Use the best FPU strategies on 32-bits Arm Cortex
Subsystems (2)
cover Use the best FPU strategies on 32-bits Arm Cortex
Hi All,
this patchset aims to enable the best FPU strategy for every board with 32-bits Arm Cortex actually present in Buildroot. I don't own these boards so I can't test these changes. What is about Allwinner doesn't worry me because I've tested a lot of cases with Olimex boards, but the other changes are still to be tested.
So I ask to the board maintainers to test the patches that involve their boards if possible. Anyway I've checked well all the SoCs Datasheet and I think I've made it correctly, even if in some of them it's not specified if VFPv4 means -D32. I assume it like that because otherwise -D16 is usually specified.
Best regards
---
Giulio Benetti
Benetti Engineering sas
Giulio Benetti (28):
configs/acmesystems_acqua_a5_256mb_defconfig: enable NEON/VFPV4 FPU
strategy
configs/acmesystems_acqua_a5_512mb_defconfig: enable NEON/VFPV4 FPU
strategy
configs/bananapi_m2_plus_defconfig: enable NEON/VFPV4 FPU strategy configs/bananapi_m2_plus_defconfig: enable NEON/VFPV4 FPU strategy configs/beagleboardx15_defconfig: enable NEON/VFPV4 FPU strategy configs/beagleboneai_defconfig: enable NEON/VFPV4 FPU strategy configs/chromebook_snow_defconfig: enable NEON/VFPV4 FPU strategy configs/chromebook_snow_defconfig: enable NEON/VFPV4 FPU strategy configs/friendlyarm_nanopi_m1_defconfig: enable NEON/VFPV4 FPU
strategy
configs/friendlyarm_nanopi_m1_plus_defconfig: enable NEON/VFPV4 FPU
strategy
configs/friendlyarm_nanopi_r1_defconfig: enable NEON/VFPV4 FPU
strategy
configs/friendlyarm_nanopi_r1_defconfig: enable NEON/VFPV4 FPU
strategy
configs/grinn_liteboard_defconfig: enable NEON/VFPV4 FPU strategy configs/licheepi_zero_defconfig: enable NEON/VFPV4 FPU strategy configs/microchip_sama5d27_wlsom1_ek_mmc_defconfig: enable NEON/VFPV4
FPU strategy
configs/microchip_sama5d27_wlsom1_ek_mmc_dev_defconfig: enable
NEON/VFPV4 FPU strategy
configs/microchip_sama5d2_icp_mmc_dev_defconfig: enable NEON/VFPV4 FPU
strategy
configs/microchip_sama5d2_icp_mmc_dev_defconfig: enable NEON/VFPV4 FPU
strategy
configs/orangepi_lite_defconfig: enable NEON/VFPV4 FPU strategy configs/orangepi_one_defconfig: enable NEON/VFPV4 FPU strategy configs/orangepi_pc_defconfig: enable NEON/VFPV4 FPU strategy configs/orangepi_pc_plus_defconfig: enable NEON/VFPV4 FPU strategy configs/orangepi_plus_defconfig: enable NEON/VFPV4 FPU strategy configs/orangepi_r1_defconfig: enable NEON/VFPV4 FPU strategy configs/orangepi_zero_defconfig: enable NEON/VFPV4 FPU strategy configs/stm32mp157a_dk1_defconfig: enable NEON/VFPV4 FPU strategy configs/stm32mp157c_dk2_defconfig: enable NEON/VFPV4 FPU strategy configs/stm32mp157c_odyssey_defconfig: enable NEON/VFPV4 FPU strategy
configs/acmesystems_acqua_a5_256mb_defconfig | 1 + configs/acmesystems_acqua_a5_512mb_defconfig | 1 + configs/bananapi_m2_plus_defconfig | 1 + configs/bananapi_m2_ultra_defconfig | 1 + configs/beagleboardx15_defconfig | 1 + configs/beagleboneai_defconfig | 1 + configs/chromebook_snow_defconfig | 1 + configs/freescale_imx7dsabresd_defconfig | 1 + configs/friendlyarm_nanopi_m1_defconfig | 2 +- configs/friendlyarm_nanopi_m1_plus_defconfig | 2 +- configs/friendlyarm_nanopi_neo_defconfig | 2 +- configs/friendlyarm_nanopi_r1_defconfig | 2 +- configs/grinn_liteboard_defconfig | 1 + configs/licheepi_zero_defconfig | 2 +- configs/microchip_sama5d27_wlsom1_ek_mmc_defconfig | 1 + configs/microchip_sama5d27_wlsom1_ek_mmc_dev_defconfig | 1 + configs/microchip_sama5d2_icp_mmc_defconfig | 1 + configs/microchip_sama5d2_icp_mmc_dev_defconfig | 1 + configs/orangepi_lite_defconfig | 2 +- configs/orangepi_one_defconfig | 2 +- configs/orangepi_pc_defconfig | 2 +- configs/orangepi_pc_plus_defconfig | 2 +- configs/orangepi_plus_defconfig | 2 +- configs/orangepi_r1_defconfig | 2 +- configs/orangepi_zero_defconfig | 2 +- configs/stm32mp157a_dk1_defconfig | 1 + configs/stm32mp157c_dk2_defconfig | 1 + configs/stm32mp157c_odyssey_defconfig | 1 + 28 files changed, 28 insertions(+), 12 deletions(-)
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 1/28 configs/acmesystems_acqua_a5_256mb_defconfig: enable NEON/VFPV4 FPU strategy
As pointed by SAMA5D2 Datasheet[1]:
``` The Cortex-A5 NEON Media Processing Engine (MPE) extends the Cortex-A5 functionality to provide support for the ARM v7 Advanced SIMD v2 and Vector Floating-Point v4 (VFPv4) instruction sets. ```
So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
--- configs/acmesystems_acqua_a5_256mb_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/acmesystems_acqua_a5_256mb_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a5=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_ENABLE_VFP=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_4=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 01/28] configs/acmesystems_acqua_a5_256mb_defconfig: enable NEON/VFPV4 FPU strategy
Hi All,
Quoted 1 line · Giulio Benetti
As pointed by SAMA5D2 Datasheet[1]:
As pointed by Edgar on Patch 02/28 Acqua A5 doesn't have a SAMA5D2 but a SAMA5D3 that doesn't have VFPV4+NEON, but a VFPv4-D16 and NEON, so this can't work.
I've marked it as Rejected in Patchwork as patch 01/28 and please drop it.
Thank you and sorry for the noise.
Best regards
Signature · 2 lines
Giulio Benetti Benetti Engineering sas
Quoted 21 lines
``` The Cortex-A5 NEON Media Processing Engine (MPE) extends the Cortex-A5 functionality to provide support for the ARM v7 Advanced SIMD v2 and Vector Floating-Point v4 (VFPv4) instruction sets. ```So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
Signed-off-by Giulio Benetti <[email protected]>--- configs/acmesystems_acqua_a5_256mb_defconfig | 1 + 1 file changed, 1 insertion(+)configs/acmesystems_acqua_a5_256mb_defconfig +0
-0
configs/acmesystems_acqua_a5_256mb_defconfig line 6
Quoted 6 lines
configs/acmesystems_acqua_a5_256mb_defconfig +1
-0@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a5=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_ENABLE_VFP=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_4=y
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 2/28 configs/acmesystems_acqua_a5_512mb_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiAs pointed by SAMA5D2 Datasheet[1]:
``` The Cortex-A5 NEON Media Processing Engine (MPE) extends the Cortex-A5 functionality to provide support for the ARM v7 Advanced SIMD v2 and Vector Floating-Point v4 (VFPv4) instruction sets. ```
So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
--- configs/acmesystems_acqua_a5_512mb_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/acmesystems_acqua_a5_512mb_defconfig +1 -0
@@ -1,6 +1,7 @@ BR2_arm=y BR2_cortex_a5=y BR2_ARM_ENABLE_VFP=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_4=y BR2_SYSTEM_DHCP="eth0"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 02/28] configs/acmesystems_acqua_a5_512mb_defconfig: enable NEON/VFPV4 FPU strategy
Hi!
[Replying to list only, hoping it is the right etiquette.]
configs/acmesystems_acqua_a5_512mb_defconfig line 7
Quoted 10 lines · Giulio Benetti
configs/acmesystems_acqua_a5_512mb_defconfig +1
-0@@ -1,6 +1,7 @@ BR2_arm=y BR2_cortex_a5=y BR2_ARM_ENABLE_VFP=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_4=y BR2_SYSTEM_DHCP="eth0"
I think this may be missing some dependencies, as this change does not survive `save defconfig'. I tried the following from a clean tree:
git pull # from master
git am ~/tmp/saved-patch.email # the patch above make acmesystems_acqua_a5_512mb_defconfig make nconfig
# no changes to configuration
<F6> Save
<F9> Exit
make savedefconfig
git diff
And git showed me that the patch had just been reverted on the worktree. According to arch/Config.in.arm, BR2_ARM_FPU_NEON_VFPV4 depends on BR2_ARM_CPU_HAS_NEON, which is not mentioned in my .config. It also appears that BR2_ARM_ENABLE_NEON selects BR2_ARM_CPU_HAS_NEON. Would BR2_ARM_ENABLE_NEON be the canonical way of making BR2_ARM_FPU_NEON_VFPV4 acceptable? I am a bit confused by the VFP selection logic in arch/Config.in.arm, so I am not sure that would be the right approach.
As a follow up question, if this issue is fixed, what would be the simplest way of testing the floating point implementation on the board? I was thinking about adding gnuplot, which doesn't have any dependencies, and having it do some computations. I do not know, however, whether this would exercise the NEON SIMD extensions.
Best regards,
Edgar.
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 02/28] configs/acmesystems_acqua_a5_512mb_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiHi again!
Quoted 2 lines · Giulio Benetti
As pointed by SAMA5D2 Datasheet [...] So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
BTW, the Acqua is a SAMA5D3[1] (SAMA5D31 or SAMA5D36, depending on the version), not a SAMA5D2. The datasheet[2] of the SAMA5D3 states:
The Floating-Point Unit (FPU) supports the ARMv7 VFPv4-D16 architecture without Advanced SIMD extensions (NEON).
So I guess the patch is not relevant to this board.
[1] https://www.acmesystems.it/acqua
[2] https://ww1.microchip.com/downloads/en/DeviceDoc/SAMA5D3-Series-Data-sheet-DS60001609b.pdf
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 02/28] configs/acmesystems_acqua_a5_512mb_defconfig: enable NEON/VFPV4 FPU strategy
Hi Edgar,
Quoted 11 lines · Edgar Bonet
Hi again!
Quoted 2 lines · Giulio Benetti
As pointed by SAMA5D2 Datasheet [...] So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
BTW, the Acqua is a SAMA5D3[1] (SAMA5D31 or SAMA5D36, depending on the version), not a SAMA5D2. The datasheet[2] of the SAMA5D3 states:
The Floating-Point Unit (FPU) supports the ARMv7 VFPv4-D16 architecture without Advanced SIMD extensions (NEON).
Ah, thank you. I had the SAMA5D3 opened and I couldn't find anymore which board used it. I thought it was this one. Thanks for pointing.
Quoted 1 line
So I guess the patch is not relevant to this board.
Exactly. So please drop this patch and I've set it as rejected in Patchwork.
I'll go in depth later with patch 01/28.
Thank you for the answers!
Best regards
Signature · 6 lines
Giulio Benetti Benetti Engineering sas _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 3/28 configs/bananapi_m2_plus_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis boards has Allwinner H3 that supports NEON/VFPV4 FPU strategy, so let's enable it.
--- configs/bananapi_m2_plus_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/bananapi_m2_plus_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a7=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_0=y BR2_TARGET_GENERIC_ISSUE="Welcome to Bananapi M2 Plus" BR2_ROOTFS_POST_IMAGE_SCRIPT="support/scripts/genimage.sh"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 4/28 configs/bananapi_m2_ultra_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis boards has Allwinner A40i/R40/V40 that support NEON/VFPV4 FPU strategy, so let's enable it.
--- configs/bananapi_m2_ultra_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/bananapi_m2_ultra_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a7=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_4_18=y BR2_TARGET_GENERIC_ISSUE="Welcome to Bananapi M2 Ultra" BR2_ROOTFS_POST_IMAGE_SCRIPT="support/scripts/genimage.sh"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 5/28 configs/beagleboardx15_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has TI AM5728 that is a cortex-A15 and supports NEON/VFPV4 FPU strategy, so let's enable it.
--- configs/beagleboardx15_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/beagleboardx15_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a15=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_4_9=y BR2_ROOTFS_POST_IMAGE_SCRIPT="support/scripts/genimage.sh" BR2_ROOTFS_POST_SCRIPT_ARGS="-c board/beagleboardx15/genimage.cfg"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 6/28 configs/beagleboneai_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has TI AM5729 that is a cortex-A15 and supports NEON/VFPV4 FPU strategy, so let's enable it.
--- configs/beagleboneai_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/beagleboneai_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a15=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_GLOBAL_PATCH_DIR="board/beagleboneai/patches" BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_SYSTEM_DHCP="eth0"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 7/28 configs/chromebook_snow_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has NVIDIA Tegra K1 that is a cortex-A15 and supports NEON/VFPV4 FPU strategy, so let's enable it.
--- configs/chromebook_snow_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/chromebook_snow_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a15=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_4_15=y BR2_TARGET_GENERIC_GETTY_PORT="tty1" BR2_TARGET_GENERIC_GETTY_TERM="linux"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 8/28 configs/freescale_imx7dsabresd_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has i.M7D that supports NEON/VFPV4 FPU strategy, so let's enable it.
--- configs/freescale_imx7dsabresd_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/freescale_imx7dsabresd_defconfig +1 -0
@@ -1,6 +1,7 @@ # architecture BR2_arm=y BR2_cortex_a7=y +BR2_ARM_FPU_NEON_VFPV4=y # Linux headers same as kernel, a 5.10 series BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 9/28 configs/friendlyarm_nanopi_m1_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/friendlyarm_nanopi_m1_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/friendlyarm_nanopi_m1_defconfig +1 -1
@@ -1,7 +1,7 @@ # Architecture BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y # Linux headers same as kernel, a 5.0 series BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_0=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 10/28 configs/friendlyarm_nanopi_m1_plus_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/friendlyarm_nanopi_m1_plus_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/friendlyarm_nanopi_m1_plus_defconfig +1 -1
@@ -1,7 +1,7 @@ # Architecture BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y # Linux headers same as kernel, a 5.0 series BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_0=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 11/28 configs/friendlyarm_nanopi_neo_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/friendlyarm_nanopi_neo_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/friendlyarm_nanopi_neo_defconfig +1 -1
@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_TARGET_GENERIC_HOSTNAME="nanopi-neo" BR2_TARGET_GENERIC_ISSUE="Welcome to Buildroot for the NanoPi NEO"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 12/28 configs/friendlyarm_nanopi_r1_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/friendlyarm_nanopi_r1_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/friendlyarm_nanopi_r1_defconfig +1 -1
@@ -1,7 +1,7 @@ # Architecture BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y # Linux headers same as kernel, a 5.0 series BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_0=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 13/28 configs/grinn_liteboard_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has i.MX6UL that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it.
--- configs/grinn_liteboard_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/grinn_liteboard_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a7=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_14=y BR2_SYSTEM_DHCP="eth0" BR2_ROOTFS_POST_IMAGE_SCRIPT="support/scripts/genimage.sh"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 14/28 configs/licheepi_zero_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner V3s that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/licheepi_zero_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/licheepi_zero_defconfig +1 -1
@@ -1,7 +1,7 @@ # Target options BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y # Toolchain BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_3=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 15/28 configs/microchip_sama5d27_wlsom1_ek_mmc_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiAs pointed by SAMA5D2 Datasheet[1]:
``` The Cortex-A5 NEON Media Processing Engine (MPE) extends the Cortex-A5 functionality to provide support for the ARM v7 Advanced SIMD v2 and Vector Floating-Point v4 (VFPv4) instruction sets. ```
So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
--- configs/microchip_sama5d27_wlsom1_ek_mmc_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/microchip_sama5d27_wlsom1_ek_mmc_defconfig +1 -0
@@ -2,6 +2,7 @@ BR2_arm=y BR2_cortex_a5=y BR2_ARM_ENABLE_NEON=y BR2_ARM_ENABLE_VFP=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_ROOTFS_POST_IMAGE_SCRIPT="support/scripts/genimage.sh"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 16/28 configs/microchip_sama5d27_wlsom1_ek_mmc_dev_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiAs pointed by SAMA5D2 Datasheet[1]:
``` The Cortex-A5 NEON Media Processing Engine (MPE) extends the Cortex-A5 functionality to provide support for the ARM v7 Advanced SIMD v2 and Vector Floating-Point v4 (VFPv4) instruction sets. ```
So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
--- configs/microchip_sama5d27_wlsom1_ek_mmc_dev_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/microchip_sama5d27_wlsom1_ek_mmc_dev_defconfig +1 -0
@@ -2,6 +2,7 @@ BR2_arm=y BR2_cortex_a5=y BR2_ARM_ENABLE_NEON=y BR2_ARM_ENABLE_VFP=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_TOOLCHAIN_BUILDROOT_WCHAR=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 17/28 configs/microchip_sama5d2_icp_mmc_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiAs pointed by SAMA5D2 Datasheet[1]:
``` The Cortex-A5 NEON Media Processing Engine (MPE) extends the Cortex-A5 functionality to provide support for the ARM v7 Advanced SIMD v2 and Vector Floating-Point v4 (VFPv4) instruction sets. ```
So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
--- configs/microchip_sama5d2_icp_mmc_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/microchip_sama5d2_icp_mmc_defconfig +1 -0
@@ -2,6 +2,7 @@ BR2_arm=y BR2_cortex_a5=y BR2_ARM_ENABLE_NEON=y BR2_ARM_ENABLE_VFP=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_ROOTFS_POST_IMAGE_SCRIPT="support/scripts/genimage.sh"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 18/28 configs/microchip_sama5d2_icp_mmc_dev_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiAs pointed by SAMA5D2 Datasheet[1]:
``` The Cortex-A5 NEON Media Processing Engine (MPE) extends the Cortex-A5 functionality to provide support for the ARM v7 Advanced SIMD v2 and Vector Floating-Point v4 (VFPv4) instruction sets. ```
So let's enable VFPV4/NEON FPU strategy instead of the default VFPV4-D16.
--- configs/microchip_sama5d2_icp_mmc_dev_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/microchip_sama5d2_icp_mmc_dev_defconfig +1 -0
@@ -2,6 +2,7 @@ BR2_arm=y BR2_cortex_a5=y BR2_ARM_ENABLE_NEON=y BR2_ARM_ENABLE_VFP=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ARM_INSTRUCTIONS_THUMB2=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_TOOLCHAIN_BUILDROOT_WCHAR=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 19/28 configs/orangepi_lite_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/orangepi_lite_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/orangepi_lite_defconfig +1 -1
@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_15=y BR2_TARGET_GENERIC_HOSTNAME="OrangePi_Lite" BR2_TARGET_GENERIC_ISSUE="Welcome to Buildroot for the Orange Pi Lite"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 20/28 configs/orangepi_one_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/orangepi_one_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/orangepi_one_defconfig +1 -1
@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_12=y BR2_TARGET_GENERIC_HOSTNAME="OrangePi_One" BR2_TARGET_GENERIC_ISSUE="Welcome to Buildroot for the Orange Pi One"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 21/28 configs/orangepi_pc_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/orangepi_pc_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/orangepi_pc_defconfig +1 -1
@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_TARGET_GENERIC_HOSTNAME="OrangePi_PC" BR2_TARGET_GENERIC_ISSUE="Welcome to Buildroot for the Orange Pi PC"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 21/28] configs/orangepi_pc_defconfig: enable NEON/VFPV4 FPU strategy
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 22/28 configs/orangepi_pc_plus_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/orangepi_pc_plus_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/orangepi_pc_plus_defconfig +1 -1
@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_ROOTFS_DEVICE_CREATION_DYNAMIC_MDEV=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_12=y BR2_TARGET_GENERIC_HOSTNAME="OrangePi_PC_Plus"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 23/28 configs/orangepi_plus_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H3 that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/orangepi_plus_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/orangepi_plus_defconfig +1 -1
@@ -1,7 +1,7 @@ # Architecture BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y # Linux headers same as kernel, a 5.0 series BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_0=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 24/28 configs/orangepi_r1_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H2+(compatible with H3) that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/orangepi_r1_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/orangepi_r1_defconfig +1 -1
@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_TARGET_GENERIC_HOSTNAME="OrangePi_R1" BR2_TARGET_GENERIC_ISSUE="Welcome to Buildroot for the Orange Pi R1"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 25/28 configs/orangepi_zero_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has Allwinner H2+(compatible with H3) that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
--- configs/orangepi_zero_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
configs/orangepi_zero_defconfig +1 -1
@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_GLOBAL_PATCH_DIR="board/orangepi/orangepi-zero/patches" BR2_ROOTFS_DEVICE_CREATION_DYNAMIC_MDEV=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_12=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 25/28] configs/orangepi_zero_defconfig: enable NEON/VFPV4 FPU strategy
18.01.2022 13:43, Giulio Benetti:
Quoted 13 lines
This board has Allwinner H2+(compatible with H3) that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
Signed-off-by Giulio Benetti <[email protected]>--- configs/orangepi_zero_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)configs/orangepi_zero_defconfig +0
-0
configs/orangepi_zero_defconfig line 6
Quoted 7 lines
configs/orangepi_zero_defconfig +1
-1@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_GLOBAL_PATCH_DIR="board/orangepi/orangepi-zero/patches" BR2_ROOTFS_DEVICE_CREATION_DYNAMIC_MDEV=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_12=y
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 25/28] configs/orangepi_zero_defconfig: enable NEON/VFPV4 FPU strategy
Hi Sergey,
thank you for the Tested-by, but you should put it…
Quoted 6 lines
Il giorno 23 gen 2022, alle ore 19:08, Sergey Kuzminov <[email protected]> ha scritto:
18.01.2022 13:43, Giulio Benetti:
Quoted 3 lines
This board has Allwinner H2+(compatible with H3) that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
Signed-off-by Giulio Benetti <[email protected]>
…here, after my SoB.
And when answering to Mailing List(ML), always use “Reply to All”.
Honestly it’s since years I always use “Reply to All” because it works both for multiple recipients and for single recipient too.
This way I don’t forget to do it.
Best regards!
Giulio
Quoted 17 lines
Quoted 8 lines
--- configs/orangepi_zero_defconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)configs/orangepi_zero_defconfig +0
-0configs/orangepi_zero_defconfig line 6
Quoted 7 lines
configs/orangepi_zero_defconfig +1
-1@@ -1,6 +1,6 @@ BR2_arm=y BR2_cortex_a7=y -BR2_ARM_FPU_VFPV4=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_GLOBAL_PATCH_DIR="board/orangepi/orangepi-zero/patches" BR2_ROOTFS_DEVICE_CREATION_DYNAMIC_MDEV=y BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_12=yTested-by Sergey Kuzminov <[email protected]>
Quoted 4 lines
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH 25/28] configs/orangepi_zero_defconfig: enable NEON/VFPV4 FPU strategy
Giulio, All,
On 2022-01-23 20:12 +0100, Giulio Benetti spake thusly:
Quoted 4 lines
Hi Sergey,
thank you for the Tested-by, but you should put it…
Quoted 8 lines
Quoted 6 lines
Il giorno 23 gen 2022, alle ore 19:08, Sergey Kuzminov <[email protected]> ha scritto:
18.01.2022 13:43, Giulio Benetti:
Quoted 3 lines
This board has Allwinner H2+(compatible with H3) that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of only VFPV4.
Signed-off-by Giulio Benetti <[email protected]>…here, after my SoB.
No, not needed: Patchwork will correctly catch it wherever it is as long as the mail is a proper reply, which it is here, and you can check it on
patchwork: the 'T' column is now set to '1' for this patch, and the tested-by tag is properly added to the mbox (which is what we maintainers use when applying patches):
$ curl 'https://patchwork.ozlabs.org/project/buildroot/patch/[email protected]/mbox/' [...] Signed-off-by: Giulio Benetti <[email protected]> Tested-by: Sergey Kuzminov <[email protected]> [...]
Regards,
Yann E. MORIN.
Signature · 10 lines
.-----------------.--------------------.------------------.--------------------. | Yann E. MORIN | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: | | +33 662 376 056 | Software Designer | \ / CAMPAIGN | ___ | | +33 561 099 427 `------------.-------: X AGAINST | \e/ There is no | | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL | v conspiracy. | '------------------------------^-------^------------------^--------------------' _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 26/28 configs/stm32mp157a_dk1_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has stm32mp157a that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of the default VFPV4-D16.
--- configs/stm32mp157a_dk1_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/stm32mp157a_dk1_defconfig +1 -0
@@ -1,6 +1,7 @@ # Architecture BR2_arm=y BR2_cortex_a7=y +BR2_ARM_FPU_NEON_VFPV4=y # Linux headers same as kernel, a 5.13 series BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_13=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 27/28 configs/stm32mp157c_dk2_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has stm32mp157c that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of the default VFPV4-D16.
--- configs/stm32mp157c_dk2_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/stm32mp157c_dk2_defconfig +1 -0
@@ -1,6 +1,7 @@ # Architecture BR2_arm=y BR2_cortex_a7=y +BR2_ARM_FPU_NEON_VFPV4=y # Linux headers same as kernel, a 5.13 series BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_13=y
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 28/28 configs/stm32mp157c_odyssey_defconfig: enable NEON/VFPV4 FPU strategy
In reply to Giulio BenettiThis board has stm32mp157c that is a cortex-A7 and supports NEON/VFPV4 FPU strategy, so let's enable it instead of the default VFPV4-D16.
--- configs/stm32mp157c_odyssey_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/stm32mp157c_odyssey_defconfig +1 -0
@@ -1,5 +1,6 @@ BR2_arm=y BR2_cortex_a7=y +BR2_ARM_FPU_NEON_VFPV4=y BR2_GLOBAL_PATCH_DIR="board/seeed/stm32mp157c-odyssey/patches" BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_10=y BR2_ROOTFS_OVERLAY="board/seeed/stm32mp157c-odyssey/overlay/"
Signature · 6 lines
2.25.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hello,
Quoted 11 lines · On Tue, 18 Jan 2022 11:43:10 +0100 Giulio Benetti…
this patchset aims to enable the best FPU strategy for every board with 32-bits Arm Cortex actually present in Buildroot. I don't own these boards so I can't test these changes. What is about Allwinner doesn't worry me because I've tested a lot of cases with Olimex boards, but the other changes are still to be tested.
So I ask to the board maintainers to test the patches that involve their boards if possible. Anyway I've checked well all the SoCs Datasheet and I think I've made it correctly, even if in some of them it's not specified if VFPv4 means -D32. I assume it like that because otherwise -D16 is usually specified.
Do we have a good understand of what -mfpu=neon-vfpv4 does? Your patch series basically converts many defconfigs to use this -mfpu value, but it's not clear to me how it works. What does it mean to combine NEON and VFPv4 instructions?
Regarding VFPv4 D16 vs. D32,
https://developer.arm.com/documentation/den0018/a/Compiling-NEON-Instructions/GCC-command-line-options/Option-to-specify-the-FPU
tells us:
VFPv3 and VFPv4 implementations provide 32 double-precision registers. However, when NEON unit is not present, the top sixteen registers (D16-D31) become optional. This is shown by the -d16 in the option name, which means that the top sixteen D registers are not available.
So, my understanding is that when NEON is available, the VFPv4 is guaranteed to have the 32 double-precision registers (D32).
Thomas
Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Thomas, Giulio, All,
On 2022-01-22 15:32 +0100, Thomas Petazzoni spake thusly:
Quoted 15 lines
Quoted 11 lines · On Tue, 18 Jan 2022 11:43:10 +0100 Giulio Benetti…
this patchset aims to enable the best FPU strategy for every board with 32-bits Arm Cortex actually present in Buildroot. I don't own these boards so I can't test these changes. What is about Allwinner doesn't worry me because I've tested a lot of cases with Olimex boards, but the other changes are still to be tested.
So I ask to the board maintainers to test the patches that involve their boards if possible. Anyway I've checked well all the SoCs Datasheet and I think I've made it correctly, even if in some of them it's not specified if VFPv4 means -D32. I assume it like that because otherwise -D16 is usually specified.
Quoted 4 lines
Do we have a good understand of what -mfpu=neon-vfpv4 does? Your patch series basically converts many defconfigs to use this -mfpu value, but it's not clear to me how it works. What does it mean to combine NEON and VFPv4 instructions?
The gcc man page states that specifying Neon as part of the fpu setting has no effect, unless the -funsafe-math-optimizations is also specified, because Neon is not compliant zith IEEE 754:
If the selected floating-point hardware includes the NEON extension (e.g. -mfpu=neon), note that floating-point operations are not generated by GCC's auto-vectorization pass unless -funsafe-math-optimizations is also specified. This is because NEON hardware does not fully implement the IEEE 754 standard for floating-point arithmetic (in particular denormal values are treated as zero), so the use of NEON instructions may lead to a loss of precision.
So it is my understanding that using Neon is not a good idea overall. It should only be requested on-demand by people who know what they are doing, and most probably, be setting appropriate CFLAGS on a per-package basis.
Additionally, changing the default FPU setting on those defconfigs is not really interesting. Indeed, the base system that (most of) those defconfig build are probably not exercising the FPU setting much (the busybox login and shell are probably not using much FPU insns).
So, for me, this series is a no-go, first because it has not been tsted on actual hardware, second because some of the changes introduce a dubious feature which is actually a no-op at best, or worse will generate incorrect code.
Regards,
Yann E. MORIN.
Quoted 18 lines
Regarding VFPv4 D16 vs. D32,
https://developer.arm.com/documentation/den0018/a/Compiling-NEON-Instructions/GCC-command-line-options/Option-to-specify-the-FPU
tells us:VFPv3 and VFPv4 implementations provide 32 double-precision registers. However, when NEON unit is not present, the top sixteen registers (D16-D31) become optional. This is shown by the -d16 in the option name, which means that the top sixteen D registers are not available.
So, my understanding is that when NEON is available, the VFPv4 is guaranteed to have the 32 double-precision registers (D32).
Thomas
Signature · 3 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com
Signature · 10 lines
.-----------------.--------------------.------------------.--------------------. | Yann E. MORIN | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: | | +33 662 376 056 | Software Designer | \ / CAMPAIGN | ___ | | +33 561 099 427 `------------.-------: X AGAINST | \e/ There is no | | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL | v conspiracy. | '------------------------------^-------^------------------^--------------------' _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hi Yann, Thomas, All,
Quoted 46 lines · Yann E. MORIN
Thomas, Giulio, All,
On 2022-01-22 15:32 +0100, Thomas Petazzoni spake thusly:
Quoted 15 lines
Quoted 11 lines · On Tue, 18 Jan 2022 11:43:10 +0100 Giulio Benetti…
this patchset aims to enable the best FPU strategy for every board with 32-bits Arm Cortex actually present in Buildroot. I don't own these boards so I can't test these changes. What is about Allwinner doesn't worry me because I've tested a lot of cases with Olimex boards, but the other changes are still to be tested.
So I ask to the board maintainers to test the patches that involve their boards if possible. Anyway I've checked well all the SoCs Datasheet and I think I've made it correctly, even if in some of them it's not specified if VFPv4 means -D32. I assume it like that because otherwise -D16 is usually specified.
Quoted 4 lines
Do we have a good understand of what -mfpu=neon-vfpv4 does? Your patch series basically converts many defconfigs to use this -mfpu value, but it's not clear to me how it works. What does it mean to combine NEON and VFPv4 instructions?
The gcc man page states that specifying Neon as part of the fpu setting has no effect, unless the -funsafe-math-optimizations is also specified, because Neon is not compliant zith IEEE 754:
If the selected floating-point hardware includes the NEON extension (e.g. -mfpu=neon), note that floating-point operations are not generated by GCC's auto-vectorization pass unless -funsafe-math-optimizations is also specified. This is because NEON hardware does not fully implement the IEEE 754 standard for floating-point arithmetic (in particular denormal values are treated as zero), so the use of NEON instructions may lead to a loss of precision.
So it is my understanding that using Neon is not a good idea overall. It should only be requested on-demand by people who know what they are doing, and most probably, be setting appropriate CFLAGS on a per-package basis.
Additionally, changing the default FPU setting on those defconfigs is not really interesting. Indeed, the base system that (most of) those defconfig build are probably not exercising the FPU setting much (the busybox login and shell are probably not using much FPU insns).
Quoted 4 lines
So, for me, this series is a no-go, first because it has not been tsted on actual hardware, second because some of the changes introduce a dubious feature which is actually a no-op at best, or worse will generate incorrect code.
I agree with you, it's too dangerous. I mean, I've tested on all Allwinner Axx series, but haven't check that in the end the assembly code corresponding. What I plan to do later, so now you can reject this patchset, is to use the smallest package that uses fpu instructions for sure, then enable for example:
-mfpu=vfpv4
check
AND
-mfpu=neon-vfpv4
This way I will have my real answer before eventually re-submitting while explaining this in detail.
Also, once I will have cleared this topic and it comes out that fpu is not use with neon-vfpv4, then I'll send patches to remove that fpu strategy on the boards I've set it to it.
I thought it was easier, instead this is an insidious topic.
So yes, let's drop this and thank you all for helping me to clarify.
Kind regards!
Signature · 2 lines
Giulio Benetti Benetti Engineering sas
Quoted 22 lines
Regards,
Yann E. MORIN.Quoted 18 lines
Regarding VFPv4 D16 vs. D32,
https://developer.arm.com/documentation/den0018/a/Compiling-NEON-Instructions/GCC-command-line-options/Option-to-specify-the-FPU
tells us:VFPv3 and VFPv4 implementations provide 32 double-precision registers. However, when NEON unit is not present, the top sixteen registers (D16-D31) become optional. This is shown by the -d16 in the option name, which means that the top sixteen D registers are not available.
So, my understanding is that when NEON is available, the VFPv4 is guaranteed to have the 32 double-precision registers (D32).
Thomas
Signature · 3 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot