Allow custom default address spaces and parse `p-` specifications in the datalayout string #143182

xdoardo · 2025-06-29T10:17:45Z

Some targets, such as CHERI, use as default an address space different from the "normal" default address space 0 (in the case of CHERI, 200 is used). Currently, rustc does not allow to specify custom address spaces and does not take into consideration p- specifications in the datalayout string.

This patch tries to mitigate these problems by allowing targets to define a custom default address space (while keeping the default value to address space 0) and adding the code to parse the p- specifications in rustc_abi. The main changes are that TargetDataLayout now uses functions to refer to pointer-related informations, instead of having specific fields for the size and alignment of pointers in the default address space; furthermore, the two pointer_size and pointer_align fields in TargetDataLayout are replaced with an FxHashMap that holds info for all the possible address spaces, as parsed by the p- specifications.

The potential performance drawbacks of not having ad-hoc fields for the default address space will be tested in this PR's CI run.

r? workingjubilee

rustbot · 2025-06-29T10:17:50Z

workingjubilee is currently at their maximum review capacity.
They may take a while to respond.

rustbot · 2025-06-29T10:17:52Z

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri

These commits modify compiler targets.
(See the Target Tier Policy.)

Some changes occurred in compiler/rustc_ast_lowering/src/format.rs

cc @m-ou-se

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri, @RalfJung, @oli-obk, @lcnr

Some changes occurred in cfg and check-cfg configuration

cc @Urgau

Some changes occurred to the CTFE machinery

cc @RalfJung, @oli-obk, @lcnr

Some changes occurred in compiler/rustc_codegen_ssa

cc @WaffleLapkin

workingjubilee · 2025-06-29T10:28:43Z

@bors2 try @rust-timer queue

rust-bors · 2025-06-29T10:28:47Z

⌛ Trying commit ef6e4ef with merge a6d8fc2…

To cancel the try build, run the command @bors2 try cancel.

Allow custom default address spaces and parse `p-` specifications in the datalayout string Some targets, such as CHERI, use as default an address space different from the "normal" default address space `0` (in the case of CHERI, [200 is used](https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-877.pdf)). Currently, `rustc` does not allow to specify custom address spaces and does not take into consideration [`p-` specifications in the datalayout string](https://llvm.org/docs/LangRef.html#langref-datalayout). This patch tries to mitigate these problems by allowing target to define a custom default address space (while keeping the default value to address space `0`) and adding the code to parse the `p-` specifications in `rustc_abi`. The main changes are that `TargetDataLayout` now uses functions to refer to pointer-related informations, instead of having specific fields for the size and alignment of pointers in the default address space: the potential performance drawbacks of not having ad-hoc fields will be tested in this PR's CI run. r? workingjubilee

rust-bors · 2025-06-29T12:47:13Z

☀️ Try build successful (CI)
Build commit: a6d8fc2 (a6d8fc279459764136b16847f5b4411d4298525d, parent: 5ca574e85b67cec0a6fc3fddfe398cbe676c9c69)

rust-timer · 2025-06-29T14:13:30Z

Finished benchmarking commit (a6d8fc2): comparison URL.

Overall result: ❌ regressions - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	1.0%	[0.7%, 1.7%]	8
Regressions ❌ (secondary)	2.0%	[0.9%, 4.7%]	38
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	1.0%	[0.7%, 1.7%]	8

Max RSS (memory usage)

Results (primary -2.6%, secondary -8.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.6%	[-3.6%, -1.5%]	2
Improvements ✅ (secondary)	-8.0%	[-8.0%, -8.0%]	1
All ❌✅ (primary)	-2.6%	[-3.6%, -1.5%]	2

Cycles

Results (secondary 2.6%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.6%	[2.4%, 2.7%]	3
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 695.013s -> 695.668s (0.09%)
Artifact size: 371.77 MiB -> 372.01 MiB (0.06%)

rustbot · 2025-06-29T17:34:17Z

Some changes occurred in compiler/rustc_codegen_gcc

cc @antoyo, @GuillaumeGomez

rust-analyzer is developed in its own repository. If possible, consider making this change to rust-lang/rust-analyzer instead.

cc @rust-lang/rust-analyzer

Some changes occurred in compiler/rustc_codegen_cranelift

cc @bjorn3

Some changes occurred in src/tools/clippy

cc @rust-lang/clippy

workingjubilee · 2025-06-29T20:27:21Z

@bors2 try

rust-bors · 2025-06-29T20:27:25Z

⌛ Trying commit fce06e3 with merge 92687cf…

To cancel the try build, run the command @bors2 try cancel.

Allow custom default address spaces and parse `p-` specifications in the datalayout string Some targets, such as CHERI, use as default an address space different from the "normal" default address space `0` (in the case of CHERI, [200 is used](https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-877.pdf)). Currently, `rustc` does not allow to specify custom address spaces and does not take into consideration [`p-` specifications in the datalayout string](https://llvm.org/docs/LangRef.html#langref-datalayout). This patch tries to mitigate these problems by allowing targets to define a custom default address space (while keeping the default value to address space `0`) and adding the code to parse the `p-` specifications in `rustc_abi`. The main changes are that `TargetDataLayout` now uses functions to refer to pointer-related informations, instead of having specific fields for the size and alignment of pointers in the default address space; furthermore, the two `pointer_size` and `pointer_align` fields in `TargetDataLayout` are replaced with an `FxHashMap` that holds info for all the possible address spaces, as parsed by the `p-` specifications. The potential performance drawbacks of not having ad-hoc fields for the default address space will be tested in this PR's CI run. r? workingjubilee

rust-bors · 2025-06-29T22:43:09Z

☀️ Try build successful (CI)
Build commit: 92687cf (92687cf86ed4b58afb2653d78a1c735272295f02, parent: ed2d759783dc9de134bbb3f01085b1e6dbf539f3)

workingjubilee · 2025-06-30T06:53:02Z

@rust-timer build 92687cf

rust-timer · 2025-06-30T09:01:27Z

Finished benchmarking commit (92687cf): comparison URL.

Overall result: ❌ regressions - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	0.4%	[0.4%, 0.6%]	4
Regressions ❌ (secondary)	1.1%	[0.3%, 1.9%]	29
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.4%	[0.4%, 0.6%]	4

Max RSS (memory usage)

Results (primary -2.3%, secondary 2.1%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.1%	[2.1%, 2.1%]	1
Improvements ✅ (primary)	-2.3%	[-3.1%, -1.4%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-2.3%	[-3.1%, -1.4%]	2

Cycles

Results (primary -1.1%, secondary 0.3%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	1.9%	[1.9%, 1.9%]	1
Improvements ✅ (primary)	-1.1%	[-1.2%, -0.9%]	2
Improvements ✅ (secondary)	-1.2%	[-1.2%, -1.2%]	1
All ❌✅ (primary)	-1.1%	[-1.2%, -0.9%]	2

Binary size

Results (secondary 0.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.0%	[0.0%, 0.0%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Bootstrap: 694.715s -> 695.8s (0.16%)
Artifact size: 371.77 MiB -> 371.86 MiB (0.02%)

RalfJung · 2025-07-01T16:46:14Z

@bors try
@rust-timer queue

Allow custom default address spaces and parse `p-` specifications in the datalayout string Some targets, such as CHERI, use as default an address space different from the "normal" default address space `0` (in the case of CHERI, [200 is used](https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-877.pdf)). Currently, `rustc` does not allow to specify custom address spaces and does not take into consideration [`p-` specifications in the datalayout string](https://llvm.org/docs/LangRef.html#langref-datalayout). This patch tries to mitigate these problems by allowing targets to define a custom default address space (while keeping the default value to address space `0`) and adding the code to parse the `p-` specifications in `rustc_abi`. The main changes are that `TargetDataLayout` now uses functions to refer to pointer-related informations, instead of having specific fields for the size and alignment of pointers in the default address space; furthermore, the two `pointer_size` and `pointer_align` fields in `TargetDataLayout` are replaced with an `FxHashMap` that holds info for all the possible address spaces, as parsed by the `p-` specifications. The potential performance drawbacks of not having ad-hoc fields for the default address space will be tested in this PR's CI run. r? workingjubilee

bors · 2025-07-01T16:47:27Z

⌛ Trying commit afec259 with merge d439e37...

bors · 2025-07-01T18:55:06Z

☀️ Try build successful - checks-actions
Build commit: d439e37 (d439e37333c92ecce0cd62a651f35235fd04cb18)

rust-timer · 2025-07-01T20:29:03Z

Finished benchmarking commit (d439e37): comparison URL.

Overall result: ✅ improvements - no action needed

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

@bors rollup=never
@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.0%	[-1.1%, -0.9%]	4
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

Results (primary 0.2%, secondary 0.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	2.1%	[1.2%, 3.2%]	3
Regressions ❌ (secondary)	2.8%	[2.7%, 2.9%]	3
Improvements ✅ (primary)	-1.6%	[-2.3%, -1.1%]	3
Improvements ✅ (secondary)	-1.9%	[-3.0%, -0.6%]	3
All ❌✅ (primary)	0.2%	[-2.3%, 3.2%]	6

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 461.796s -> 461.853s (0.01%)
Artifact size: 372.30 MiB -> 371.85 MiB (-0.12%)

arichardson · 2025-07-02T08:02:04Z

compiler/rustc_abi/src/lib.rs

-                    dl.pointer_size = parse_size(s, p)?;
-                    dl.pointer_align = parse_align(a, p)?;
+                [p, s, a @ ..] if p.starts_with("p") => {
+                    // Some targets, such as CHERI, use the 'f' suffix in the p- spec to signal that


I am planning to add more flags upstream for non-integral pointer properties so ignoring any letter for now is great for future proofing.

We generally try to reduce future-proofing in this code: deviation from the LLVM datalayout string will cause rustc to error, quite deliberately.

If we're making rustc correct with respect to address spaces, then if you add a character that would be relevant for how the platform is handled by codegen, why should we ignore it? And if we don't know what it means, why should we assume it is irrelevant?

Yeah, just ignoring things we don't know about is generally bad since we don't know what we are ignoring, so there's a high risk we are ignoring something important.

I will change it so that only p and pf are stripped. Do you think that would make sense? Once more flags are added we can take them into account accordingly.

Yes, that seems fine. I wonder if there's something we should do in response to "pf" but nothing immediately occurs to me.

In CHERI targets the f is used to signal that the pointer is non integral (or fat). Upstream LLVM, in practice, just ignores it and sets to false the according field in the pointer spec.

Another possibility would be to wait for f to be taken into account at all until a target that uses it is introduced.

That's fair and it probably makes sense to wait until such uses exist. I am planning to introduce 'e', 'n', 'u' flags which are basically finger-grained properties for the current ni:<addrspace> (llvm/llvm-project#105735), but to allow building rustc against the CHERI LLVM fork, ignoring 'f' would be quite nice since it doesn't affect the rest of rustc.

ccc2083 removes the stripping of f. Instead, upon seeing a pX specification, an error that tells the user that the given pointer specification is unknown is returned.

arichardson · 2025-07-02T08:07:03Z

compiler/rustc_abi/src/lib.rs

+                        }
+                    }
+                }
+                [p, s, _pr, i, a @ ..] if p.starts_with("p") => {


Suggested change

[p, s, _pr, i, a @ ..] if p.starts_with("p") => {

[p, s, _pr, a, idc @ ..] if p.starts_with("p") => {

Index size is the last component?

Thanks, I missed that. Do you think it should it be, then,

[p, s, a , _pr, i @ ..] if p.starts_with("p") => {

instead?

For reference, the relevant bit from LLVM's datalayout string spec:

p[n]:<size>:<abi>[:<pref>[:<idx>]]

I can also add the pref field to AddressSpaceInfo, which perhaps would make even more sense.

commit ccc2083 adds the parsing of :<pref> with functions to retrieve that value: I set it to default to <abi> when not specified.

I also had to slightly change the utility parse_align function, which took as parameter a &[&str], whose semantics were actually to ignore the potential :<pref>:<more...> following <abi>.

arichardson · 2025-07-02T08:13:27Z

compiler/rustc_abi/src/lib.rs

@@ -1104,7 +1287,7 @@ impl Primitive {
            // FIXME(erikdesjardins): ignoring address space is technically wrong, pointers in
            // different address spaces can have different sizes
            // (but TargetDataLayout doesn't currently parse that part of the DL string)
-            Pointer(_) => dl.pointer_size,
+            Pointer(a) => dl.pointer_size_in(a),


Seems like this Todo can be removed now?

arichardson · 2025-07-02T08:13:38Z

compiler/rustc_abi/src/lib.rs

@@ -1118,7 +1301,7 @@ impl Primitive {
            // FIXME(erikdesjardins): ignoring address space is technically wrong, pointers in
            // different address spaces can have different alignments
            // (but TargetDataLayout doesn't currently parse that part of the DL string)
-            Pointer(_) => dl.pointer_align,
+            Pointer(a) => dl.pointer_align_in(a),


Remove Todo?

arichardson · 2025-07-02T08:15:08Z

compiler/rustc_abi/src/lib.rs

@@ -298,6 +323,7 @@ impl TargetDataLayout {
    /// determined from llvm string.
    pub fn parse_from_llvm_datalayout_string<'a>(
        input: &'a str,
+        default_address_space: AddressSpace,


IMO this should be part of the data layout but currently none of alloca/globals/program AS can be used for this, so passing it in seems reasonable.

…default data address space

…arse preferred alignment spec in `p-`

rustbot · 2025-07-03T15:44:23Z

Some changes occurred in coverage instrumentation.

cc @Zalathar

rustbot assigned workingjubilee Jun 29, 2025