arm64_pointer_compression_rel_ng
Bug: v8:8977, v8:7703
Change-Id: I93e024d13af34d484086b7983f379265d16ac154
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1602702
Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Commit-Queue: Santiago Aboy Solanes <solanes@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61473}
