[APFloat] Add support for f8E3M4 IEEE 754 type #99698
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
clang
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-llvm-adt Author: Alexander Pivovarov (apivovarov) ChangesThis PR adds
f8E3M4 (IEEE 754)
- Exponent bias: 3
- Maximum stored exponent value: 6 (binary 110)
- Maximum unbiased exponent value: 6 - 3 = 3
- Minimum stored exponent value: 1 (binary 001)
- Minimum unbiased exponent value: 1 − 3 = −2
- Precision specifies the total number of bits used for the significand (mantissa),
including implicit leading integer bit = 4 + 1 = 5
- Follows IEEE 754 conventions for representation of special values
- Has Positive and Negative zero
- Has Positive and Negative infinity
- Has NaNs
Additional details:
- Max exp (unbiased): 3
- Min exp (unbiased): -2
- Infinities (+/-): S.1111.000
- Zeros (+/-): S.0000.000
- NaNs: S.111.{0,1}⁴ except S.111.0000
- Max normal number: S.110.1111 = +/-2^(6-3) x (1 + 15/16) = +/-2^3 x 31 x 2^(-4) = +/-15.5
- Min normal number: S.001.0000 = +/-2^(1-3) x (1 + 0) = +/-2^(-2)
- Max subnormal number: S.000.1111 = +/-2^(-2) x 15/16 = +/-2^(-2) x 15 x 2^(-4) = +/-15 x 2^(-6)
- Min subnormal number: S.000.0001 = +/-2^(-2) x 1/16 = +/-2^(-2) x 2^(-4) = +/-2^(-6)Related PRs:
Full diff: https://github.com/llvm/llvm-project/pull/99698.diff 4 Files Affected:
diff --git a/clang/lib/AST/MicrosoftMangle.cpp b/clang/lib/AST/MicrosoftMangle.cpp
index 4016043df62ed..75627b5d4f4f7 100644
--- a/clang/lib/AST/MicrosoftMangle.cpp
+++ b/clang/lib/AST/MicrosoftMangle.cpp
@@ -986,6 +986,7 @@ void MicrosoftCXXNameMangler::mangleFloat(llvm::APFloat Number) {
case APFloat::S_Float8E5M2FNUZ:
case APFloat::S_Float8E4M3FNUZ:
case APFloat::S_Float8E4M3B11FNUZ:
+ case APFloat::S_Float8E3M4:
case APFloat::S_FloatTF32:
case APFloat::S_Float6E3M2FN:
case APFloat::S_Float6E2M3FN:
diff --git a/llvm/include/llvm/ADT/APFloat.h b/llvm/include/llvm/ADT/APFloat.h
index bff8e6490d1de..7039e961bff82 100644
--- a/llvm/include/llvm/ADT/APFloat.h
+++ b/llvm/include/llvm/ADT/APFloat.h
@@ -188,6 +188,9 @@ struct APFloatBase {
// This format's exponent bias is 11, instead of the 7 (2 ** (4 - 1) - 1)
// that IEEE precedent would imply.
S_Float8E4M3B11FNUZ,
+ // 8-bit floating point number following IEEE-754 conventions with bit
+ // layout S1E3M4.
+ S_Float8E3M4,
// Floating point number that occupies 32 bits or less of storage, providing
// improved range compared to half (16-bit) formats, at (potentially)
// greater throughput than single precision (32-bit) formats.
@@ -224,6 +227,7 @@ struct APFloatBase {
static const fltSemantics &Float8E4M3FN() LLVM_READNONE;
static const fltSemantics &Float8E4M3FNUZ() LLVM_READNONE;
static const fltSemantics &Float8E4M3B11FNUZ() LLVM_READNONE;
+ static const fltSemantics &Float8E3M4() LLVM_READNONE;
static const fltSemantics &FloatTF32() LLVM_READNONE;
static const fltSemantics &Float6E3M2FN() LLVM_READNONE;
static const fltSemantics &Float6E2M3FN() LLVM_READNONE;
@@ -646,6 +650,7 @@ class IEEEFloat final : public APFloatBase {
APInt convertFloat8E4M3FNAPFloatToAPInt() const;
APInt convertFloat8E4M3FNUZAPFloatToAPInt() const;
APInt convertFloat8E4M3B11FNUZAPFloatToAPInt() const;
+ APInt convertFloat8E3M4APFloatToAPInt() const;
APInt convertFloatTF32APFloatToAPInt() const;
APInt convertFloat6E3M2FNAPFloatToAPInt() const;
APInt convertFloat6E2M3FNAPFloatToAPInt() const;
@@ -665,6 +670,7 @@ class IEEEFloat final : public APFloatBase {
void initFromFloat8E4M3FNAPInt(const APInt &api);
void initFromFloat8E4M3FNUZAPInt(const APInt &api);
void initFromFloat8E4M3B11FNUZAPInt(const APInt &api);
+ void initFromFloat8E3M4APInt(const APInt &api);
void initFromFloatTF32APInt(const APInt &api);
void initFromFloat6E3M2FNAPInt(const APInt &api);
void initFromFloat6E2M3FNAPInt(const APInt &api);
diff --git a/llvm/lib/Support/APFloat.cpp b/llvm/lib/Support/APFloat.cpp
index 26b4f8e55448f..7f68c5ab9b7cf 100644
--- a/llvm/lib/Support/APFloat.cpp
+++ b/llvm/lib/Support/APFloat.cpp
@@ -143,6 +143,7 @@ static constexpr fltSemantics semFloat8E4M3FNUZ = {
7, -7, 4, 8, fltNonfiniteBehavior::NanOnly, fltNanEncoding::NegativeZero};
static constexpr fltSemantics semFloat8E4M3B11FNUZ = {
4, -10, 4, 8, fltNonfiniteBehavior::NanOnly, fltNanEncoding::NegativeZero};
+static constexpr fltSemantics semFloat8E3M4 = {3, -2, 5, 8};
static constexpr fltSemantics semFloatTF32 = {127, -126, 11, 19};
static constexpr fltSemantics semFloat6E3M2FN = {
4, -2, 3, 6, fltNonfiniteBehavior::FiniteOnly};
@@ -217,6 +218,8 @@ const llvm::fltSemantics &APFloatBase::EnumToSemantics(Semantics S) {
return Float8E4M3FNUZ();
case S_Float8E4M3B11FNUZ:
return Float8E4M3B11FNUZ();
+ case S_Float8E3M4:
+ return Float8E3M4();
case S_FloatTF32:
return FloatTF32();
case S_Float6E3M2FN:
@@ -257,6 +260,8 @@ APFloatBase::SemanticsToEnum(const llvm::fltSemantics &Sem) {
return S_Float8E4M3FNUZ;
else if (&Sem == &llvm::APFloat::Float8E4M3B11FNUZ())
return S_Float8E4M3B11FNUZ;
+ else if (&Sem == &llvm::APFloat::Float8E3M4())
+ return S_Float8E3M4;
else if (&Sem == &llvm::APFloat::FloatTF32())
return S_FloatTF32;
else if (&Sem == &llvm::APFloat::Float6E3M2FN())
@@ -287,6 +292,7 @@ const fltSemantics &APFloatBase::Float8E4M3FNUZ() { return semFloat8E4M3FNUZ; }
const fltSemantics &APFloatBase::Float8E4M3B11FNUZ() {
return semFloat8E4M3B11FNUZ;
}
+const fltSemantics &APFloatBase::Float8E3M4() { return semFloat8E3M4; }
const fltSemantics &APFloatBase::FloatTF32() { return semFloatTF32; }
const fltSemantics &APFloatBase::Float6E3M2FN() { return semFloat6E3M2FN; }
const fltSemantics &APFloatBase::Float6E2M3FN() { return semFloat6E2M3FN; }
@@ -3643,6 +3649,11 @@ APInt IEEEFloat::convertFloat8E4M3B11FNUZAPFloatToAPInt() const {
return convertIEEEFloatToAPInt<semFloat8E4M3B11FNUZ>();
}
+APInt IEEEFloat::convertFloat8E3M4APFloatToAPInt() const {
+ assert(partCount() == 1);
+ return convertIEEEFloatToAPInt<semFloat8E3M4>();
+}
+
APInt IEEEFloat::convertFloatTF32APFloatToAPInt() const {
assert(partCount() == 1);
return convertIEEEFloatToAPInt<semFloatTF32>();
@@ -3704,6 +3715,9 @@ APInt IEEEFloat::bitcastToAPInt() const {
if (semantics == (const llvm::fltSemantics *)&semFloat8E4M3B11FNUZ)
return convertFloat8E4M3B11FNUZAPFloatToAPInt();
+ if (semantics == (const llvm::fltSemantics *)&semFloat8E3M4)
+ return convertFloat8E3M4APFloatToAPInt();
+
if (semantics == (const llvm::fltSemantics *)&semFloatTF32)
return convertFloatTF32APFloatToAPInt();
@@ -3932,6 +3946,10 @@ void IEEEFloat::initFromFloat8E4M3B11FNUZAPInt(const APInt &api) {
initFromIEEEAPInt<semFloat8E4M3B11FNUZ>(api);
}
+void IEEEFloat::initFromFloat8E3M4APInt(const APInt &api) {
+ initFromIEEEAPInt<semFloat8E3M4>(api);
+}
+
void IEEEFloat::initFromFloatTF32APInt(const APInt &api) {
initFromIEEEAPInt<semFloatTF32>(api);
}
@@ -3977,6 +3995,8 @@ void IEEEFloat::initFromAPInt(const fltSemantics *Sem, const APInt &api) {
return initFromFloat8E4M3FNUZAPInt(api);
if (Sem == &semFloat8E4M3B11FNUZ)
return initFromFloat8E4M3B11FNUZAPInt(api);
+ if (Sem == &semFloat8E3M4)
+ return initFromFloat8E3M4APInt(api);
if (Sem == &semFloatTF32)
return initFromFloatTF32APInt(api);
if (Sem == &semFloat6E3M2FN)
diff --git a/llvm/unittests/ADT/APFloatTest.cpp b/llvm/unittests/ADT/APFloatTest.cpp
index d50bdf4a65dcb..8a9488165b45a 100644
--- a/llvm/unittests/ADT/APFloatTest.cpp
+++ b/llvm/unittests/ADT/APFloatTest.cpp
@@ -2141,6 +2141,8 @@ TEST(APFloatTest, getZero) {
{&APFloat::Float8E4M3FNUZ(), true, false, {0, 0}, 1},
{&APFloat::Float8E4M3B11FNUZ(), false, false, {0, 0}, 1},
{&APFloat::Float8E4M3B11FNUZ(), true, false, {0, 0}, 1},
+ {&APFloat::Float8E3M4(), false, true, {0, 0}, 1},
+ {&APFloat::Float8E3M4(), true, true, {0x80ULL, 0}, 1},
{&APFloat::FloatTF32(), false, true, {0, 0}, 1},
{&APFloat::FloatTF32(), true, true, {0x40000ULL, 0}, 1},
{&APFloat::Float6E3M2FN(), false, true, {0, 0}, 1},
@@ -6636,6 +6638,44 @@ TEST(APFloatTest, Float8E4M3FNUZToDouble) {
EXPECT_TRUE(std::isnan(QNaN.convertToDouble()));
}
+TEST(APFloatTest, Float8E3M4ToDouble) {
+ APFloat PosZero = APFloat::getZero(APFloat::Float8E3M4(), false);
+ APFloat PosZeroToDouble(PosZero.convertToDouble());
+ EXPECT_TRUE(PosZeroToDouble.isPosZero());
+ APFloat NegZero = APFloat::getZero(APFloat::Float8E3M4(), true);
+ APFloat NegZeroToDouble(NegZero.convertToDouble());
+ EXPECT_TRUE(NegZeroToDouble.isNegZero());
+
+ APFloat One(APFloat::Float8E3M4(), "1.0");
+ EXPECT_EQ(1.0, One.convertToDouble());
+ APFloat Two(APFloat::Float8E3M4(), "2.0");
+ EXPECT_EQ(2.0, Two.convertToDouble());
+ APFloat PosLargest = APFloat::getLargest(APFloat::Float8E3M4(), false);
+ EXPECT_EQ(15.5F, PosLargest.convertToDouble());
+ APFloat NegLargest = APFloat::getLargest(APFloat::Float8E3M4(), true);
+ EXPECT_EQ(-15.5F, NegLargest.convertToDouble());
+ APFloat PosSmallest =
+ APFloat::getSmallestNormalized(APFloat::Float8E3M4(), false);
+ EXPECT_EQ(0x1.p-2, PosSmallest.convertToDouble());
+ APFloat NegSmallest =
+ APFloat::getSmallestNormalized(APFloat::Float8E3M4(), true);
+ EXPECT_EQ(-0x1.p-2, NegSmallest.convertToDouble());
+
+ APFloat PosSmallestDenorm = APFloat::getSmallest(APFloat::Float8E3M4(), false);
+ EXPECT_TRUE(PosSmallestDenorm.isDenormal());
+ EXPECT_EQ(0x1.p-6, PosSmallestDenorm.convertToDouble());
+ APFloat NegSmallestDenorm = APFloat::getSmallest(APFloat::Float8E3M4(), true);
+ EXPECT_TRUE(NegSmallestDenorm.isDenormal());
+ EXPECT_EQ(-0x1.p-6, NegSmallestDenorm.convertToDouble());
+
+ APFloat PosInf = APFloat::getInf(APFloat::Float8E3M4());
+ EXPECT_EQ(std::numeric_limits<double>::infinity(), PosInf.convertToDouble());
+ APFloat NegInf = APFloat::getInf(APFloat::Float8E3M4(), true);
+ EXPECT_EQ(-std::numeric_limits<double>::infinity(), NegInf.convertToDouble());
+ APFloat QNaN = APFloat::getQNaN(APFloat::Float8E3M4());
+ EXPECT_TRUE(std::isnan(QNaN.convertToDouble()));
+}
+
TEST(APFloatTest, FloatTF32ToDouble) {
APFloat One(APFloat::FloatTF32(), "1.0");
EXPECT_EQ(1.0, One.convertToDouble());
@@ -6944,6 +6984,45 @@ TEST(APFloatTest, Float8E4M3FNToFloat) {
EXPECT_TRUE(std::isnan(QNaN.convertToFloat()));
}
+TEST(APFloatTest, Float8E3M4ToFloat) {
+ APFloat PosZero = APFloat::getZero(APFloat::Float8E3M4(), false);
+ APFloat PosZeroToFloat(PosZero.convertToFloat());
+ EXPECT_TRUE(PosZeroToFloat.isPosZero());
+ APFloat NegZero = APFloat::getZero(APFloat::Float8E3M4(), true);
+ APFloat NegZeroToFloat(NegZero.convertToFloat());
+ EXPECT_TRUE(NegZeroToFloat.isNegZero());
+
+ APFloat One(APFloat::Float8E3M4(), "1.0");
+ EXPECT_EQ(1.0F, One.convertToFloat());
+ APFloat Two(APFloat::Float8E3M4(), "2.0");
+ EXPECT_EQ(2.0F, Two.convertToFloat());
+
+ APFloat PosLargest = APFloat::getLargest(APFloat::Float8E3M4(), false);
+ EXPECT_EQ(15.5F, PosLargest.convertToFloat());
+ APFloat NegLargest = APFloat::getLargest(APFloat::Float8E3M4(), true);
+ EXPECT_EQ(-15.5F, NegLargest.convertToFloat());
+ APFloat PosSmallest =
+ APFloat::getSmallestNormalized(APFloat::Float8E3M4(), false);
+ EXPECT_EQ(0x1.p-2, PosSmallest.convertToFloat());
+ APFloat NegSmallest =
+ APFloat::getSmallestNormalized(APFloat::Float8E3M4(), true);
+ EXPECT_EQ(-0x1.p-2, NegSmallest.convertToFloat());
+
+ APFloat PosSmallestDenorm = APFloat::getSmallest(APFloat::Float8E3M4(), false);
+ EXPECT_TRUE(PosSmallestDenorm.isDenormal());
+ EXPECT_EQ(0x1.p-6, PosSmallestDenorm.convertToFloat());
+ APFloat NegSmallestDenorm = APFloat::getSmallest(APFloat::Float8E3M4(), true);
+ EXPECT_TRUE(NegSmallestDenorm.isDenormal());
+ EXPECT_EQ(-0x1.p-6, NegSmallestDenorm.convertToFloat());
+
+ APFloat PosInf = APFloat::getInf(APFloat::Float8E3M4());
+ EXPECT_EQ(std::numeric_limits<float>::infinity(), PosInf.convertToFloat());
+ APFloat NegInf = APFloat::getInf(APFloat::Float8E3M4(), true);
+ EXPECT_EQ(-std::numeric_limits<float>::infinity(), NegInf.convertToFloat());
+ APFloat QNaN = APFloat::getQNaN(APFloat::Float8E3M4());
+ EXPECT_TRUE(std::isnan(QNaN.convertToFloat()));
+}
+
TEST(APFloatTest, FloatTF32ToFloat) {
APFloat PosZero = APFloat::getZero(APFloat::FloatTF32());
APFloat PosZeroToFloat(PosZero.convertToFloat());
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
apivovarov
force-pushed
the
f8E3M4_APFloat
branch
from
fdcc285
to
1bddc5c
Compare
|
Hi Thomas, when you have a moment, could you please review this PR? It is similar to prev PR-97179 |
|
@tschuett, Could you please help review this change? |
I did. I gave you the float reviewer of the house. |
ThomasRaoux
approved these changes
Jul 26, 2024
|
If there are no further questions, comments, or objections, I will merge this PR in a few hours. |
apivovarov
added a commit
that referenced
this pull request
Aug 2, 2024
This PR adds `f8E3M4` type to mlir.
`f8E3M4` type follows IEEE 754 convention
```c
f8E3M4 (IEEE 754)
- Exponent bias: 3
- Maximum stored exponent value: 6 (binary 110)
- Maximum unbiased exponent value: 6 - 3 = 3
- Minimum stored exponent value: 1 (binary 001)
- Minimum unbiased exponent value: 1 − 3 = −2
- Precision specifies the total number of bits used for the significand (mantissa),
including implicit leading integer bit = 4 + 1 = 5
- Follows IEEE 754 conventions for representation of special values
- Has Positive and Negative zero
- Has Positive and Negative infinity
- Has NaNs
Additional details:
- Max exp (unbiased): 3
- Min exp (unbiased): -2
- Infinities (+/-): S.111.0000
- Zeros (+/-): S.000.0000
- NaNs: S.111.{0,1}⁴ except S.111.0000
- Max normal number: S.110.1111 = +/-2^(6-3) x (1 + 15/16) = +/-2^3 x 31 x 2^(-4) = +/-15.5
- Min normal number: S.001.0000 = +/-2^(1-3) x (1 + 0) = +/-2^(-2)
- Max subnormal number: S.000.1111 = +/-2^(-2) x 15/16 = +/-2^(-2) x 15 x 2^(-4) = +/-15 x 2^(-6)
- Min subnormal number: S.000.0001 = +/-2^(-2) x 1/16 = +/-2^(-2) x 2^(-4) = +/-2^(-6)
```
Related PRs:
- [PR-99698](#99698) [APFloat]
Add support for f8E3M4 IEEE 754 type
- [PR-97118](#97118) [MLIR] Add
f8E4M3 IEEE 754 type
banach-space
pushed a commit
to banach-space/llvm-project
that referenced
this pull request
Aug 7, 2024
This PR adds `f8E4M3` type to APFloat.
`f8E3M4` type follows IEEE 754 convention
```c
f8E3M4 (IEEE 754)
- Exponent bias: 3
- Maximum stored exponent value: 6 (binary 110)
- Maximum unbiased exponent value: 6 - 3 = 3
- Minimum stored exponent value: 1 (binary 001)
- Minimum unbiased exponent value: 1 − 3 = −2
- Precision specifies the total number of bits used for the significand (mantissa),
including implicit leading integer bit = 4 + 1 = 5
- Follows IEEE 754 conventions for representation of special values
- Has Positive and Negative zero
- Has Positive and Negative infinity
- Has NaNs
Additional details:
- Max exp (unbiased): 3
- Min exp (unbiased): -2
- Infinities (+/-): S.111.0000
- Zeros (+/-): S.000.0000
- NaNs: S.111.{0,1}⁴ except S.111.0000
- Max normal number: S.110.1111 = +/-2^(6-3) x (1 + 15/16) = +/-2^3 x 31 x 2^(-4) = +/-15.5
- Min normal number: S.001.0000 = +/-2^(1-3) x (1 + 0) = +/-2^(-2)
- Max subnormal number: S.000.1111 = +/-2^(-2) x 15/16 = +/-2^(-2) x 15 x 2^(-4) = +/-15 x 2^(-6)
- Min subnormal number: S.000.0001 = +/-2^(-2) x 1/16 = +/-2^(-2) x 2^(-4) = +/-2^(-6)
```
Related PRs:
- [PR-97179](llvm#97179) [APFloat]
Add support for f8E4M3 IEEE 754 type
banach-space
pushed a commit
to banach-space/llvm-project
that referenced
this pull request
Aug 7, 2024
This PR adds `f8E3M4` type to mlir.
`f8E3M4` type follows IEEE 754 convention
```c
f8E3M4 (IEEE 754)
- Exponent bias: 3
- Maximum stored exponent value: 6 (binary 110)
- Maximum unbiased exponent value: 6 - 3 = 3
- Minimum stored exponent value: 1 (binary 001)
- Minimum unbiased exponent value: 1 − 3 = −2
- Precision specifies the total number of bits used for the significand (mantissa),
including implicit leading integer bit = 4 + 1 = 5
- Follows IEEE 754 conventions for representation of special values
- Has Positive and Negative zero
- Has Positive and Negative infinity
- Has NaNs
Additional details:
- Max exp (unbiased): 3
- Min exp (unbiased): -2
- Infinities (+/-): S.111.0000
- Zeros (+/-): S.000.0000
- NaNs: S.111.{0,1}⁴ except S.111.0000
- Max normal number: S.110.1111 = +/-2^(6-3) x (1 + 15/16) = +/-2^3 x 31 x 2^(-4) = +/-15.5
- Min normal number: S.001.0000 = +/-2^(1-3) x (1 + 0) = +/-2^(-2)
- Max subnormal number: S.000.1111 = +/-2^(-2) x 15/16 = +/-2^(-2) x 15 x 2^(-4) = +/-15 x 2^(-6)
- Min subnormal number: S.000.0001 = +/-2^(-2) x 1/16 = +/-2^(-2) x 2^(-4) = +/-2^(-6)
```
Related PRs:
- [PR-99698](llvm#99698) [APFloat]
Add support for f8E3M4 IEEE 754 type
- [PR-97118](llvm#97118) [MLIR] Add
f8E4M3 IEEE 754 type
This was referenced Aug 9, 2024
GleasonK
pushed a commit
to openxla/stablehlo
that referenced
this pull request
Sep 3, 2024
### Summary This is a proposal to add `Float8E4M3` and `Float8E3M4` floating point types to StableHLO. Feedback welcome, see [RFC: Float8E4M3 and Float8E3M4](https://github.com/apivovarov/stablehlo/blob/rfc_f8E4M3_f8E3M4/rfcs/20240808-f8E4M3_f8E3M4.md) for more details. ### References and Links - LLVM [PR-97179](llvm/llvm-project#97179) [APFloat] Add support for f8E4M3 IEEE 754 type (Merged) - LLVM [PR-97118](llvm/llvm-project#97118) [MLIR] Add f8E4M3 IEEE 754 type (Merged) - LLVM [PR-99698](llvm/llvm-project#99698) [APFloat] Add support for f8E3M4 IEEE 754 type (Merged) - LLVM [PR-101230](llvm/llvm-project#101230) [MLIR] Add f8E3M4 IEEE 754 type (Merged) - [RFC: FP8 in StableHLO](https://github.com/openxla/stablehlo/blob/main/rfcs/20221031-fp8.md) - [RFC: Float8E4M3FNUZ and Float8E5M2FNUZ](https://github.com/openxla/stablehlo/blob/main/rfcs/20230321-fp8_fnuz.md) - StableHLO [PR-2482](#2482) Add f8E4M3 and f8E3M4 types support - [Amazon EC2 Trn1 Instances](https://aws.amazon.com/ec2/instance-types/trn1/) - ml_dtypes [PR-161](jax-ml/ml_dtypes#161) Add float8_e4m3 (Merged) - ml_dtypes [PR-171](jax-ml/ml_dtypes#171) Add float8_e3m4 (Merged) - XLA [PR-16585](openxla/xla#16585) Add support for float8_e4m3
GleasonK
pushed a commit
to openxla/stablehlo
that referenced
this pull request
Sep 4, 2024
This PR adds f8E4M3 and f8E3M4 types support.
f8E4M3 and f8E3M4 types follow IEEE 754 convention.
```c
f8E4M3 (IEEE 754)
- Exponent bias: 7
- Maximum stored exponent value: 14 (binary 1110)
- Maximum unbiased exponent value: 14 - 7 = 7
- Minimum stored exponent value: 1 (binary 0001)
- Minimum unbiased exponent value: 1 − 7 = −6
- Precision specifies the total number of bits used for the significand (mantisa),
including implicit leading integer bit = 3 + 1 = 4
- Follows IEEE 754 conventions for representation of special values
- Has Positive and Negative zero
- Has Positive and Negative infinity
- Has NaNs
Additional details:
- Max exp (unbiased): 7
- Min exp (unbiased): -6
- Infinities (+/-): S.1111.000
- Zeros (+/-): S.0000.000
- NaNs: S.1111.{001, 010, 011, 100, 101, 110, 111}
- Max normal number: S.1110.111 = +/-2^(7) x (1 + 0.875) = +/-240
- Min normal number: S.0001.000 = +/-2^(-6)
- Max subnormal number: S.0000.111 = +/-2^(-6) x 0.875 = +/-2^(-9) x 7
- Min subnormal number: S.0000.001 = +/-2^(-6) x 0.125 = +/-2^(-9)
```
```c
f8E3M4 (IEEE 754)
- Exponent bias: 3
- Maximum stored exponent value: 6 (binary 110)
- Maximum unbiased exponent value: 6 - 3 = 3
- Minimum stored exponent value: 1 (binary 001)
- Minimum unbiased exponent value: 1 − 3 = −2
- Precision specifies the total number of bits used for the significand (mantissa),
including implicit leading integer bit = 4 + 1 = 5
- Follows IEEE 754 conventions for representation of special values
- Has Positive and Negative zero
- Has Positive and Negative infinity
- Has NaNs
Additional details:
- Max exp (unbiased): 3
- Min exp (unbiased): -2
- Infinities (+/-): S.111.0000
- Zeros (+/-): S.000.0000
- NaNs: S.111.{0,1}⁴ except S.111.0000
- Max normal number: S.110.1111 = +/-2^(6-3) x (1 + 15/16) = +/-2^3 x 31 x 2^(-4) = +/-15.5
- Min normal number: S.001.0000 = +/-2^(1-3) x (1 + 0) = +/-2^(-2)
- Max subnormal number: S.000.1111 = +/-2^(-2) x 15/16 = +/-2^(-2) x 15 x 2^(-4) = +/-15 x 2^(-6)
- Min subnormal number: S.000.0001 = +/-2^(-2) x 1/16 = +/-2^(-2) x 2^(-4) = +/-2^(-6)
```
Related PRs:
- LLVM [PR-97179](llvm/llvm-project#97179)
[APFloat] Add support for f8E4M3 IEEE 754 type (Merged)
- LLVM [PR-97118](llvm/llvm-project#97118)
[MLIR] Add f8E4M3 IEEE 754 type (Merged)
- LLVM [PR-99698](llvm/llvm-project#99698)
[APFloat] Add support for f8E3M4 IEEE 754 type (Merged)
- LLVM [PR-101230](llvm/llvm-project#101230)
[MLIR] Add f8E3M4 IEEE 754 type (Merged)
- StableHLO [PR-2486](#2486)
[RFC] Add f8E4M3 and f8E3M4 types support
- ml_dtypes [PR-161](jax-ml/ml_dtypes#161) Add
float8_e4m3 (Merged)
- ml_dtypes [PR-171](jax-ml/ml_dtypes#171) Add
float8_e3m4 (Merged)
- XLA [PR-16585](openxla/xla#16585) Add support
for float8_e4m3
This was referenced Sep 23, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds
f8E4M3type to APFloat.f8E3M4type follows IEEE 754 conventionRelated PRs: