Implement INumber, IFloatingPoint, and the modulus operator for the IEEE 754 decimal types by tannergooding · Pull Request #130890 · dotnet/runtime

tannergooding · 2026-07-16T17:21:05Z

Continues the incremental buildout of the IEEE 754 decimal types (part of #81376). #130807 added the conversion surface and INumberBase<TSelf>; this one adds the next set of interfaces and the modulus operator.

Adds to Decimal32, Decimal64, and Decimal128:

IFloatingPointConstants<TSelf> -- E, Pi, and Tau, correctly rounded to each format's precision.
ISignedNumber<TSelf> -- NegativeOne.
INumber<TSelf> -- Max/Min/MaxNumber/MinNumber/Clamp/Sign/CopySign and friends, mirroring System.Half's semantics (IEEE 2019 minimum/maximum cohort/signed-zero/NaN handling).
IFloatingPoint<TSelf> -- Round (all overloads), the GetExponentByteCount/GetSignificandByteCount/TryWrite* bit APIs, and the remaining classification members.
The modulus operator (%, IModulusOperators<TSelf, TSelf, TSelf>).

The % operator follows the same truncated-remainder (fmod) semantics as float/double/Half -- x - Truncate(x / y) * y -- which is distinct from IEEE754Remainder (round-to-nearest quotient); the latter arrives with IFloatingPointIeee754.

Non-trivial members are ported from the Intel BID reference and validated bit-exact against its reference vectors (modulus and round), plus the existing Python-oracle-generated coverage. The multi-argument Round(value, digits, mode) overloads are the one bespoke piece -- Intel has no direct baseline for them, so they are built on the faithful single-argument rounding primitive.

IFloatingPointIeee754<TSelf>/IDecimalFloatingPointIeee754<TSelf> (and the transcendental sub-interfaces) are deferred to a follow-up that stacks on this.

Note

This PR description was drafted by GitHub Copilot.

E, Pi, and Tau already exist as public constants; this just declares the interface on Decimal32/64/128. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

NegativeOne already exists as a public constant; this just declares the interface on Decimal32/64/128. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Implements the truncated remainder (x - Truncate(x / y) * y) matching the C# floating-point % operator on double/float/Half, not the round-to-nearest IEEE 754 remainder. The result carries the dividend's sign and is computed exactly at the preferred exponent min(exp(x), exp(y)) by reducing the dividend coefficient modulo the divisor coefficient with word-level integer arithmetic. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

All INumber members (Clamp/Max/Min/Sign/CopySign and friends) already exist as public static members; the modulus operator and relational comparison operators satisfy the remaining IModulusOperators/IComparisonOperators requirements, so this declares the interface on Decimal32/64/128. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Collapses the IFloatingPointConstants/INumber/INumberBase/ISignedNumber declarations into IFloatingPoint<TSelf> on Decimal32/64/128 and adds the member set: Ceiling/Floor/Truncate/Round, ConvertToInteger(Native), and the IFloatingPoint exponent/significand writers. Exponent is exposed as int (its UnbiasedExponent type) so NaN/Inf exponent fields don't overflow a narrower type; the significand is the full unpacked coefficient. Round shares Number.RoundDecimalIeee754. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Covers Round (all midpoint modes, curated NaN/Inf/signed-zero/tie edge cases plus oracle-derived random vectors), operator %, the Ceiling/Floor/Truncate/ Round convenience overloads, and the exponent/significand writers + ConvertToInteger. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…tel reference vectors Wires bidNN_fmod and bidNN_round_integral_* from the Intel readtest.in vectors into the existing DecimalIeee754IntelTestData harness. The round vectors surfaced two canonicalization gaps in RoundDecimalIeee754 (non-canonical finite and infinity operands were returned verbatim), now fixed in the implementation. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

azure-pipelines · 2026-07-16T17:21:46Z

Azure Pipelines: Successfully started running 3 pipeline(s). 12 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run.

dotnet-policy-service · 2026-07-16T17:22:12Z

Tagging subscribers to this area: @dotnet/area-system-numerics
See info in area-owners.md if you want to be subscribed.

azure-pipelines · 2026-07-16T17:25:42Z

Azure Pipelines: Successfully started running 3 pipeline(s). 12 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run.

Copilot

Pull request overview

This PR expands the IEEE 754 decimal floating-point types (System.Numerics.Decimal32, Decimal64, Decimal128) with additional generic-math surface (notably IFloatingPoint-related members) and implements the % (truncating remainder / fmod) operator, with accompanying runtime and reference-vector-based tests.

Changes:

Add % operator support by introducing Number.RemainderDecimalIeee754 and wiring it into Decimal32/64/128.
Implement IFloatingPoint<TSelf> members (rounding helpers, ConvertToInteger*, and exponent/significand byte-writing APIs) for Decimal32/64/128 using new Number.RoundDecimalIeee754.
Extend tests and Intel reference-vector enumerators to cover modulus and integral rounding behavior, plus the new exponent/significand bit APIs.

Show a summary per file

File	Description
src/libraries/System.Private.CoreLib/src/System/Number.DecimalIeee754.cs	Adds shared helpers for truncated remainder (`%`) and rounding to fractional digits used by decimal IEEE types.
src/libraries/System.Private.CoreLib/src/System/Numerics/Decimal32.cs	Adds `%` operator and `IFloatingPoint<Decimal32>` implementations (Round/Ceiling/Floor/Truncate + exponent/significand write APIs).
src/libraries/System.Private.CoreLib/src/System/Numerics/Decimal64.cs	Adds `%` operator and `IFloatingPoint<Decimal64>` implementations (Round/Ceiling/Floor/Truncate + exponent/significand write APIs).
src/libraries/System.Private.CoreLib/src/System/Numerics/Decimal128.cs	Adds `%` operator and `IFloatingPoint<Decimal128>` implementations (Round/Ceiling/Floor/Truncate + exponent/significand write APIs).
src/libraries/System.Runtime/ref/System.Runtime.cs	Updates public ref surface for `Decimal32/64/128` to include the new interfaces and members (including `%` and rounding APIs).
src/libraries/System.Runtime/tests/System.Runtime.Tests/System/DecimalIeee754IntelTestData.cs	Adds Intel readtest.in operation families for `fmod` and integral-rounding vectors.
src/libraries/System.Runtime/tests/System.Runtime.Tests/System/Decimal32Tests.cs	Adds modulus, round-to-digits, convenience overload, and exponent/significand write API tests (plus Intel vector theories).
src/libraries/System.Runtime/tests/System.Runtime.Tests/System/Decimal64Tests.cs	Adds modulus, round-to-digits, convenience overload, and exponent/significand write API tests (plus Intel vector theories).
src/libraries/System.Runtime/tests/System.Runtime.Tests/System/Decimal128Tests.cs	Adds modulus, round-to-digits, convenience overload, and exponent/significand write API tests (plus Intel vector theories).

Copilot's findings

Files reviewed: 9/9 changed files
Comments generated: 2

The reduction loop scaled the running remainder one decimal digit at a time, so a large exponent gap spun far more iterations than necessary and kept going after the remainder reached zero. Fold as many trailing zeros per step as the running product can hold without overflowing the integer width (up to the largest cached power of ten) and stop once the remainder is zero. A small divisor now reaches the full cached power; output is bit-exact-unchanged against the Intel reference vectors. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Comments suppressed due to low confidence (1)

src/libraries/System.Runtime/ref/System.Runtime.cs:11436

This PR adds new public API surface (e.g., Decimal32/64/128 now implement additional generic-math interfaces and expose new members like %/Round/Ceiling/Floor/Truncate/bit APIs). Issue #81376 is labeled api-approved, but I couldn’t locate the “approved API shape” comment (code-fenced C# snippet) required by the API-approval procedure. Please link the specific approval comment (or add it) so reviewers can verify the ref changes match the approved shape.

Files reviewed: 9/9 changed files
Comments generated: 4

…est files Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…4 decimal types Covers GetExponentShortestBitLength (both exponent-sign branches), GetSignificandBitLength, and the big-endian significand and exponent writers. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Ceiling/Floor/Round/Truncate propagated a NaN operand's bits unchanged, so a signaling or out-of-range-payload NaN escaped as a non-canonical result. Route it through PropagateNaN, matching the binary operators. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

PranavSenthilnathan · 2026-07-16T19:25:17Z

LGTM after null propagation is fixed

Replace the manual divide and multiply-subtract with TValue.DivRem, matching the remainder computation used elsewhere in the file. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…amily IEEE 754-2019 5.1 requires general-computational operations to produce canonical results, so a returned NaN operand must be re-canonicalized. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 9/9 changed files
Comments generated: 1

tannergooding · 2026-07-16T20:56:14Z

@PranavSenthilnathan should all be handled. Still needs explicit sign-off

tannergooding · 2026-07-17T03:55:41Z

/ba-g the known net481 build break and an unrelated timeout

tannergooding and others added 7 commits July 16, 2026 10:16

Implement IFloatingPointConstants for the IEEE 754 decimal types

40a369e

E, Pi, and Tau already exist as public constants; this just declares the interface on Decimal32/64/128. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Implement ISignedNumber for the IEEE 754 decimal types

1549546

NegativeOne already exists as a public constant; this just declares the interface on Decimal32/64/128. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings July 16, 2026 17:21

tannergooding added the area-System.Numerics label Jul 16, 2026

tannergooding self-assigned this Jul 16, 2026

Copilot started reviewing on behalf of tannergooding July 16, 2026 17:21 View session

tannergooding marked this pull request as ready for review July 16, 2026 17:25

Copilot AI reviewed Jul 16, 2026

View reviewed changes

Comment thread src/libraries/System.Private.CoreLib/src/System/Number.DecimalIeee754.cs Outdated

Comment thread src/libraries/System.Runtime/ref/System.Runtime.cs

Copilot AI review requested due to automatic review settings July 16, 2026 18:09

Copilot started reviewing on behalf of tannergooding July 16, 2026 18:10 View session

PranavSenthilnathan reviewed Jul 16, 2026

View reviewed changes

Comment thread src/libraries/System.Private.CoreLib/src/System/Numerics/Decimal64.cs

Copilot AI reviewed Jul 16, 2026

View reviewed changes

PranavSenthilnathan reviewed Jul 16, 2026

View reviewed changes

Comment thread src/libraries/System.Private.CoreLib/src/System/Numerics/Decimal64.cs

Remove the erroneously introduced byte order marks from the decimal t…

0da9c08

…est files Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

PranavSenthilnathan reviewed Jul 16, 2026

View reviewed changes

Comment thread src/libraries/System.Private.CoreLib/src/System/Number.DecimalIeee754.cs Outdated

tannergooding and others added 3 commits July 16, 2026 11:52

Update the decimal remainder remarks to match the chunked reduction

7b01738

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

PranavSenthilnathan reviewed Jul 16, 2026

View reviewed changes

Comment thread src/libraries/System.Private.CoreLib/src/System/Number.DecimalIeee754.cs Outdated

Use DivRem when splitting the rounded coefficient

94d0c69

Replace the manual divide and multiply-subtract with TValue.DivRem, matching the remainder computation used elsewhere in the file. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings July 16, 2026 20:38

Copilot started reviewing on behalf of tannergooding July 16, 2026 20:38 View session

Copilot AI reviewed Jul 16, 2026

View reviewed changes

Comment thread src/libraries/System.Runtime/ref/System.Runtime.cs

tannergooding enabled auto-merge (squash) July 16, 2026 20:55

PranavSenthilnathan approved these changes Jul 16, 2026

View reviewed changes

build-analysis Bot mentioned this pull request Jul 17, 2026

The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008

Open

3 tasks

tannergooding merged commit 874b9e9 into dotnet:main Jul 17, 2026
141 of 145 checks passed

tannergooding deleted the tannergooding-decimal-ieee754-ifloatingpoint branch July 17, 2026 03:55

Uh oh!

Conversation

tannergooding commented Jul 16, 2026

Uh oh!

azure-pipelines Bot commented Jul 16, 2026

Uh oh!

dotnet-policy-service Bot commented Jul 16, 2026

Uh oh!

azure-pipelines Bot commented Jul 16, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

PranavSenthilnathan commented Jul 16, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

tannergooding commented Jul 16, 2026

Uh oh!

tannergooding commented Jul 17, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants