maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/lib/Headers/fmaintrin.h

231 lines
7.7 KiB
C
Raw Normal View History

[X86] Fix the file header name on fmaintrin.h llvm-svn: 332108
2018-05-12 01:37:40 +08:00
/*===---- fmaintrin.h - FMA intrinsics -------------------------------------===
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*
*===-----------------------------------------------------------------------===
*/
#ifndef __IMMINTRIN_H
#error "Never use <fmaintrin.h> directly; include <immintrin.h> instead."
#endif
#ifndef __FMAINTRIN_H
#define __FMAINTRIN_H
Use a define for per-file function attributes for the Intel intrinsic headers. This is a precursor to changing them to use the new target attribute code. llvm-svn: 239882
2015-06-17 15:09:20 +08:00
/* Define the default attributes for the functions in this file. */
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
#define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("fma"), __min_vector_width__(128)))
#define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("fma"), __min_vector_width__(256)))
Use a define for per-file function attributes for the Intel intrinsic headers. This is a precursor to changing them to use the new target attribute code. llvm-svn: 239882
2015-06-17 15:09:20 +08:00
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmadd_ps(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmadd_pd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmadd_ss(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Rename the VEX scalar fma builtins to end with a '3' to match gcc I think we need to use different builtins for the FMA4 instructions since those instructions zero the upper bits and FMA3 instructions pass the bits through. So this moves the existing builtins to be the FMA3 versions. New versions will be added for FMA4. llvm-svn: 317766
2017-11-09 12:10:46 +08:00
return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmadd_sd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Rename the VEX scalar fma builtins to end with a '3' to match gcc I think we need to use different builtins for the FMA4 instructions since those instructions zero the upper bits and FMA3 instructions pass the bits through. So this moves the existing builtins to be the FMA3 versions. New versions will be added for FMA4. llvm-svn: 317766
2017-11-09 12:10:46 +08:00
return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, (__v2df)__B, (__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmsub_ps(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmsub_pd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmsub_ss(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Use separate builtins for fma4 scalar intrinsics. Use negations to remove some of the scalar fma3 builtins. fma4 instructions zero the upper bits of the xmm register. fma3 instructions leave the bits unmodified. This requires separate builtins for the different semantics. While we're cleaning up the scalar builtins this also removes the fma3 fmsub/fnmadd/fnmsub builtins by using negates in the header file. llvm-svn: 318985
2017-11-26 03:32:12 +08:00
return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmsub_sd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Use separate builtins for fma4 scalar intrinsics. Use negations to remove some of the scalar fma3 builtins. fma4 instructions zero the upper bits of the xmm register. fma3 instructions leave the bits unmodified. This requires separate builtins for the different semantics. While we're cleaning up the scalar builtins this also removes the fma3 fmsub/fnmadd/fnmsub builtins by using negates in the header file. llvm-svn: 318985
2017-11-26 03:32:12 +08:00
return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmadd_ps(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmadd_pd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmadd_ss(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Use separate builtins for fma4 scalar intrinsics. Use negations to remove some of the scalar fma3 builtins. fma4 instructions zero the upper bits of the xmm register. fma3 instructions leave the bits unmodified. This requires separate builtins for the different semantics. While we're cleaning up the scalar builtins this also removes the fma3 fmsub/fnmadd/fnmsub builtins by using negates in the header file. llvm-svn: 318985
2017-11-26 03:32:12 +08:00
return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, -(__v4sf)__B, (__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmadd_sd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Use separate builtins for fma4 scalar intrinsics. Use negations to remove some of the scalar fma3 builtins. fma4 instructions zero the upper bits of the xmm register. fma3 instructions leave the bits unmodified. This requires separate builtins for the different semantics. While we're cleaning up the scalar builtins this also removes the fma3 fmsub/fnmadd/fnmsub builtins by using negates in the header file. llvm-svn: 318985
2017-11-26 03:32:12 +08:00
return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, -(__v2df)__B, (__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmsub_ps(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmsub_pd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmsub_ss(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Use separate builtins for fma4 scalar intrinsics. Use negations to remove some of the scalar fma3 builtins. fma4 instructions zero the upper bits of the xmm register. fma3 instructions leave the bits unmodified. This requires separate builtins for the different semantics. While we're cleaning up the scalar builtins this also removes the fma3 fmsub/fnmadd/fnmsub builtins by using negates in the header file. llvm-svn: 318985
2017-11-26 03:32:12 +08:00
return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, -(__v4sf)__B, -(__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fnmsub_sd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Use separate builtins for fma4 scalar intrinsics. Use negations to remove some of the scalar fma3 builtins. fma4 instructions zero the upper bits of the xmm register. fma3 instructions leave the bits unmodified. This requires separate builtins for the different semantics. While we're cleaning up the scalar builtins this also removes the fma3 fmsub/fnmadd/fnmsub builtins by using negates in the header file. llvm-svn: 318985
2017-11-26 03:32:12 +08:00
return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, -(__v2df)__B, -(__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmaddsub_ps(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmaddsub_pd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128 __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmsubadd_ps(__m128 __A, __m128 __B, __m128 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m128d __DEFAULT_FN_ATTRS128
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm_fmsubadd_pd(__m128d __A, __m128d __B, __m128d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256 __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmadd_ps(__m256 __A, __m256 __B, __m256 __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256d __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmadd_pd(__m256d __A, __m256d __B, __m256d __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256 __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmsub_ps(__m256 __A, __m256 __B, __m256 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256d __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmsub_pd(__m256d __A, __m256d __B, __m256d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256 __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fnmadd_ps(__m256 __A, __m256 __B, __m256 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256d __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fnmadd_pd(__m256d __A, __m256d __B, __m256d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256 __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fnmsub_ps(__m256 __A, __m256 __B, __m256 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256d __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fnmsub_pd(__m256d __A, __m256d __B, __m256d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256 __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmaddsub_ps(__m256 __A, __m256 __B, __m256 __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256d __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmaddsub_pd(__m256d __A, __m256d __B, __m256d __C)
{
[X86] Add typecasts to remove most assumptions about what __m128i/__m256i is defined as. Add similar typecasts for the fp types as well. llvm-svn: 269632
2016-05-16 14:38:42 +08:00
return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256 __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmsubadd_ps(__m256 __A, __m256 __B, __m256 __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
static __inline__ __m256d __DEFAULT_FN_ATTRS256
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
_mm256_fmsubadd_pd(__m256d __A, __m256d __B, __m256d __C)
{
[X86] Reduce the number of FMA builtins needed by the frontend by adding negates to operands of the fmadd and fmaddsub builtins. The backend should be able to combine the negates to create fmsub, fnmadd, and fnmsub. faddsub converting to fsubadd still needs work I think, but should be very doable. This matches what we already do for the masked builtins. This only covers the packed builtins. Scalar builtins will be done after FMA4 is fixed. llvm-svn: 317873
2017-11-10 13:20:32 +08:00
return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
}
[Builtins][Attributes][X86] Tag all X86 builtins with their required vector width. Add a min_vector_width function attribute and tag all x86 instrinsics with it This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter. Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible. To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future. To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin. To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute. To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins. There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch. Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue. Differential Revision: https://reviews.llvm.org/D48617 llvm-svn: 336583
2018-07-10 03:00:16 +08:00
#undef __DEFAULT_FN_ATTRS128
#undef __DEFAULT_FN_ATTRS256
Use a define for per-file function attributes for the Intel intrinsic headers. This is a precursor to changing them to use the new target attribute code. llvm-svn: 239882
2015-06-17 15:09:20 +08:00
Add fma3 intrinsic header file. llvm-svn: 157913
2012-06-04 11:42:47 +08:00
#endif /* __FMAINTRIN_H */