[clang] [Headers][X86] Add rounding and exception notes to conversions (PR #83447)
Paul T Robinson via cfe-commits
cfe-commits at lists.llvm.org
Wed Mar 13 09:00:39 PDT 2024
https://github.com/pogo59 updated https://github.com/llvm/llvm-project/pull/83447
>From 64cdd358d0bf359383a5dd3d1da236a219644c9e Mon Sep 17 00:00:00 2001
From: Paul Robinson <paul.robinson at sony.com>
Date: Thu, 29 Feb 2024 08:59:26 -0800
Subject: [PATCH 1/4] [Headers][X86] Add rounding and exception notes to
conversions
Consistently describe rounding/truncating on convert intrinsics.
Add notes where an out-of-range result can raise an exception.
---
clang/lib/Headers/avxintrin.h | 36 +++++++---
clang/lib/Headers/emmintrin.h | 91 ++++++++++++++++++-------
clang/lib/Headers/xmmintrin.h | 125 ++++++++++++++++++++++++----------
3 files changed, 182 insertions(+), 70 deletions(-)
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index f116d8bc3a94c7..51c7d76e75ca1a 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -2180,7 +2180,8 @@ _mm256_cvtepi32_pd(__m128i __a)
return (__m256d)__builtin_convertvector((__v4si)__a, __v4df);
}
-/// Converts a vector of [8 x i32] into a vector of [8 x float].
+/// Converts a vector of [8 x i32] into a vector of [8 x float]. Rounds inexact
+/// results according to the rounding control bits in the MXCSR register.
///
/// \headerfile <x86intrin.h>
///
@@ -2196,7 +2197,8 @@ _mm256_cvtepi32_ps(__m256i __a)
}
/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of
-/// [4 x float].
+/// [4 x float]. Rounds inexact results according to the rounding control
+/// bits in the MXCSR register.
///
/// \headerfile <x86intrin.h>
///
@@ -2211,7 +2213,12 @@ _mm256_cvtpd_ps(__m256d __a)
return (__m128)__builtin_ia32_cvtpd2ps256((__v4df) __a);
}
-/// Converts a vector of [8 x float] into a vector of [8 x i32].
+/// Converts a vector of [8 x float] into a vector of [8 x i32]. Rounds inexact
+/// results according to the rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -2243,8 +2250,11 @@ _mm256_cvtps_pd(__m128 __a)
}
/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4
-/// x i32], truncating the result by rounding towards zero when it is
-/// inexact.
+/// x i32], truncating inexact results.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -2259,9 +2269,13 @@ _mm256_cvttpd_epi32(__m256d __a)
return (__m128i)__builtin_ia32_cvttpd2dq256((__v4df) __a);
}
-/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4
-/// x i32]. When a conversion is inexact, the value returned is rounded
-/// according to the rounding control bits in the MXCSR register.
+/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of
+/// [4 x i32]. Rounds inexact results according to the rounding control bits
+/// in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -2277,7 +2291,11 @@ _mm256_cvtpd_epi32(__m256d __a)
}
/// Converts a vector of [8 x float] into a vector of [8 x i32],
-/// truncating the result by rounding towards zero when it is inexact.
+/// truncating inexact results.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h
index 96e3ebdecbdf83..9ae7fe2c809ac6 100644
--- a/clang/lib/Headers/emmintrin.h
+++ b/clang/lib/Headers/emmintrin.h
@@ -1245,7 +1245,8 @@ static __inline__ int __DEFAULT_FN_ATTRS _mm_ucomineq_sd(__m128d __a,
/// Converts the two double-precision floating-point elements of a
/// 128-bit vector of [2 x double] into two single-precision floating-point
/// values, returned in the lower 64 bits of a 128-bit vector of [4 x float].
-/// The upper 64 bits of the result vector are set to zero.
+/// Rounds inexact results according to the rounding control bits in the
+/// MXCSR register. The upper 64 bits of the result vector are set to zero.
///
/// \headerfile <x86intrin.h>
///
@@ -1300,9 +1301,13 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_cvtepi32_pd(__m128i __a) {
}
/// Converts the two double-precision floating-point elements of a
-/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
-/// returned in the lower 64 bits of a 128-bit vector of [4 x i32]. The upper
-/// 64 bits of the result vector are set to zero.
+/// 128-bit vector of [2 x double] into two signed 32-bit integer values.
+/// Rounds inexact results according to the rounding control bits in the
+/// MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1317,7 +1322,12 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvtpd_epi32(__m128d __a) {
}
/// Converts the low-order element of a 128-bit vector of [2 x double]
-/// into a 32-bit signed integer value.
+/// into a 32-bit signed integer value. Rounds inexact results according to
+/// the rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1333,9 +1343,10 @@ static __inline__ int __DEFAULT_FN_ATTRS _mm_cvtsd_si32(__m128d __a) {
/// Converts the lower double-precision floating-point element of a
/// 128-bit vector of [2 x double], in the second parameter, into a
-/// single-precision floating-point value, returned in the lower 32 bits of a
-/// 128-bit vector of [4 x float]. The upper 96 bits of the result vector are
-/// copied from the upper 96 bits of the first parameter.
+/// single-precision floating-point value, returned in the lower 32 bits of
+/// a 128-bit vector of [4 x float]. Rounds inexact results according to the
+/// rounding control bits in the MXCSR register. The upper 96 bits of the
+/// result vector are copied from the upper 96 bits of the first parameter.
///
/// \headerfile <x86intrin.h>
///
@@ -1405,11 +1416,11 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_cvtss_sd(__m128d __a,
/// Converts the two double-precision floating-point elements of a
/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
-/// returned in the lower 64 bits of a 128-bit vector of [4 x i32].
+/// truncating inexact results.
///
-/// If the result of either conversion is inexact, the result is truncated
-/// (rounded towards zero) regardless of the current MXCSR setting. The upper
-/// 64 bits of the result vector are set to zero.
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1425,7 +1436,11 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvttpd_epi32(__m128d __a) {
}
/// Converts the low-order element of a [2 x double] vector into a 32-bit
-/// signed integer value, truncating the result when it is inexact.
+/// signed integer value, truncating inexact results.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1441,8 +1456,13 @@ static __inline__ int __DEFAULT_FN_ATTRS _mm_cvttsd_si32(__m128d __a) {
}
/// Converts the two double-precision floating-point elements of a
-/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
-/// returned in a 64-bit vector of [2 x i32].
+/// 128-bit vector of [2 x double] into two signed 32-bit integer values.
+/// Rounds inexact results according to the rounding control bits in the
+/// MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1457,10 +1477,11 @@ static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX _mm_cvtpd_pi32(__m128d __a) {
/// Converts the two double-precision floating-point elements of a
/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
-/// returned in a 64-bit vector of [2 x i32].
+/// truncating inexact results.
///
-/// If the result of either conversion is inexact, the result is truncated
-/// (rounded towards zero) regardless of the current MXCSR setting.
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3185,8 +3206,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cmplt_epi32(__m128i __a,
#ifdef __x86_64__
/// Converts a 64-bit signed integer value from the second operand into a
/// double-precision value and returns it in the lower element of a [2 x
-/// double] vector; the upper element of the returned vector is copied from
-/// the upper element of the first operand.
+/// double] vector. Rounds inexact results according to the rounding control
+/// bits in the MXCSR register. The upper element of the returned vector is
+/// copied from the upper element of the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -3207,7 +3229,12 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_cvtsi64_sd(__m128d __a,
}
/// Converts the first (lower) element of a vector of [2 x double] into a
-/// 64-bit signed integer value, according to the current rounding mode.
+/// 64-bit signed integer value. Rounds inexact results according to the
+/// rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3222,7 +3249,11 @@ static __inline__ long long __DEFAULT_FN_ATTRS _mm_cvtsd_si64(__m128d __a) {
}
/// Converts the first (lower) element of a vector of [2 x double] into a
-/// 64-bit signed integer value, truncating the result when it is inexact.
+/// 64-bit signed integer value, truncating inexact results.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3238,7 +3269,8 @@ static __inline__ long long __DEFAULT_FN_ATTRS _mm_cvttsd_si64(__m128d __a) {
}
#endif
-/// Converts a vector of [4 x i32] into a vector of [4 x float].
+/// Converts a vector of [4 x i32] into a vector of [4 x float]. Rounds inexact
+/// results according to the rounding control bits in the MXCSR register.
///
/// \headerfile <x86intrin.h>
///
@@ -3251,7 +3283,12 @@ static __inline__ __m128 __DEFAULT_FN_ATTRS _mm_cvtepi32_ps(__m128i __a) {
return (__m128) __builtin_convertvector((__v4si)__a, __v4sf);
}
-/// Converts a vector of [4 x float] into a vector of [4 x i32].
+/// Converts a vector of [4 x float] into a vector of [4 x i32]. Rounds inexact
+/// results according to the rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3266,7 +3303,11 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvtps_epi32(__m128 __a) {
}
/// Converts a vector of [4 x float] into a vector of [4 x i32],
-/// truncating the result when it is inexact.
+/// truncating inexact results.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
diff --git a/clang/lib/Headers/xmmintrin.h b/clang/lib/Headers/xmmintrin.h
index 47368f3c23d2d6..0feee669e1e818 100644
--- a/clang/lib/Headers/xmmintrin.h
+++ b/clang/lib/Headers/xmmintrin.h
@@ -1295,7 +1295,12 @@ _mm_ucomineq_ss(__m128 __a, __m128 __b)
}
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer.
+/// [4 x float] into a 32-bit integer. Rounds an inexact result according to
+/// the rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1313,7 +1318,12 @@ _mm_cvtss_si32(__m128 __a)
}
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer.
+/// [4 x float] into a 32-bit integer. Rounds an inexact result according to
+/// the rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1333,7 +1343,12 @@ _mm_cvt_ss2si(__m128 __a)
#ifdef __x86_64__
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 64-bit integer.
+/// [4 x float] into a 64-bit integer. Rounds an inexact result according to
+/// the rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1352,8 +1367,13 @@ _mm_cvtss_si64(__m128 __a)
#endif
-/// Converts two low-order float values in a 128-bit vector of
-/// [4 x float] into a 64-bit vector of [2 x i32].
+/// Converts two low-order float values in a 128-bit vector of [4 x float] into
+/// a 64-bit vector of [2 x i32]. Rounds inexact results according to the
+/// rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1368,8 +1388,13 @@ _mm_cvtps_pi32(__m128 __a)
return (__m64)__builtin_ia32_cvtps2pi((__v4sf)__a);
}
-/// Converts two low-order float values in a 128-bit vector of
-/// [4 x float] into a 64-bit vector of [2 x i32].
+/// Converts two low-order float values in a 128-bit vector of [4 x float] into
+/// a 64-bit vector of [2 x i32]. Rounds inexact results according to the
+/// rounding control bits in the MXCSR register.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1385,8 +1410,11 @@ _mm_cvt_ps2pi(__m128 __a)
}
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer, truncating the result when it is
-/// inexact.
+/// [4 x float] into a 32-bit integer, truncating an inexact result.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1404,8 +1432,11 @@ _mm_cvttss_si32(__m128 __a)
}
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer, truncating the result when it is
-/// inexact.
+/// [4 x float] into a 32-bit integer, truncating an inexact result.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1424,8 +1455,11 @@ _mm_cvtt_ss2si(__m128 __a)
#ifdef __x86_64__
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 64-bit integer, truncating the result when it is
-/// inexact.
+/// [4 x float] into a 64-bit integer, truncating an inexact result.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1444,8 +1478,12 @@ _mm_cvttss_si64(__m128 __a)
#endif
/// Converts two low-order float values in a 128-bit vector of
-/// [4 x float] into a 64-bit vector of [2 x i32], truncating the result
-/// when it is inexact.
+/// [4 x float] into a 64-bit vector of [2 x i32], truncating inexact
+/// results.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1462,8 +1500,11 @@ _mm_cvttps_pi32(__m128 __a)
}
/// Converts two low-order float values in a 128-bit vector of [4 x
-/// float] into a 64-bit vector of [2 x i32], truncating the result when it
-/// is inexact.
+/// float] into a 64-bit vector of [2 x i32], truncating inexact results.
+///
+/// If a converted value is larger than the maximum possible result,
+/// raises a floating-point invalid exception. If the exception is
+/// masked, returns the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1478,10 +1519,11 @@ _mm_cvtt_ps2pi(__m128 __a)
return _mm_cvttps_pi32(__a);
}
-/// Converts a 32-bit signed integer value into a floating point value
-/// and writes it to the lower 32 bits of the destination. The remaining
-/// higher order elements of the destination vector are copied from the
-/// corresponding elements in the first operand.
+/// Converts a 32-bit signed integer value into a floating point value and
+/// writes it to the lower 32 bits of the destination. Rounds an inexact
+/// result according to the rounding control bits in the MXCSR register.
+/// The remaining higher order elements of the destination vector are copied
+/// from the corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1501,10 +1543,11 @@ _mm_cvtsi32_ss(__m128 __a, int __b)
return __a;
}
-/// Converts a 32-bit signed integer value into a floating point value
-/// and writes it to the lower 32 bits of the destination. The remaining
-/// higher order elements of the destination are copied from the
-/// corresponding elements in the first operand.
+/// Converts a 32-bit signed integer value into a floating point value and
+/// writes it to the lower 32 bits of the destination. Rounds an inexact
+/// result according to the rounding control bits in the MXCSR register.
+/// The remaining higher order elements of the destination are copied from
+/// the corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1525,10 +1568,11 @@ _mm_cvt_si2ss(__m128 __a, int __b)
#ifdef __x86_64__
-/// Converts a 64-bit signed integer value into a floating point value
-/// and writes it to the lower 32 bits of the destination. The remaining
-/// higher order elements of the destination are copied from the
-/// corresponding elements in the first operand.
+/// Converts a 64-bit signed integer value into a floating point value and
+/// writes it to the lower 32 bits of the destination. Rounds an inexact
+/// result according to the rounding control bits in the MXCSR register.
+/// The remaining higher order elements of the destination are copied from
+/// the corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1552,8 +1596,10 @@ _mm_cvtsi64_ss(__m128 __a, long long __b)
/// Converts two elements of a 64-bit vector of [2 x i32] into two
/// floating point values and writes them to the lower 64-bits of the
-/// destination. The remaining higher order elements of the destination are
-/// copied from the corresponding elements in the first operand.
+/// destination. Rounds inexact results according to the rounding control
+/// bits in the MXCSR register. The remaining higher order elements of the
+/// destination are copied from the corresponding elements in the first
+/// operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1575,8 +1621,10 @@ _mm_cvtpi32_ps(__m128 __a, __m64 __b)
/// Converts two elements of a 64-bit vector of [2 x i32] into two
/// floating point values and writes them to the lower 64-bits of the
-/// destination. The remaining higher order elements of the destination are
-/// copied from the corresponding elements in the first operand.
+/// destination. Rounds inexact results according to the rounding control
+/// bits in the MXCSR register. The remaining higher order elements of the
+/// destination are copied from the corresponding elements in the first
+/// operand.
///
/// \headerfile <x86intrin.h>
///
@@ -2835,7 +2883,9 @@ _mm_cvtpu8_ps(__m64 __a)
}
/// Converts the two 32-bit signed integer values from each 64-bit vector
-/// operand of [2 x i32] into a 128-bit vector of [4 x float].
+/// operand of [2 x i32] into a 128-bit vector of [4 x float]. Rounds
+/// inexact results according to the rounding control bits in the MXCSR
+/// register.
///
/// \headerfile <x86intrin.h>
///
@@ -2864,7 +2914,9 @@ _mm_cvtpi32x2_ps(__m64 __a, __m64 __b)
/// Converts each single-precision floating-point element of a 128-bit
/// floating-point vector of [4 x float] into a 16-bit signed integer, and
-/// packs the results into a 64-bit integer vector of [4 x i16].
+/// packs the results into a 64-bit integer vector of [4 x i16]. Rounds
+/// inexact results according to the rounding control bits in the MXCSR
+/// register.
///
/// If the floating-point element is NaN or infinity, or if the
/// floating-point element is greater than 0x7FFFFFFF or less than -0x8000,
@@ -2894,7 +2946,8 @@ _mm_cvtps_pi16(__m128 __a)
/// Converts each single-precision floating-point element of a 128-bit
/// floating-point vector of [4 x float] into an 8-bit signed integer, and
/// packs the results into the lower 32 bits of a 64-bit integer vector of
-/// [8 x i8]. The upper 32 bits of the vector are set to 0.
+/// [8 x i8]. The upper 32 bits of the vector are set to 0. Rounds inexact
+/// results according to the rounding control bits in the MXCSR register.
///
/// If the floating-point element is NaN or infinity, or if the
/// floating-point element is greater than 0x7FFFFFFF or less than -0x80, it
>From bf8d6301fe6716b6f25bd1e6c9e06088881a0b6e Mon Sep 17 00:00:00 2001
From: Paul Robinson <paul.robinson at sony.com>
Date: Thu, 7 Mar 2024 18:06:56 -0800
Subject: [PATCH 2/4] Don't bother describing normal (MXCSR-based) rounding.
Rephrase the conversion note to apply to both positive and negative overflow.
---
clang/lib/Headers/avxintrin.h | 36 +++++-----
clang/lib/Headers/emmintrin.h | 101 +++++++++++++-------------
clang/lib/Headers/xmmintrin.h | 129 +++++++++++++++-------------------
3 files changed, 119 insertions(+), 147 deletions(-)
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index 51c7d76e75ca1a..11c1dfa79af37c 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -2180,8 +2180,7 @@ _mm256_cvtepi32_pd(__m128i __a)
return (__m256d)__builtin_convertvector((__v4si)__a, __v4df);
}
-/// Converts a vector of [8 x i32] into a vector of [8 x float]. Rounds inexact
-/// results according to the rounding control bits in the MXCSR register.
+/// Converts a vector of [8 x i32] into a vector of [8 x float].
///
/// \headerfile <x86intrin.h>
///
@@ -2197,8 +2196,7 @@ _mm256_cvtepi32_ps(__m256i __a)
}
/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of
-/// [4 x float]. Rounds inexact results according to the rounding control
-/// bits in the MXCSR register.
+/// [4 x float].
///
/// \headerfile <x86intrin.h>
///
@@ -2213,12 +2211,11 @@ _mm256_cvtpd_ps(__m256d __a)
return (__m128)__builtin_ia32_cvtpd2ps256((__v4df) __a);
}
-/// Converts a vector of [8 x float] into a vector of [8 x i32]. Rounds inexact
-/// results according to the rounding control bits in the MXCSR register.
+/// Converts a vector of [8 x float] into a vector of [8 x i32].
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -2252,9 +2249,9 @@ _mm256_cvtps_pd(__m128 __a)
/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4
/// x i32], truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -2270,12 +2267,11 @@ _mm256_cvttpd_epi32(__m256d __a)
}
/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of
-/// [4 x i32]. Rounds inexact results according to the rounding control bits
-/// in the MXCSR register.
+/// [4 x i32].
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -2293,9 +2289,9 @@ _mm256_cvtpd_epi32(__m256d __a)
/// Converts a vector of [8 x float] into a vector of [8 x i32],
/// truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h
index 9ae7fe2c809ac6..a934a078af51f7 100644
--- a/clang/lib/Headers/emmintrin.h
+++ b/clang/lib/Headers/emmintrin.h
@@ -1245,8 +1245,7 @@ static __inline__ int __DEFAULT_FN_ATTRS _mm_ucomineq_sd(__m128d __a,
/// Converts the two double-precision floating-point elements of a
/// 128-bit vector of [2 x double] into two single-precision floating-point
/// values, returned in the lower 64 bits of a 128-bit vector of [4 x float].
-/// Rounds inexact results according to the rounding control bits in the
-/// MXCSR register. The upper 64 bits of the result vector are set to zero.
+/// The upper 64 bits of the result vector are set to zero.
///
/// \headerfile <x86intrin.h>
///
@@ -1301,13 +1300,13 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_cvtepi32_pd(__m128i __a) {
}
/// Converts the two double-precision floating-point elements of a
-/// 128-bit vector of [2 x double] into two signed 32-bit integer values.
-/// Rounds inexact results according to the rounding control bits in the
-/// MXCSR register.
+/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
+/// returned in the lower 64 bits of a 128-bit vector of [4 x i32]. The upper
+/// 64 bits of the result vector are set to zero.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1322,12 +1321,11 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvtpd_epi32(__m128d __a) {
}
/// Converts the low-order element of a 128-bit vector of [2 x double]
-/// into a 32-bit signed integer value. Rounds inexact results according to
-/// the rounding control bits in the MXCSR register.
+/// into a 32-bit signed integer value.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1343,10 +1341,9 @@ static __inline__ int __DEFAULT_FN_ATTRS _mm_cvtsd_si32(__m128d __a) {
/// Converts the lower double-precision floating-point element of a
/// 128-bit vector of [2 x double], in the second parameter, into a
-/// single-precision floating-point value, returned in the lower 32 bits of
-/// a 128-bit vector of [4 x float]. Rounds inexact results according to the
-/// rounding control bits in the MXCSR register. The upper 96 bits of the
-/// result vector are copied from the upper 96 bits of the first parameter.
+/// single-precision floating-point value, returned in the lower 32 bits of a
+/// 128-bit vector of [4 x float]. The upper 96 bits of the result vector are
+/// copied from the upper 96 bits of the first parameter.
///
/// \headerfile <x86intrin.h>
///
@@ -1416,11 +1413,12 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_cvtss_sd(__m128d __a,
/// Converts the two double-precision floating-point elements of a
/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
+/// returned in the lower 64 bits of a 128-bit vector of [4 x i32],
/// truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1438,9 +1436,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvttpd_epi32(__m128d __a) {
/// Converts the low-order element of a [2 x double] vector into a 32-bit
/// signed integer value, truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1456,13 +1454,12 @@ static __inline__ int __DEFAULT_FN_ATTRS _mm_cvttsd_si32(__m128d __a) {
}
/// Converts the two double-precision floating-point elements of a
-/// 128-bit vector of [2 x double] into two signed 32-bit integer values.
-/// Rounds inexact results according to the rounding control bits in the
-/// MXCSR register.
+/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
+/// returned in a 64-bit vector of [2 x i32].
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1477,11 +1474,11 @@ static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX _mm_cvtpd_pi32(__m128d __a) {
/// Converts the two double-precision floating-point elements of a
/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
-/// truncating inexact results.
+/// returned in a 64-bit vector of [2 x i32], truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3206,9 +3203,8 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cmplt_epi32(__m128i __a,
#ifdef __x86_64__
/// Converts a 64-bit signed integer value from the second operand into a
/// double-precision value and returns it in the lower element of a [2 x
-/// double] vector. Rounds inexact results according to the rounding control
-/// bits in the MXCSR register. The upper element of the returned vector is
-/// copied from the upper element of the first operand.
+/// double] vector; the upper element of the returned vector is copied from
+/// the upper element of the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -3229,12 +3225,11 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_cvtsi64_sd(__m128d __a,
}
/// Converts the first (lower) element of a vector of [2 x double] into a
-/// 64-bit signed integer value. Rounds inexact results according to the
-/// rounding control bits in the MXCSR register.
+/// 64-bit signed integer value.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 64-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3251,9 +3246,9 @@ static __inline__ long long __DEFAULT_FN_ATTRS _mm_cvtsd_si64(__m128d __a) {
/// Converts the first (lower) element of a vector of [2 x double] into a
/// 64-bit signed integer value, truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 64-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3269,8 +3264,7 @@ static __inline__ long long __DEFAULT_FN_ATTRS _mm_cvttsd_si64(__m128d __a) {
}
#endif
-/// Converts a vector of [4 x i32] into a vector of [4 x float]. Rounds inexact
-/// results according to the rounding control bits in the MXCSR register.
+/// Converts a vector of [4 x i32] into a vector of [4 x float].
///
/// \headerfile <x86intrin.h>
///
@@ -3283,12 +3277,11 @@ static __inline__ __m128 __DEFAULT_FN_ATTRS _mm_cvtepi32_ps(__m128i __a) {
return (__m128) __builtin_convertvector((__v4si)__a, __v4sf);
}
-/// Converts a vector of [4 x float] into a vector of [4 x i32]. Rounds inexact
-/// results according to the rounding control bits in the MXCSR register.
+/// Converts a vector of [4 x float] into a vector of [4 x i32].
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -3305,9 +3298,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvtps_epi32(__m128 __a) {
/// Converts a vector of [4 x float] into a vector of [4 x i32],
/// truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
diff --git a/clang/lib/Headers/xmmintrin.h b/clang/lib/Headers/xmmintrin.h
index 0feee669e1e818..3dd9f6841039b3 100644
--- a/clang/lib/Headers/xmmintrin.h
+++ b/clang/lib/Headers/xmmintrin.h
@@ -1295,12 +1295,11 @@ _mm_ucomineq_ss(__m128 __a, __m128 __b)
}
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer. Rounds an inexact result according to
-/// the rounding control bits in the MXCSR register.
+/// [4 x float] into a 32-bit integer.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1318,12 +1317,11 @@ _mm_cvtss_si32(__m128 __a)
}
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer. Rounds an inexact result according to
-/// the rounding control bits in the MXCSR register.
+/// [4 x float] into a 32-bit integer.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1343,12 +1341,11 @@ _mm_cvt_ss2si(__m128 __a)
#ifdef __x86_64__
/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 64-bit integer. Rounds an inexact result according to
-/// the rounding control bits in the MXCSR register.
+/// [4 x float] into a 64-bit integer.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1367,13 +1364,12 @@ _mm_cvtss_si64(__m128 __a)
#endif
-/// Converts two low-order float values in a 128-bit vector of [4 x float] into
-/// a 64-bit vector of [2 x i32]. Rounds inexact results according to the
-/// rounding control bits in the MXCSR register.
+/// Converts two low-order float values in a 128-bit vector of
+/// [4 x float] into a 64-bit vector of [2 x i32].
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1388,13 +1384,12 @@ _mm_cvtps_pi32(__m128 __a)
return (__m64)__builtin_ia32_cvtps2pi((__v4sf)__a);
}
-/// Converts two low-order float values in a 128-bit vector of [4 x float] into
-/// a 64-bit vector of [2 x i32]. Rounds inexact results according to the
-/// rounding control bits in the MXCSR register.
+/// Converts two low-order float values in a 128-bit vector of
+/// [4 x float] into a 64-bit vector of [2 x i32].
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1412,9 +1407,9 @@ _mm_cvt_ps2pi(__m128 __a)
/// Converts a float value contained in the lower 32 bits of a vector of
/// [4 x float] into a 32-bit integer, truncating an inexact result.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1434,9 +1429,9 @@ _mm_cvttss_si32(__m128 __a)
/// Converts a float value contained in the lower 32 bits of a vector of
/// [4 x float] into a 32-bit integer, truncating an inexact result.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1457,9 +1452,9 @@ _mm_cvtt_ss2si(__m128 __a)
/// Converts a float value contained in the lower 32 bits of a vector of
/// [4 x float] into a 64-bit integer, truncating an inexact result.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If the converted value does not fit in a 64-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1481,9 +1476,9 @@ _mm_cvttss_si64(__m128 __a)
/// [4 x float] into a 64-bit vector of [2 x i32], truncating inexact
/// results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1502,9 +1497,9 @@ _mm_cvttps_pi32(__m128 __a)
/// Converts two low-order float values in a 128-bit vector of [4 x
/// float] into a 64-bit vector of [2 x i32], truncating inexact results.
///
-/// If a converted value is larger than the maximum possible result,
-/// raises a floating-point invalid exception. If the exception is
-/// masked, returns the most negative integer.
+/// If a converted value does not fit in a 32-bit integer, raises a
+/// floating-point invalid exception. If the exception is masked, returns
+/// the most negative integer.
///
/// \headerfile <x86intrin.h>
///
@@ -1519,11 +1514,10 @@ _mm_cvtt_ps2pi(__m128 __a)
return _mm_cvttps_pi32(__a);
}
-/// Converts a 32-bit signed integer value into a floating point value and
-/// writes it to the lower 32 bits of the destination. Rounds an inexact
-/// result according to the rounding control bits in the MXCSR register.
-/// The remaining higher order elements of the destination vector are copied
-/// from the corresponding elements in the first operand.
+/// Converts a 32-bit signed integer value into a floating point value
+/// and writes it to the lower 32 bits of the destination. The remaining
+/// higher order elements of the destination vector are copied from the
+/// corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1543,11 +1537,10 @@ _mm_cvtsi32_ss(__m128 __a, int __b)
return __a;
}
-/// Converts a 32-bit signed integer value into a floating point value and
-/// writes it to the lower 32 bits of the destination. Rounds an inexact
-/// result according to the rounding control bits in the MXCSR register.
-/// The remaining higher order elements of the destination are copied from
-/// the corresponding elements in the first operand.
+/// Converts a 32-bit signed integer value into a floating point value
+/// and writes it to the lower 32 bits of the destination. The remaining
+/// higher order elements of the destination are copied from the
+/// corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1568,11 +1561,10 @@ _mm_cvt_si2ss(__m128 __a, int __b)
#ifdef __x86_64__
-/// Converts a 64-bit signed integer value into a floating point value and
-/// writes it to the lower 32 bits of the destination. Rounds an inexact
-/// result according to the rounding control bits in the MXCSR register.
-/// The remaining higher order elements of the destination are copied from
-/// the corresponding elements in the first operand.
+/// Converts a 64-bit signed integer value into a floating point value
+/// and writes it to the lower 32 bits of the destination. The remaining
+/// higher order elements of the destination are copied from the
+/// corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1596,10 +1588,8 @@ _mm_cvtsi64_ss(__m128 __a, long long __b)
/// Converts two elements of a 64-bit vector of [2 x i32] into two
/// floating point values and writes them to the lower 64-bits of the
-/// destination. Rounds inexact results according to the rounding control
-/// bits in the MXCSR register. The remaining higher order elements of the
-/// destination are copied from the corresponding elements in the first
-/// operand.
+/// destination. The remaining higher order elements of the destination are
+/// copied from the corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -1621,10 +1611,8 @@ _mm_cvtpi32_ps(__m128 __a, __m64 __b)
/// Converts two elements of a 64-bit vector of [2 x i32] into two
/// floating point values and writes them to the lower 64-bits of the
-/// destination. Rounds inexact results according to the rounding control
-/// bits in the MXCSR register. The remaining higher order elements of the
-/// destination are copied from the corresponding elements in the first
-/// operand.
+/// destination. The remaining higher order elements of the destination are
+/// copied from the corresponding elements in the first operand.
///
/// \headerfile <x86intrin.h>
///
@@ -2883,9 +2871,7 @@ _mm_cvtpu8_ps(__m64 __a)
}
/// Converts the two 32-bit signed integer values from each 64-bit vector
-/// operand of [2 x i32] into a 128-bit vector of [4 x float]. Rounds
-/// inexact results according to the rounding control bits in the MXCSR
-/// register.
+/// operand of [2 x i32] into a 128-bit vector of [4 x float].
///
/// \headerfile <x86intrin.h>
///
@@ -2914,9 +2900,7 @@ _mm_cvtpi32x2_ps(__m64 __a, __m64 __b)
/// Converts each single-precision floating-point element of a 128-bit
/// floating-point vector of [4 x float] into a 16-bit signed integer, and
-/// packs the results into a 64-bit integer vector of [4 x i16]. Rounds
-/// inexact results according to the rounding control bits in the MXCSR
-/// register.
+/// packs the results into a 64-bit integer vector of [4 x i16].
///
/// If the floating-point element is NaN or infinity, or if the
/// floating-point element is greater than 0x7FFFFFFF or less than -0x8000,
@@ -2946,8 +2930,7 @@ _mm_cvtps_pi16(__m128 __a)
/// Converts each single-precision floating-point element of a 128-bit
/// floating-point vector of [4 x float] into an 8-bit signed integer, and
/// packs the results into the lower 32 bits of a 64-bit integer vector of
-/// [8 x i8]. The upper 32 bits of the vector are set to 0. Rounds inexact
-/// results according to the rounding control bits in the MXCSR register.
+/// [8 x i8]. The upper 32 bits of the vector are set to 0.
///
/// If the floating-point element is NaN or infinity, or if the
/// floating-point element is greater than 0x7FFFFFFF or less than -0x80, it
>From bdc909605b93d9f1be9d9baf7923ae1d325d5776 Mon Sep 17 00:00:00 2001
From: Paul Robinson <paul.robinson at sony.com>
Date: Wed, 13 Mar 2024 08:45:40 -0700
Subject: [PATCH 3/4] Rephrase the truncation info.
---
clang/lib/Headers/avxintrin.h | 8 ++++----
clang/lib/Headers/emmintrin.h | 18 +++++++++---------
clang/lib/Headers/xmmintrin.h | 23 ++++++++++++-----------
3 files changed, 25 insertions(+), 24 deletions(-)
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index 11c1dfa79af37c..c9de580a1de75e 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -2246,8 +2246,8 @@ _mm256_cvtps_pd(__m128 __a)
return (__m256d)__builtin_convertvector((__v4sf)__a, __v4df);
}
-/// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4
-/// x i32], truncating inexact results.
+/// Converts a 256-bit vector of [4 x double] into signed truncated (rounded
+/// toward zero) 32-bit integers returned in a 128-bit vector of [4 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -2286,8 +2286,8 @@ _mm256_cvtpd_epi32(__m256d __a)
return (__m128i)__builtin_ia32_cvtpd2dq256((__v4df) __a);
}
-/// Converts a vector of [8 x float] into a vector of [8 x i32],
-/// truncating inexact results.
+/// Converts a vector of [8 x float] into signed truncated (rounded toward
+/// zero) 32-bit integers returned in a vector of [8 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h
index a934a078af51f7..da56a359d1778d 100644
--- a/clang/lib/Headers/emmintrin.h
+++ b/clang/lib/Headers/emmintrin.h
@@ -1412,9 +1412,9 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_cvtss_sd(__m128d __a,
}
/// Converts the two double-precision floating-point elements of a
-/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
-/// returned in the lower 64 bits of a 128-bit vector of [4 x i32],
-/// truncating inexact results.
+/// 128-bit vector of [2 x double] into two signed truncated (rounded
+/// toward zero) 32-bit integer values, returned in the lower 64 bits
+/// of a 128-bit vector of [4 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -1434,7 +1434,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvttpd_epi32(__m128d __a) {
}
/// Converts the low-order element of a [2 x double] vector into a 32-bit
-/// signed integer value, truncating inexact results.
+/// signed truncated (rounded toward zero) integer value.
///
/// If the converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -1473,8 +1473,8 @@ static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX _mm_cvtpd_pi32(__m128d __a) {
}
/// Converts the two double-precision floating-point elements of a
-/// 128-bit vector of [2 x double] into two signed 32-bit integer values,
-/// returned in a 64-bit vector of [2 x i32], truncating inexact results.
+/// 128-bit vector of [2 x double] into two signed truncated (rounded toward
+/// zero) 32-bit integer values, returned in a 64-bit vector of [2 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -3244,7 +3244,7 @@ static __inline__ long long __DEFAULT_FN_ATTRS _mm_cvtsd_si64(__m128d __a) {
}
/// Converts the first (lower) element of a vector of [2 x double] into a
-/// 64-bit signed integer value, truncating inexact results.
+/// 64-bit signed truncated (rounded toward zero) integer value.
///
/// If a converted value does not fit in a 64-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -3295,8 +3295,8 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cvtps_epi32(__m128 __a) {
return (__m128i)__builtin_ia32_cvtps2dq((__v4sf)__a);
}
-/// Converts a vector of [4 x float] into a vector of [4 x i32],
-/// truncating inexact results.
+/// Converts a vector of [4 x float] into four signed truncated (rounded toward
+/// zero) 32-bit integers, returned in a vector of [4 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
diff --git a/clang/lib/Headers/xmmintrin.h b/clang/lib/Headers/xmmintrin.h
index 3dd9f6841039b3..feca532925257a 100644
--- a/clang/lib/Headers/xmmintrin.h
+++ b/clang/lib/Headers/xmmintrin.h
@@ -1404,8 +1404,8 @@ _mm_cvt_ps2pi(__m128 __a)
return _mm_cvtps_pi32(__a);
}
-/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer, truncating an inexact result.
+/// Converts the lower (first) element of a vector of [4 x float] into a signed
+/// truncated (rounded toward zero) 32-bit integer.
///
/// If the converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -1426,8 +1426,8 @@ _mm_cvttss_si32(__m128 __a)
return __builtin_ia32_cvttss2si((__v4sf)__a);
}
-/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 32-bit integer, truncating an inexact result.
+/// Converts the lower (first) element of a vector of [4 x float] into a signed
+/// truncated (rounded toward zero) 32-bit integer.
///
/// If the converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -1449,8 +1449,8 @@ _mm_cvtt_ss2si(__m128 __a)
}
#ifdef __x86_64__
-/// Converts a float value contained in the lower 32 bits of a vector of
-/// [4 x float] into a 64-bit integer, truncating an inexact result.
+/// Converts the lower (first) element of a vector of [4 x float] into a signed
+/// truncated (rounded toward zero) 64-bit integer.
///
/// If the converted value does not fit in a 64-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -1472,9 +1472,9 @@ _mm_cvttss_si64(__m128 __a)
}
#endif
-/// Converts two low-order float values in a 128-bit vector of
-/// [4 x float] into a 64-bit vector of [2 x i32], truncating inexact
-/// results.
+/// Converts the lower (first) two elements of a 128-bit vector of [4 x float]
+/// into two signed truncated (rounded toward zero) 32-bit integers,
+/// returned in a 64-bit vector of [2 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -1494,8 +1494,9 @@ _mm_cvttps_pi32(__m128 __a)
return (__m64)__builtin_ia32_cvttps2pi((__v4sf)__a);
}
-/// Converts two low-order float values in a 128-bit vector of [4 x
-/// float] into a 64-bit vector of [2 x i32], truncating inexact results.
+/// Converts the lower (first) two elements of a 128-bit vector of [4 x float]
+/// into two signed truncated (rounded toward zero) 64-bit integers,
+/// returned in a 64-bit vector of [2 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
>From c8b3c498ef5ba9496e10852c3e91c01c537d7215 Mon Sep 17 00:00:00 2001
From: Paul Robinson <paul.robinson at sony.com>
Date: Wed, 13 Mar 2024 09:00:00 -0700
Subject: [PATCH 4/4] Minor additional changes.
---
clang/lib/Headers/avxintrin.h | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index c9de580a1de75e..4b2c65d4d6ea2f 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -2246,8 +2246,9 @@ _mm256_cvtps_pd(__m128 __a)
return (__m256d)__builtin_convertvector((__v4sf)__a, __v4df);
}
-/// Converts a 256-bit vector of [4 x double] into signed truncated (rounded
-/// toward zero) 32-bit integers returned in a 128-bit vector of [4 x i32].
+/// Converts a 256-bit vector of [4 x double] into four signed truncated
+/// (rounded toward zero) 32-bit integers returned in a 128-bit vector of
+/// [4 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
@@ -2286,8 +2287,8 @@ _mm256_cvtpd_epi32(__m256d __a)
return (__m128i)__builtin_ia32_cvtpd2dq256((__v4df) __a);
}
-/// Converts a vector of [8 x float] into signed truncated (rounded toward
-/// zero) 32-bit integers returned in a vector of [8 x i32].
+/// Converts a vector of [8 x float] into eight signed truncated (rounded
+/// toward zero) 32-bit integers returned in a vector of [8 x i32].
///
/// If a converted value does not fit in a 32-bit integer, raises a
/// floating-point invalid exception. If the exception is masked, returns
More information about the cfe-commits
mailing list