maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/Driver/arm-mfpu.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

443 lines
24 KiB
C
Raw Normal View History

More tests for ARM FPU features. Also remove the svn:eol-style property from the test file. llvm-svn: 147947
2012-01-11 20:36:39 +08:00
// Test that different values of -mfpu pick correct ARM FPU target-feature(s).
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi %s -### -o %t.o 2>&1 \
More tests for ARM FPU features. Also remove the svn:eol-style property from the test file. llvm-svn: 147947
2012-01-11 20:36:39 +08:00
// RUN: | FileCheck --check-prefix=CHECK-DEFAULT %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// CHECK-DEFAULT-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-DEFAULT-DAG: "-target-feature" "+soft-float-abi"
More tests for ARM FPU features. Also remove the svn:eol-style property from the test file. llvm-svn: 147947
2012-01-11 20:36:39 +08:00
// CHECK-DEFAULT-NOT: "-target-feature" "+vfp2"
// CHECK-DEFAULT-NOT: "-target-feature" "+vfp3"
// CHECK-DEFAULT-NOT: "-target-feature" "+neon"
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fpa %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-FPA %s
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fpe2 %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-FPA %s
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fpe3 %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-FPA %s
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=maverick %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-FPA %s
[ARM] Use TargetParser to determine FPU subtarget features The main effect of this is to fix anomalies where certain -mfpu options didn't disable everything that they should causing strange behaviour when combined with -mcpu or -march values that themselves enabled fpu subtarget features, e.g. -mfpu=fpv5-dp-d16 with -march=armv7em previously behaved the same as -mfpu=fpv5-sp-d16 due to fp-only-sp not being disabled. Invalid -mfpu options now also give an error, which is consistent with the handling of the .fpu directive. Differential Revision: http://reviews.llvm.org/D10239 llvm-svn: 239152
2015-06-05 21:34:11 +08:00
// CHECK-FPA: error: {{.*}} does not support '-mfpu={{fpa|fpe|fpe2|fpe3|maverick}}'
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfp %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-VFP %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfp %s -mfloat-abi=soft -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-2 %s
// CHECK-VFP-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP-DAG: "-target-feature" "+vfp2"
// CHECK-VFP-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-VFP-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-VFP-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-2-DAG: "-target-feature" "+soft-float-abi"
// CHECK-SOFT-ABI-FP-2-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-SOFT-ABI-FP-2-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-SOFT-ABI-FP-2-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-SOFT-ABI-FP-2-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-2-DAG: "-target-feature" "-crypto"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-2-DAG: "-target-feature" "-vfp2sp"
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfp3 %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-VFP3 %s
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3 %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-VFP3 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-3 %s
// CHECK-VFP3-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP3-DAG: "-target-feature" "+vfp3"
// CHECK-VFP3-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-VFP3-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP3-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-3-DAG: "-target-feature" "+soft-float-abi"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-3-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-3-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-SOFT-ABI-FP-3-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-SOFT-ABI-FP-3-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-3-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-3-DAG: "-target-feature" "-vfp3d16sp"
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3-fp16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP3-FP16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3-fp16 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-3 %s
// CHECK-VFP3-FP16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3-FP16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP3-FP16-DAG: "-target-feature" "+vfp3"
// CHECK-VFP3-FP16-DAG: "-target-feature" "+fp16"
// CHECK-VFP3-FP16-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-VFP3-FP16-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP3-FP16-DAG: "-target-feature" "+fp64"
// CHECK-VFP3-FP16-DAG: "-target-feature" "+d32"
// CHECK-VFP3-FP16-DAG: "-target-feature" "-neon"
// CHECK-VFP3-FP16-DAG: "-target-feature" "-crypto"
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
Add -mfpu=neon-vfpv4 This enables user to architecturally specify ARMv7A + VFPv4 + NEON. Change-Id: I779b01fef5c47e5e4ac702ae24ed2f76a0e4c63f llvm-svn: 222932
2014-11-29 04:39:54 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfp3-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP3-D16 %s
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP3-D16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3-d16 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-3 %s
// CHECK-VFP3-D16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3-D16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP3-D16-DAG: "-target-feature" "+vfp3d16"
// CHECK-VFP3-D16-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-VFP3-D16-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP3-D16-DAG: "-target-feature" "+fp64"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-VFP3-D16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3-D16-DAG: "-target-feature" "-neon"
Add -mfpu=neon-vfpv4 This enables user to architecturally specify ARMv7A + VFPv4 + NEON. Change-Id: I779b01fef5c47e5e4ac702ae24ed2f76a0e4c63f llvm-svn: 222932
2014-11-29 04:39:54 +08:00
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3-d16-fp16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP3-D16-FP16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3-d16-fp16 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-3 %s
// CHECK-VFP3-D16-FP16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "+vfp3d16"
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "+fp16"
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "+fp64"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "-neon"
// CHECK-VFP3-D16-FP16-DAG: "-target-feature" "-crypto"
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3xd %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP3XD %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3xd -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-3 %s
// CHECK-VFP3XD-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3XD-DAG: "-target-feature" "+soft-float-abi"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-VFP3XD-DAG: "-target-feature" "-fp64"
// CHECK-VFP3XD-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3XD-DAG: "-target-feature" "+vfp3d16sp"
// CHECK-VFP3XD-DAG: "-target-feature" "-fp16"
// CHECK-VFP3XD-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-VFP3XD-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP3XD-DAG: "-target-feature" "-neon"
// CHECK-VFP3XD-DAG: "-target-feature" "-crypto"
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3xd-fp16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP3XD-FP16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv3xd-fp16 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-3 %s
// CHECK-VFP3XD-FP16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "+vfp3d16sp"
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "+fp16"
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "-fp-armv8d16sp"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "-fp64"
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "-neon"
// CHECK-VFP3XD-FP16-DAG: "-target-feature" "-crypto"
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
[ARM] add basic support for Cortex-A7 and VFPv4 to Clang llvm-svn: 195359
2013-11-21 22:04:38 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfp4 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP4 %s
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv4 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP4 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv4 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-4 %s
// CHECK-VFP4-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP4-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP4-DAG: "-target-feature" "+vfp4"
// CHECK-VFP4-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP4-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-4-DAG: "-target-feature" "+soft-float-abi"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-4-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-4-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-SOFT-ABI-FP-4-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-SOFT-ABI-FP-4-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-4-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-4-DAG: "-target-feature" "-vfp4d16sp"
[ARM] add basic support for Cortex-A7 and VFPv4 to Clang llvm-svn: 195359
2013-11-21 22:04:38 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfp4-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP4-D16 %s
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv4-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-VFP4-D16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=vfpv4-d16 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-4 %s
// CHECK-VFP4-D16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP4-D16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-VFP4-D16-DAG: "-target-feature" "+vfp4d16"
// CHECK-VFP4-D16-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-VFP4-D16-DAG: "-target-feature" "+fp64"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-VFP4-D16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-VFP4-D16-DAG: "-target-feature" "-neon"
[ARM] add basic support for Cortex-A7 and VFPv4 to Clang llvm-svn: 195359
2013-11-21 22:04:38 +08:00
Add support for FPv4-SP to the clang driver Added two new options for -mfpu when targetting ARM: * fpv4-sp-d16 * fp4-sp-d16 The first is the same spelling as gcc. The lack of a leading `v' is correct, this is consistent with ARM's documentation and gcc's spelling of the option. llvm-svn: 201846
2014-02-21 18:39:15 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fp4-sp-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FP4-SP-D16 %s
// RUN: %clang -target arm-linux-eabi -mfpu=fpv4-sp-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FP4-SP-D16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fpv4-sp-d16 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-4 %s
// CHECK-FP4-SP-D16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-FP4-SP-D16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-FP4-SP-D16-DAG: "-target-feature" "+vfp4d16sp"
// CHECK-FP4-SP-D16-DAG: "-target-feature" "-fp-armv8d16sp"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-FP4-SP-D16-DAG: "-target-feature" "-fp64"
// CHECK-FP4-SP-D16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-FP4-SP-D16-DAG: "-target-feature" "-neon"
[ARM] add basic support for Cortex-A7 and VFPv4 to Clang llvm-svn: 195359
2013-11-21 22:04:38 +08:00
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modeled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. llvm-svn: 218748
2014-10-01 17:03:02 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fp5-sp-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FP5-SP-D16 %s
// RUN: %clang -target arm-linux-eabi -mfpu=fpv5-sp-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FP5-SP-D16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fp-armv8-sp-d16 -mfloat-abi=soft %s -### -o %t.o \
// RUN: 2>&1 | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// CHECK-FP5-SP-D16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-FP5-SP-D16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-FP5-SP-D16-DAG: "-target-feature" "+fp-armv8d16sp"
// CHECK-FP5-SP-D16-DAG: "-target-feature" "-neon"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-FP5-SP-D16-DAG: "-target-feature" "-fp64"
// CHECK-FP5-SP-D16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-FP5-SP-D16-DAG: "-target-feature" "-crypto"
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modeled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. llvm-svn: 218748
2014-10-01 17:03:02 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fp5-dp-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FP5-DP-D16 %s
// RUN: %clang -target arm-linux-eabi -mfpu=fpv5-dp-d16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FP5-DP-D16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=fpv5-dp-d16 %s -mfloat-abi=soft -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-5 %s
// CHECK-FP5-DP-D16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-FP5-DP-D16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-FP5-DP-D16-DAG: "-target-feature" "+fp-armv8d16"
// CHECK-FP5-DP-D16-DAG: "-target-feature" "+fp64"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-FP5-DP-D16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-FP5-DP-D16-DAG: "-target-feature" "-neon"
// CHECK-FP5-DP-D16-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "+soft-float"
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "+soft-float-abi"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-5-DAG: "-target-feature" "-fp-armv8d16sp"
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modeled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. llvm-svn: 218748
2014-10-01 17:03:02 +08:00
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon %s -### -o %t.o 2>&1 \
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
// RUN: | FileCheck --check-prefix=CHECK-NEON %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-6 %s
// CHECK-NEON-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NEON-DAG: "-target-feature" "+neon"
// CHECK-SOFT-ABI-FP-6-DAG: "-target-feature" "+soft-float-abi"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-6-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-6-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-SOFT-ABI-FP-6-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-SOFT-ABI-FP-6-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-6-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-SOFT-ABI-FP-6-DAG: "-target-feature" "-neon"
Fix -mfpu parsing on ARM. - Support gcc-compatible vfpv3 name in addition to vfp3. - Support vfpv3-d16. - Disable neon feature for -mfpu=vfp* (yes, we were emitting Neon instructions for those!). llvm-svn: 147943
2012-01-11 19:21:31 +08:00
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon-fp16 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-NEON-FP16 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon-fp16 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-6 %s
// CHECK-NEON-FP16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NEON-FP16-DAG: "-target-feature" "+soft-float-abi"
// CHECK-NEON-FP16-DAG: "-target-feature" "+vfp3"
// CHECK-NEON-FP16-DAG: "-target-feature" "+fp16"
// CHECK-NEON-FP16-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-NEON-FP16-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-NEON-FP16-DAG: "-target-feature" "+fp64"
// CHECK-NEON-FP16-DAG: "-target-feature" "+d32"
// CHECK-NEON-FP16-DAG: "-target-feature" "+neon"
// CHECK-NEON-FP16-DAG: "-target-feature" "-crypto"
[ARM]: Extend -mfpu options for half-precision and vfpv3xd Some of the the permissible ARM -mfpu options, which are supported in GCC, are currently not present in llvm/clang.This patch adds the options: 'neon-fp16', 'vfpv3-fp16', 'vfpv3-d16-fp16', 'vfpv3xd' and 'vfpv3xd-fp16. These are related to half-precision floating-point and single precision. Reviewers: rengolin, ranjeet.singh Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D10764 llvm-svn: 240929
2015-06-29 17:30:19 +08:00
Add additional arguments for -mfpu options Add neon-vfpv3 to allow specifying both at the same time. This is not an option that GCC supports, but follows the same track and should be non-controversial. Change-Id: Id9ec157c835937d7d11ad0f49dbe5171fac17658 llvm-svn: 222933
2014-11-29 04:39:59 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon-vfpv3 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-NEON-VFPV3 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon-vfpv3 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-6 %s
// CHECK-NEON-VFPV3-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NEON-VFPV3-DAG: "-target-feature" "+soft-float-abi"
// CHECK-NEON-VFPV3-DAG: "-target-feature" "+vfp3"
// CHECK-NEON-VFPV3-DAG: "-target-feature" "+neon"
Add additional arguments for -mfpu options Add neon-vfpv3 to allow specifying both at the same time. This is not an option that GCC supports, but follows the same track and should be non-controversial. Change-Id: Id9ec157c835937d7d11ad0f49dbe5171fac17658 llvm-svn: 222933
2014-11-29 04:39:59 +08:00
Add -mfpu=neon-vfpv4 This enables user to architecturally specify ARMv7A + VFPv4 + NEON. Change-Id: I779b01fef5c47e5e4ac702ae24ed2f76a0e4c63f llvm-svn: 222932
2014-11-29 04:39:54 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon-vfpv4 %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-NEON-VFPV4 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target arm-linux-eabi -mfpu=neon-vfpv4 -mfloat-abi=soft %s -### -o %t.o 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-7 %s
// CHECK-NEON-VFPV4-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NEON-VFPV4-DAG: "-target-feature" "+soft-float-abi"
// CHECK-NEON-VFPV4-DAG: "-target-feature" "+vfp4"
// CHECK-NEON-VFPV4-DAG: "-target-feature" "+neon"
// CHECK-SOFT-ABI-FP-7-DAG: "-target-feature" "+soft-float-abi"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-7-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-7-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-SOFT-ABI-FP-7-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-SOFT-ABI-FP-7-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-7-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-SOFT-ABI-FP-7-DAG: "-target-feature" "-neon"
Add -mfpu=neon-vfpv4 This enables user to architecturally specify ARMv7A + VFPv4 + NEON. Change-Id: I779b01fef5c47e5e4ac702ae24ed2f76a0e4c63f llvm-svn: 222932
2014-11-29 04:39:54 +08:00
rename -ccc-host-triple into -target llvm-svn: 148582
2012-01-21 06:01:23 +08:00
// RUN: %clang -target arm-linux-eabi -msoft-float %s -### -o %t.o 2>&1 \
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
[ARM] Fix NEON wrongly being enabled with soft-float when targeting armv8/Cortex-A53/A57. This was caused by r200708 which enabled the crypto feature for these cores. llvm-svn: 201223
2014-02-12 18:22:35 +08:00
// RUN: %clang -target armv8 %s -### 2>&1 \
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv8a -mfpu=neon %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP-8 %s
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-8-DAG: "-target-feature" "+soft-float-abi"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-8-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-8-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-SOFT-ABI-FP-8-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-SOFT-ABI-FP-8-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-8-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-SOFT-ABI-FP-8-DAG: "-target-feature" "-neon"
[ARM] Fix NEON wrongly being enabled with soft-float when targeting armv8/Cortex-A53/A57. This was caused by r200708 which enabled the crypto feature for these cores. llvm-svn: 201223
2014-02-12 18:22:35 +08:00
Add driver support for FP, SIMD and crypto defaults. Although we wire up a bit for v8fp for macro setting purposes, we don't set a macro yet. Need to ask list about that. Change-Id: Ic9819593ce00882fbec72757ffccc6f0b18160a0 llvm-svn: 193367
2013-10-25 02:32:51 +08:00
// RUN: %clang -target armv8 -mfpu=fp-armv8 %s -### 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-ARMV8-SOFT-FLOAT %s
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-ARMV8-SOFT-FLOAT-DAG: "-target-feature" "+soft-float"
// CHECK-ARMV8-SOFT-FLOAT-DAG: "-target-feature" "+soft-float-abi"
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// NOT-CHECK-ARMV8-SOFT-FLOAT: "-target-feature" "+fp-armv8"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-ARMV9-SOFT-FLOAT-DAG: "-target-feature" "-neon"
// CHECK-ARMV8-SOFT-FLOAT-DAG: "-target-feature" "-crypto"
Add driver support for FP, SIMD and crypto defaults. Although we wire up a bit for v8fp for macro setting purposes, we don't set a macro yet. Need to ask list about that. Change-Id: Ic9819593ce00882fbec72757ffccc6f0b18160a0 llvm-svn: 193367
2013-10-25 02:32:51 +08:00
Add support for passing v8fp options via -mfpu. llvm-svn: 185075
2013-06-27 21:19:54 +08:00
// RUN: %clang -target armv8-linux-gnueabihf -mfpu=fp-armv8 %s -### 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FP-ARMV8 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// CHECK-FP-ARMV8-NOT: "-target-feature" "+soft-float"
// CHECK-FP-ARMV8-NOT: "-target-feature" "+soft-float-abi"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-FP-ARMV8-DAG: "-target-feature" "+fp-armv8"
// CHECK-FP-ARMV8-DAG: "-target-feature" "-neon"
// CHECK-FP-ARMV8-DAG: "-target-feature" "-crypto"
Add support for passing v8fp options via -mfpu. llvm-svn: 185075
2013-06-27 21:19:54 +08:00
// RUN: %clang -target armv8-linux-gnueabihf -mfpu=neon-fp-armv8 %s -### 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-NEON-FP-ARMV8 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// CHECK-NEON-FP-ARMV8-NOT: "-target-feature" "+soft-float"
// CHECK-NEON-FP-ARMV8-NOT: "-target-feature" "+soft-float-abi"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NEON-FP-ARMV8-DAG: "-target-feature" "+fp-armv8"
// CHECK-NEON-FP-ARMV8-DAG: "-target-feature" "+neon"
// CHECK-NEON-FP-ARMV8-DAG: "-target-feature" "-crypto"
ARM: add test for last commit This time I managed to forget the test. Patch still by Stephen Kelly. llvm-svn: 188797
2013-08-20 21:19:43 +08:00
[ARMv8] Add crypto-neon-fp-armv8 -mfpu option in driver. llvm-svn: 191001
2013-09-19 21:54:03 +08:00
// RUN: %clang -target armv8-linux-gnueabihf -mfpu=crypto-neon-fp-armv8 %s -### 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-CRYPTO-NEON-FP-ARMV8 %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// CHECK-CRYPTO-NEON-FP-ARMV8-NOT: "-target-feature" "+soft-float"
// CHECK-CRYPTO-NEON-FP-ARMV8-NOT: "-target-feature" "+soft-float-abi"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-CRYPTO-NEON-FP-ARMV8-DAG: "-target-feature" "+fp-armv8"
// CHECK-CRYPTO-NEON-FP-ARMV8-DAG: "-target-feature" "+crypto"
[ARMv8] Add crypto-neon-fp-armv8 -mfpu option in driver. llvm-svn: 191001
2013-09-19 21:54:03 +08:00
[ARM] Add -mfpu=none option to the driver. llvm-svn: 191736
2013-10-01 18:20:54 +08:00
// RUN: %clang -target armv8-linux-gnueabi -mfpu=none %s -### 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-NO-FP %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// CHECK-NO-FP-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NO-FP-DAG: "-target-feature" "+soft-float-abi"
// CHECK-NO-FP-DAG: "-target-feature" "-fpregs"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-NO-FP-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NO-FP-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-NO-FP-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-NO-FP-DAG: "-target-feature" "-fp-armv8d16sp"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-NO-FP-DAG: "-target-feature" "-fp64"
// CHECK-NO-FP-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-NO-FP-DAG: "-target-feature" "-neon"
// CHECK-NO-FP-DAG: "-target-feature" "-crypto"
[ARM] Add -mfpu=none option to the driver. llvm-svn: 191736
2013-10-01 18:20:54 +08:00
ARM: add test for last commit This time I managed to forget the test. Patch still by Stephen Kelly. llvm-svn: 188797
2013-08-20 21:19:43 +08:00
// RUN: %clang -target arm-linux-gnueabihf %s -### 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-HF %s
Add support for musl-libc on ARM Linux. Patch by Lei Zhang! llvm-svn: 273735
2016-06-25 05:35:06 +08:00
// RUN: %clang -target arm-linux-musleabihf %s -### 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-HF %s
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// CHECK-HF-NOT: "-target-feature" "+soft-float"
// CHECK-HF-NOT: "-target-feature" "+soft-float-abi"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-HF-DAG: "-target-cpu" "arm1176jzf-s"
Don't pass +soft-float, +soft-float-abi to the arm assembler. llvm-svn: 195551
2013-11-23 22:36:40 +08:00
// RUN: %clang -target armv7-apple-darwin -x assembler %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=ASM %s
// ASM-NOT: -target-feature
[ARM] disable FPU features when using soft floating point. To be compatible with GCC if soft floating point is in effect any FPU specified is effectively ignored, eg, -mfloat-abi=soft -fpu=neon If any floating point features which require FPU hardware are enabled they must be disable. There was some support for doing this for NEON, but it did not handle VFP, nor did it prevent the backend from emitting the build attribute Tag_FP_arch describing the generated code as using the floating point hardware if a FPU was specified (even though soft float does not use the FPU). Disabling the hardware floating point features for targets which are compiling for soft float has meant that some tests which were incorrectly checking for hardware support also needed to be updated. In such cases, where appropriate the tests have been updated to check compiling for soft float and a non-soft float variant (usually softfp). This was usually because the target specified in the test defaulted to soft float. Differential Revision: https://reviews.llvm.org/D42569 llvm-svn: 325492
2018-02-19 20:40:26 +08:00
// RUN: %clang -target armv8-linux-gnueabi -mfloat-abi=soft -mfpu=none %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv7-linux-gnueabi -mfloat-abi=soft -mfpu=none %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv6-linux-gnueabi -mfloat-abi=soft -mfpu=none %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv5-linux-gnueabi -mfloat-abi=soft -mfpu=none %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv4-linux-gnueabi -mfloat-abi=soft -mfpu=none %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv8-linux-gnueabi -msoft-float -mfpu=none %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv8-linux-gnueabi -mfloat-abi=soft %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
// RUN: %clang -target armv8-linux-gnueabi -msoft-float %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFT-ABI-FP %s
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "+soft-float"
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "+soft-float-abi"
[ARM] Update clang for removal of vfp2d16 and vfp2d16sp Matching fix for https://reviews.llvm.org/D67375 (r372186). Differential Revision: https://reviews.llvm.org/D67467 llvm-svn: 372187
2019-09-18 05:43:19 +08:00
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "-vfp2sp"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "-neon"
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "-crypto"
// CHECK-SOFT-ABI-FP-DAG: "-target-feature" "-fpregs"
[Driver] Add defaults for Android ARM FPUs. Summary: Android mandates that devices have at least vfpv3-d16 until Marshmallow and NEON after that. Still honor the user's decision, but raise the defaults for Android targets. Reviewers: srhines, pirama, javed.absar, kristof.beyls, peter.smith Reviewed By: peter.smith Subscribers: peter.smith, rengolin, kristof.beyls, chrib, cfe-commits Differential Revision: https://reviews.llvm.org/D53121 llvm-svn: 344367
2018-10-13 01:06:31 +08:00
// RUN: %clang -target arm-linux-androideabi21 %s -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-ARM5-ANDROID-FP-DEFAULT %s
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-ARM5-ANDROID-FP-DEFAULT-DAG: "-target-feature" "+soft-float"
// CHECK-ARM5-ANDROID-FP-DEFAULT-DAG: "-target-feature" "+soft-float-abi"
[ARM] Replace fp-only-sp and d16 with fp64 and d32. Those two subtarget features were awkward because their semantics are reversed: each one indicates the _lack_ of support for something in the architecture, rather than the presence. As a consequence, you don't get the behavior you want if you combine two sets of feature bits. Each SubtargetFeature for an FP architecture version now comes in four versions, one for each combination of those options. So you can still say (for example) '+vfp2' in a feature string and it will mean what it's always meant, but there's a new string '+vfp2d16sp' meaning the version without those extra options. A lot of this change is just mechanically replacing positive checks for the old features with negative checks for the new ones. But one more interesting change is that I've rearranged getFPUFeatures() so that the main FPU feature is appended to the output list *before* rather than after the features derived from the Restriction field, so that -fp64 and -d32 can override defaults added by the main feature. Reviewers: dmgreen, samparker, SjoerdMeijer Subscribers: srhines, javed.absar, eraman, kristof.beyls, hiraditya, zzheng, Petar.Avramovic, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D60691 llvm-svn: 361845
2019-05-29 00:13:20 +08:00
// CHECK-ARM5-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+d32"
[Driver] Add defaults for Android ARM FPUs. Summary: Android mandates that devices have at least vfpv3-d16 until Marshmallow and NEON after that. Still honor the user's decision, but raise the defaults for Android targets. Reviewers: srhines, pirama, javed.absar, kristof.beyls, peter.smith Reviewed By: peter.smith Subscribers: peter.smith, rengolin, kristof.beyls, chrib, cfe-commits Differential Revision: https://reviews.llvm.org/D53121 llvm-svn: 344367
2018-10-13 01:06:31 +08:00
// CHECK-ARM5-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+vfp3"
// CHECK-ARM5-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+vfp4"
// CHECK-ARM5-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+fp-armv8"
// CHECK-ARM5-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+neon"
// CHECK-ARM5-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+crypto"
// RUN: %clang -target armv7-linux-androideabi21 %s -### -c 2>&1 \
[Driver] Default all Android ARM targets to NEON. Summary: There are an insignificant number of ARM Android devices that don't support NEON. Default to using NEON since that will improve performance on the majority of devices. Users that need to target non-NEON devices can still explicitly disable NEON. Reviewers: srhines, pirama, kristof.beyls Reviewed By: pirama Subscribers: efriedma, javed.absar, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D58153 llvm-svn: 354166
2019-02-16 04:31:54 +08:00
// RUN: | FileCheck --check-prefix=CHECK-ARM7-ANDROID-FP-DEFAULT %s
// CHECK-ARM7-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-ARM7-ANDROID-FP-DEFAULT-DAG: "-target-feature" "+soft-float-abi"
// CHECK-ARM7-ANDROID-FP-DEFAULT-DAG: "-target-feature" "+vfp3"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-ARM7-ANDROID-FP-DEFAULT-DAG: "-target-feature" "-vfp4"
// CHECK-ARM7-ANDROID-FP-DEFAULT-DAG: "-target-feature" "-fp-armv8"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-ARM7-ANDROID-FP-DEFAULT-DAG: "-target-feature" "+neon"
[Driver] Default all Android ARM targets to NEON. Summary: There are an insignificant number of ARM Android devices that don't support NEON. Default to using NEON since that will improve performance on the majority of devices. Users that need to target non-NEON devices can still explicitly disable NEON. Reviewers: srhines, pirama, kristof.beyls Reviewed By: pirama Subscribers: efriedma, javed.absar, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D58153 llvm-svn: 354166
2019-02-16 04:31:54 +08:00
// CHECK-ARM7-ANDROID-FP-DEFAULT-NOT: "-target-feature" "+crypto"
// RUN: %clang -target armv7-linux-androideabi21 %s -mfpu=vfp3-d16 -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-ARM7-ANDROID-FP-D16 %s
// CHECK-ARM7-ANDROID-FP-D16-NOT: "-target-feature" "+soft-float"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-ARM7-ANDROID-FP-D16-DAG: "-target-feature" "+soft-float-abi"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-ARM7-ANDROID-FP-D16-DAG: "-target-feature" "-d32"
[ARM] Fix bugs introduced by the fp64/d32 rework. Change D60691 caused some knock-on failures that weren't caught by the existing tests. Firstly, selecting a CPU that should have had a restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs and no double precision) could give the unrestricted version, because `ARM::getFPUFeatures` returned a list of features including subtracted ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw away all the ones that didn't start with `+`. Secondly, the preprocessor macros didn't reliably match the actual compilation settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as if hardware FP was available, because the list of features on the cc1 command line would include things like `+vfp4`,`-vfp4d16` and clang didn't realise that one of those cancelled out the other. I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so that it returns a list that enables every FP-related feature compatible with the selected FPU and disables every feature not compatible, which is more verbose but means clang doesn't have to understand the dependency relationships between the backend features. Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all the various forms of the FP feature names, so that it won't miss cases where it should have set `HW_FP` to feed into feature test macros. That in turn caused an ordering problem when handling `-mcpu=foo+bar` together with `-mfpu=something_that_turns_off_bar`. To fix that, I've arranged that the `+bar` suffixes on the end of `-mcpu` and `-march` cause feature names to be put into a separate vector which is concatenated after the output of `getFPUFeatures`. Another side effect of all this is to fix a bug where `clang -target armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was because `HW_FP` was being set to a value including only the `FPARMV8` bit, but that feature test macro was testing only the `VFP4FPU` bit. Now `HW_FP` ends up with all the bits set, so it gives the right answer. Changes to tests included in this patch: * `arm-target-features.c`: I had to change basically all the expected results. (The Cortex-M4 test in there should function as a regression test for the accidental double-precision bug.) * `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG` everywhere so that those tests are no longer sensitive to the order of cc1 feature options on the command line. * `arm-acle-6.5.c`: been updated to expect the right answer to that FMA test. * `Preprocessor/arm-target-features.c`: added a regression test for the `mfpu=softvfp` issue. Reviewers: SjoerdMeijer, dmgreen, ostannard, samparker, JamesNagurne Reviewed By: ostannard Subscribers: srhines, javed.absar, kristof.beyls, hiraditya, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D62998 llvm-svn: 362791
2019-06-07 20:42:54 +08:00
// CHECK-ARM7-ANDROID-FP-D16-DAG: "-target-feature" "+vfp3d16"
[Driver] Default all Android ARM targets to NEON. Summary: There are an insignificant number of ARM Android devices that don't support NEON. Default to using NEON since that will improve performance on the majority of devices. Users that need to target non-NEON devices can still explicitly disable NEON. Reviewers: srhines, pirama, kristof.beyls Reviewed By: pirama Subscribers: efriedma, javed.absar, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D58153 llvm-svn: 354166
2019-02-16 04:31:54 +08:00
// CHECK-ARM7-ANDROID-FP-D16-NOT: "-target-feature" "+vfp4"
// CHECK-ARM7-ANDROID-FP-D16-NOT: "-target-feature" "+fp-armv8"
// CHECK-ARM7-ANDROID-FP-D16-NOT: "-target-feature" "+neon"
// CHECK-ARM7-ANDROID-FP-D16-NOT: "-target-feature" "+crypto"
[Driver,ARM] Make -mfloat-abi=soft turn off MVE. Since `-mfloat-abi=soft` is taken to mean turning off all uses of the FP registers, it should turn off the MVE vector instructions as well as NEON and scalar FP. But it wasn't doing so. So the options `-march=armv8.1-m.main+mve.fp+fp.dp -mfloat-abi=soft` would cause the underlying LLVM to //not// support MVE (because it knows the real target feature relationships and turned off MVE when the `fpregs` feature was removed), but the clang layer still thought it //was// supported, and would misleadingly define the feature macro `__ARM_FEATURE_MVE`. The ARM driver code already has a long list of feature names to turn off when `-mfloat-abi=soft` is selected. The fix is to add the missing entries `mve` and `mve.fp` to that list. Reviewers: dmgreen Subscribers: kristof.beyls, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D69025 llvm-svn: 375001
2019-10-16 21:23:39 +08:00
// RUN: %clang -target arm-none-none-eabi %s -march=armv8.1-m.main+mve.fp+fp.dp -mfloat-abi=soft -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-SOFTFLOATABI-INHIBITS-MVE %s
// CHECK-SOFTFLOATABI-INHIBITS-MVE-NOT: "-target-feature" "+mve"
// CHECK-SOFTFLOATABI-INHIBITS-MVE-DAG: "-target-feature" "-mve"
// CHECK-SOFTFLOATABI-INHIBITS-MVE-DAG: "-target-feature" "-mve.fp"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// RUN: %clang -target arm-none-none-eabi %s -march=armv8.1-m.main+mve.fp -mfpu=none -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-MVEFP-FPUNONE %s
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-vfp2sp"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-fp64"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-d32"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-neon"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-crypto"
[ARM][TargetParser] Improve handling of dependencies between target features The patch at https://reviews.llvm.org/D64048 added "negative" dependency handling in `ARM::appendArchExtFeatures`: feature "noX" removes all features, which imply "X". This patch adds the "positive" handling: feature "X" adds all the feature strings implied by "X". (This patch also comes from the suggestion here https://reviews.llvm.org/D72633#inline-658582) Differential Revision: https://reviews.llvm.org/D72762
2020-02-05 23:19:03 +08:00
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "+mve"
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "+dsp"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-MVEFP-FPUNONE-DAG: "-target-feature" "-mve.fp"
// CHECK-MVEFP-FPUNONE-NOT: "-target-feature" "-fpregs"
[ARM][MVE] Fix a corner case of checking for MVE-I with -mfpu=none -march=armv8.1-m.main+mve.fp+nomve -mfpu=none should disable FP registers and instructions moving to/from FP registers. This patch fixes the case when "+mve" (added to the feature list by "+mve.fp"), is followed by "-mve" (added by "+nomve"). Differential Revision: https://reviews.llvm.org/D72633
2020-02-11 20:01:17 +08:00
// RUN: %clang -target arm-none-none-eabi %s -march=armv8.1-m.main+mve.fp+nomve -mfpu=none -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-MVEFP-NOMVE-FPUNONE %s
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-vfp2sp"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-vfp3d16sp"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-vfp4d16sp"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-fp-armv8d16sp"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-fp64"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-d32"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-neon"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-crypto"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "+dsp"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-mve"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-mve.fp"
// CHECK-MVEFP-NOMVE-FPUNONE-DAG: "-target-feature" "-fpregs"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// RUN: %clang -target arm-none-none-eabi %s -march=armv8.1-m.main+mve -mfpu=none -### -c 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-MVEI-FPUNONE %s
// CHECK-MVEI-FPUNONE-DAG: "-target-feature" "-mve.fp"
// CHECK-MVEI-FPUNONE-DAG: "-target-feature" "+mve"
[ARM][TargetParser] Improve handling of dependencies between target features The patch at https://reviews.llvm.org/D64048 added "negative" dependency handling in `ARM::appendArchExtFeatures`: feature "noX" removes all features, which imply "X". This patch adds the "positive" handling: feature "X" adds all the feature strings implied by "X". (This patch also comes from the suggestion here https://reviews.llvm.org/D72633#inline-658582) Differential Revision: https://reviews.llvm.org/D72762
2020-02-05 23:19:03 +08:00
// CHECK-MVEI-FPUNONE-DAG: "-target-feature" "+dsp"
[ARM][MVE] MVE-I should not be disabled by -mfpu=none Architecturally, it's allowed to have MVE-I without an FPU, thus -mfpu=none should not disable MVE-I, or moves to/from FP-registers. This patch removes `+/-fpregs` from features unconditionally added to target feature list, depending on FPU and moves the logic to Clang driver, where the negative form (`-fpregs`) is conditionally added to the target features list for the cases of `-mfloat-abi=soft`, or `-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative form is added by the driver, the positive one is derived from other features in the backend. Differential Revision: https://reviews.llvm.org/D71843
2020-01-09 21:47:52 +08:00
// CHECK-MVEI-FPUNONE-NOT: "-target-feature" "-fpregs"