| Message ID | 20190521071814.1192-1-tesheng@andestech.com |
|---|---|
| State | Superseded |
| Delegated to: | Andes |
| Headers | show |
| Series | [U-Boot] riscv: add Kconfig entries for the F and D ISA extensions support | expand |
Hi Bin, > -----Original Message----- > From: Bin Meng [mailto:bmeng.cn@gmail.com] > Sent: Tuesday, May 21, 2019 3:56 PM > To: Eric Te-Sheng Lin(林德晟) > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(陳建志); > Greentime Ying-Han Hu(胡英漢); dslin1010@gmail.com > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > support > > Hi Eric, > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > This patch add Kconfig entries for the F (Single-Precision) > > adds > OK I'll correct it as adds > > and D (Double-Precision) floating point instruction-set extensions. > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > extension? > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > Signed-off-by: Eric Lin <tesheng@andestech.com> > > --- > > arch/riscv/Kconfig | 8 ++++++++ > > arch/riscv/Makefile | 12 ++++++++---- > > 2 files changed, 16 insertions(+), 4 deletions(-) > > > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index > > 362f3cdc65..a8031fa230 100644 > > --- a/arch/riscv/Kconfig > > +++ b/arch/riscv/Kconfig > > @@ -91,6 +91,14 @@ config RISCV_ISA_C > > when building U-Boot, which results in compressed instructions > in the > > U-Boot binary. > > > > +config RISCV_ISA_F > > + bool "Emit Floating-point instructions" > > + default n > > this can be dropped as default is n OK, I'll drop it > > + help > > + Adds "F" to the ISA subsets that the toolchain is allowed to emit > > + when building U-Boot, which results in Single and > > + Double-precision instructions > > This does not match what the config name says. The config name is for F and it > cannot indicate both F and D here. > OK, I'll correct it as follows to cover F and D: config RISCV_ISA_FD bool "Emit Floating-point instructions" help Adds "F" and "D" to the ISA subsets that the toolchain is allowed to emit when building U-Boot, which results in Single and Double-precision instructions in the U-Boot binary. > > + in the U-Boot binary. > > + > > config RISCV_ISA_A > > def_bool y > > > > diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile index > > 0b80eb8d86..87ec0ea4b5 100644 > > --- a/arch/riscv/Makefile > > +++ b/arch/riscv/Makefile > > @@ -5,15 +5,19 @@ > > > > ifeq ($(CONFIG_ARCH_RV64I),y) > > ARCH_BASE = rv64im > > - ABI = lp64 > > + ABI := lp64 > > endif > > ifeq ($(CONFIG_ARCH_RV32I),y) > > ARCH_BASE = rv32im > > - ABI = ilp32 > > + ABI := ilp32 > > endif > > ifeq ($(CONFIG_RISCV_ISA_A),y) > > ARCH_A = a > > endif > > +ifeq ($(CONFIG_RISCV_ISA_F),y) > > + ARCH_F = fd ifeq ($(CONFIG_RISCV_ISA_FD),y) ARCH_FD = fd > > + ABI := $(ABI)d > > +endif > > ifeq ($(CONFIG_RISCV_ISA_C),y) > > ARCH_C = c > > endif > > @@ -24,8 +28,8 @@ ifeq ($(CONFIG_CMODEL_MEDANY),y) > > CMODEL = medany > > endif > > > > -ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_C) -mabi=$(ABI) \ > > - -mcmodel=$(CMODEL) > > +ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_F)$(ARCH_C) ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_FD)$(ARCH_C) > -mabi=$(ABI) \ > > + -mcmodel=$(CMODEL) > > > > PLATFORM_CPPFLAGS += $(ARCH_FLAGS) > > CFLAGS_EFI += $(ARCH_FLAGS) > > -- > > Regards, > Bin Thanks for your review. Regards, Eric CONFIDENTIALITY NOTICE: This e-mail (and its attachments) may contain confidential and legally privileged information or information protected from disclosure. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution, or use of the information contained herein is strictly prohibited. In this case, please immediately notify the sender by return e-mail, delete the message (and any accompanying documents) and destroy all printed hard copies. Thank you for your cooperation. Copyright ANDES TECHNOLOGY CORPORATION - All Rights Reserved.
Hi Eric, On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > Hi Bin, > > > -----Original Message----- > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > Sent: Tuesday, May 21, 2019 3:56 PM > > To: Eric Te-Sheng Lin(林德晟) > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(陳建志); > > Greentime Ying-Han Hu(胡英漢); dslin1010@gmail.com > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > support > > > > Hi Eric, > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > adds > > > > OK I'll correct it as adds > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > extension? > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. I don't understand. What difference do these two toochains have? Isn't the v5d toolchain's default -march string be pre-configured to imafd? But even if the toolchain is pre-configured to generate fd instruction, I think it can be override by the compiler flags. Can you please share the details of the toolchain you used? I suspect you have to fix your toolchain, not U-Boot. > > > > Signed-off-by: Eric Lin <tesheng@andestech.com> > > > --- > > > arch/riscv/Kconfig | 8 ++++++++ > > > arch/riscv/Makefile | 12 ++++++++---- > > > 2 files changed, 16 insertions(+), 4 deletions(-) > > > > > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index > > > 362f3cdc65..a8031fa230 100644 > > > --- a/arch/riscv/Kconfig > > > +++ b/arch/riscv/Kconfig > > > @@ -91,6 +91,14 @@ config RISCV_ISA_C > > > when building U-Boot, which results in compressed instructions > > in the > > > U-Boot binary. > > > > > > +config RISCV_ISA_F > > > + bool "Emit Floating-point instructions" > > > + default n > > > > this can be dropped as default is n > > OK, I'll drop it > > > > + help > > > + Adds "F" to the ISA subsets that the toolchain is allowed to emit > > > + when building U-Boot, which results in Single and > > > + Double-precision instructions > > > > This does not match what the config name says. The config name is for F and it > > cannot indicate both F and D here. > > > > OK, I'll correct it as follows to cover F and D: > config RISCV_ISA_FD > bool "Emit Floating-point instructions" > help > Adds "F" and "D" to the ISA subsets that the toolchain is allowed to emit > when building U-Boot, which results in Single and Double-precision instructions > in the U-Boot binary. > > > > > + in the U-Boot binary. > > > + > > > config RISCV_ISA_A > > > def_bool y > > > > > > diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile index > > > 0b80eb8d86..87ec0ea4b5 100644 > > > --- a/arch/riscv/Makefile > > > +++ b/arch/riscv/Makefile > > > @@ -5,15 +5,19 @@ > > > > > > ifeq ($(CONFIG_ARCH_RV64I),y) > > > ARCH_BASE = rv64im > > > - ABI = lp64 > > > + ABI := lp64 > > > endif > > > ifeq ($(CONFIG_ARCH_RV32I),y) > > > ARCH_BASE = rv32im > > > - ABI = ilp32 > > > + ABI := ilp32 > > > endif > > > ifeq ($(CONFIG_RISCV_ISA_A),y) > > > ARCH_A = a > > > endif > > > +ifeq ($(CONFIG_RISCV_ISA_F),y) > > > + ARCH_F = fd > > ifeq ($(CONFIG_RISCV_ISA_FD),y) > ARCH_FD = fd > > > > + ABI := $(ABI)d > > > +endif > > > ifeq ($(CONFIG_RISCV_ISA_C),y) > > > ARCH_C = c > > > endif > > > @@ -24,8 +28,8 @@ ifeq ($(CONFIG_CMODEL_MEDANY),y) > > > CMODEL = medany > > > endif > > > > > > -ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_C) -mabi=$(ABI) \ > > > - -mcmodel=$(CMODEL) > > > +ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_F)$(ARCH_C) > > ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_FD)$(ARCH_C) > > > -mabi=$(ABI) \ > > > + -mcmodel=$(CMODEL) > > > > > > PLATFORM_CPPFLAGS += $(ARCH_FLAGS) > > > CFLAGS_EFI += $(ARCH_FLAGS) Regards, Bin
Hi Bin Bin Meng <bmeng.cn@gmail.com> 於 2019年5月22日 週三 下午5:25寫道： > > Hi Eric, > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > Hi Bin, > > > > > -----Original Message----- > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > To: Eric Te-Sheng Lin(林德晟) > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(陳建志); > > > Greentime Ying-Han Hu(胡英漢); dslin1010@gmail.com > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > support > > > > > > Hi Eric, > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > adds > > > > > > > OK I'll correct it as adds > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > extension? > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > I don't understand. What difference do these two toochains have? Isn't > the v5d toolchain's default -march string be pre-configured to imafd? > But even if the toolchain is pre-configured to generate fd > instruction, I think it can be override by the compiler flags. Can you > please share the details of the toolchain you used? I suspect you have > to fix your toolchain, not U-Boot. > It's seems the ABI issue. Because our toolchain don't support multilib, the v5d toolchain libraries ABI is ilp64d. If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will get link error as below: ... riscv64-linux-ld.bfd: /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o): can't link double-float modules with soft-float modules riscv64-linux-ld.bfd: failed to merge target specific data of file /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o) examples/standalone/Makefile:62: recipe for target 'examples/standalone/hello_world' failed ... so we would like to override the compiler flag mabi to lp64d. > > > > > > Signed-off-by: Eric Lin <tesheng@andestech.com> > > > > --- > > > > arch/riscv/Kconfig | 8 ++++++++ > > > > arch/riscv/Makefile | 12 ++++++++---- > > > > 2 files changed, 16 insertions(+), 4 deletions(-) > > > > > > > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index > > > > 362f3cdc65..a8031fa230 100644 > > > > --- a/arch/riscv/Kconfig > > > > +++ b/arch/riscv/Kconfig > > > > @@ -91,6 +91,14 @@ config RISCV_ISA_C > > > > when building U-Boot, which results in compressed instructions > > > in the > > > > U-Boot binary. > > > > > > > > +config RISCV_ISA_F > > > > + bool "Emit Floating-point instructions" > > > > + default n > > > > > > this can be dropped as default is n > > > > OK, I'll drop it > > > > > > + help > > > > + Adds "F" to the ISA subsets that the toolchain is allowed to emit > > > > + when building U-Boot, which results in Single and > > > > + Double-precision instructions > > > > > > This does not match what the config name says. The config name is for F and it > > > cannot indicate both F and D here. > > > > > > > OK, I'll correct it as follows to cover F and D: > > config RISCV_ISA_FD > > bool "Emit Floating-point instructions" > > help > > Adds "F" and "D" to the ISA subsets that the toolchain is allowed to emit > > when building U-Boot, which results in Single and Double-precision instructions > > in the U-Boot binary. > > > > > > > > + in the U-Boot binary. > > > > + > > > > config RISCV_ISA_A > > > > def_bool y > > > > > > > > diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile index > > > > 0b80eb8d86..87ec0ea4b5 100644 > > > > --- a/arch/riscv/Makefile > > > > +++ b/arch/riscv/Makefile > > > > @@ -5,15 +5,19 @@ > > > > > > > > ifeq ($(CONFIG_ARCH_RV64I),y) > > > > ARCH_BASE = rv64im > > > > - ABI = lp64 > > > > + ABI := lp64 > > > > endif > > > > ifeq ($(CONFIG_ARCH_RV32I),y) > > > > ARCH_BASE = rv32im > > > > - ABI = ilp32 > > > > + ABI := ilp32 > > > > endif > > > > ifeq ($(CONFIG_RISCV_ISA_A),y) > > > > ARCH_A = a > > > > endif > > > > +ifeq ($(CONFIG_RISCV_ISA_F),y) > > > > + ARCH_F = fd > > > > ifeq ($(CONFIG_RISCV_ISA_FD),y) > > ARCH_FD = fd > > > > > > + ABI := $(ABI)d > > > > +endif > > > > ifeq ($(CONFIG_RISCV_ISA_C),y) > > > > ARCH_C = c > > > > endif > > > > @@ -24,8 +28,8 @@ ifeq ($(CONFIG_CMODEL_MEDANY),y) > > > > CMODEL = medany > > > > endif > > > > > > > > -ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_C) -mabi=$(ABI) \ > > > > - -mcmodel=$(CMODEL) > > > > +ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_F)$(ARCH_C) > > > > ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_FD)$(ARCH_C) > > > > > -mabi=$(ABI) \ > > > > + -mcmodel=$(CMODEL) > > > > > > > > PLATFORM_CPPFLAGS += $(ARCH_FLAGS) > > > > CFLAGS_EFI += $(ARCH_FLAGS) > > > Regards, > Bin Thanks for your review Eric
Hi Bin Bin Meng <bmeng.cn@gmail.com> 於 2019年5月22日 週三 下午5:25寫道： > > Hi Eric, > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > Hi Bin, > > > > > -----Original Message----- > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > To: Eric Te-Sheng Lin(林德晟) > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(陳建志); > > > Greentime Ying-Han Hu(胡英漢); dslin1010@gmail.com > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > support > > > > > > Hi Eric, > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > adds > > > > > > > OK I'll correct it as adds > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > extension? > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > I don't understand. What difference do these two toochains have? Isn't > the v5d toolchain's default -march string be pre-configured to imafd? > But even if the toolchain is pre-configured to generate fd > instruction, I think it can be override by the compiler flags. Can you > please share the details of the toolchain you used? I suspect you have > to fix your toolchain, not U-Boot. > It's seems the ABI issue. Because our toolchain don't support multilib, the v5d toolchain libraries ABI is ilp64d. If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will get link error as below: ... riscv64-linux-ld.bfd: /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o): can't link double-float modules with soft-float modules riscv64-linux-ld.bfd: failed to merge target specific data of file /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o) examples/standalone/Makefile:62: recipe for target 'examples/standalone/hello_world' failed ... so we would like to override the compiler flag mabi to lp64d. > > > > > > Signed-off-by: Eric Lin <tesheng@andestech.com> > > > > --- > > > > arch/riscv/Kconfig | 8 ++++++++ > > > > arch/riscv/Makefile | 12 ++++++++---- > > > > 2 files changed, 16 insertions(+), 4 deletions(-) > > > > > > > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index > > > > 362f3cdc65..a8031fa230 100644 > > > > --- a/arch/riscv/Kconfig > > > > +++ b/arch/riscv/Kconfig > > > > @@ -91,6 +91,14 @@ config RISCV_ISA_C > > > > when building U-Boot, which results in compressed instructions > > > in the > > > > U-Boot binary. > > > > > > > > +config RISCV_ISA_F > > > > + bool "Emit Floating-point instructions" > > > > + default n > > > > > > this can be dropped as default is n > > > > OK, I'll drop it > > > > > > + help > > > > + Adds "F" to the ISA subsets that the toolchain is allowed to emit > > > > + when building U-Boot, which results in Single and > > > > + Double-precision instructions > > > > > > This does not match what the config name says. The config name is for F and it > > > cannot indicate both F and D here. > > > > > > > OK, I'll correct it as follows to cover F and D: > > config RISCV_ISA_FD > > bool "Emit Floating-point instructions" > > help > > Adds "F" and "D" to the ISA subsets that the toolchain is allowed to emit > > when building U-Boot, which results in Single and Double-precision instructions > > in the U-Boot binary. > > > > > > > > + in the U-Boot binary. > > > > + > > > > config RISCV_ISA_A > > > > def_bool y > > > > > > > > diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile index > > > > 0b80eb8d86..87ec0ea4b5 100644 > > > > --- a/arch/riscv/Makefile > > > > +++ b/arch/riscv/Makefile > > > > @@ -5,15 +5,19 @@ > > > > > > > > ifeq ($(CONFIG_ARCH_RV64I),y) > > > > ARCH_BASE = rv64im > > > > - ABI = lp64 > > > > + ABI := lp64 > > > > endif > > > > ifeq ($(CONFIG_ARCH_RV32I),y) > > > > ARCH_BASE = rv32im > > > > - ABI = ilp32 > > > > + ABI := ilp32 > > > > endif > > > > ifeq ($(CONFIG_RISCV_ISA_A),y) > > > > ARCH_A = a > > > > endif > > > > +ifeq ($(CONFIG_RISCV_ISA_F),y) > > > > + ARCH_F = fd > > > > ifeq ($(CONFIG_RISCV_ISA_FD),y) > > ARCH_FD = fd > > > > > > + ABI := $(ABI)d > > > > +endif > > > > ifeq ($(CONFIG_RISCV_ISA_C),y) > > > > ARCH_C = c > > > > endif > > > > @@ -24,8 +28,8 @@ ifeq ($(CONFIG_CMODEL_MEDANY),y) > > > > CMODEL = medany > > > > endif > > > > > > > > -ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_C) -mabi=$(ABI) \ > > > > - -mcmodel=$(CMODEL) > > > > +ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_F)$(ARCH_C) > > > > ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_FD)$(ARCH_C) > > > > > -mabi=$(ABI) \ > > > > + -mcmodel=$(CMODEL) > > > > > > > > PLATFORM_CPPFLAGS += $(ARCH_FLAGS) > > > > CFLAGS_EFI += $(ARCH_FLAGS) > > > Regards, > Bin Thanks for your review Eric
Hi Bin Bin Meng <bmeng.cn@gmail.com> 於 2019年5月22日 週三 下午5:25寫道： > > Hi Eric, > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > Hi Bin, > > > > > -----Original Message----- > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > To: Eric Te-Sheng Lin(林德晟) > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(陳建志); > > > Greentime Ying-Han Hu(胡英漢); dslin1010@gmail.com > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > support > > > > > > Hi Eric, > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > adds > > > > > > > OK I'll correct it as adds > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > extension? > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > I don't understand. What difference do these two toochains have? Isn't > the v5d toolchain's default -march string be pre-configured to imafd? > But even if the toolchain is pre-configured to generate fd > instruction, I think it can be override by the compiler flags. Can you > please share the details of the toolchain you used? I suspect you have > to fix your toolchain, not U-Boot. > It's seems the ABI issue. Because our toolchain don't support multilib, the v5d toolchain libraries ABI is ilp64d. If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will get link error as below: ... riscv64-linux-ld.bfd: /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o): can't link double-float modules with soft-float modules riscv64-linux-ld.bfd: failed to merge target specific data of file /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o) examples/standalone/Makefile:62: recipe for target 'examples/standalone/hello_world' failed ... so we would like to override the compiler flag mabi to lp64d. > > > > > > Signed-off-by: Eric Lin <tesheng@andestech.com> > > > > --- > > > > arch/riscv/Kconfig | 8 ++++++++ > > > > arch/riscv/Makefile | 12 ++++++++---- > > > > 2 files changed, 16 insertions(+), 4 deletions(-) > > > > > > > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index > > > > 362f3cdc65..a8031fa230 100644 > > > > --- a/arch/riscv/Kconfig > > > > +++ b/arch/riscv/Kconfig > > > > @@ -91,6 +91,14 @@ config RISCV_ISA_C > > > > when building U-Boot, which results in compressed instructions > > > in the > > > > U-Boot binary. > > > > > > > > +config RISCV_ISA_F > > > > + bool "Emit Floating-point instructions" > > > > + default n > > > > > > this can be dropped as default is n > > > > OK, I'll drop it > > > > > > + help > > > > + Adds "F" to the ISA subsets that the toolchain is allowed to emit > > > > + when building U-Boot, which results in Single and > > > > + Double-precision instructions > > > > > > This does not match what the config name says. The config name is for F and it > > > cannot indicate both F and D here. > > > > > > > OK, I'll correct it as follows to cover F and D: > > config RISCV_ISA_FD > > bool "Emit Floating-point instructions" > > help > > Adds "F" and "D" to the ISA subsets that the toolchain is allowed to emit > > when building U-Boot, which results in Single and Double-precision instructions > > in the U-Boot binary. > > > > > > > > + in the U-Boot binary. > > > > + > > > > config RISCV_ISA_A > > > > def_bool y > > > > > > > > diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile index > > > > 0b80eb8d86..87ec0ea4b5 100644 > > > > --- a/arch/riscv/Makefile > > > > +++ b/arch/riscv/Makefile > > > > @@ -5,15 +5,19 @@ > > > > > > > > ifeq ($(CONFIG_ARCH_RV64I),y) > > > > ARCH_BASE = rv64im > > > > - ABI = lp64 > > > > + ABI := lp64 > > > > endif > > > > ifeq ($(CONFIG_ARCH_RV32I),y) > > > > ARCH_BASE = rv32im > > > > - ABI = ilp32 > > > > + ABI := ilp32 > > > > endif > > > > ifeq ($(CONFIG_RISCV_ISA_A),y) > > > > ARCH_A = a > > > > endif > > > > +ifeq ($(CONFIG_RISCV_ISA_F),y) > > > > + ARCH_F = fd > > > > ifeq ($(CONFIG_RISCV_ISA_FD),y) > > ARCH_FD = fd > > > > > > + ABI := $(ABI)d > > > > +endif > > > > ifeq ($(CONFIG_RISCV_ISA_C),y) > > > > ARCH_C = c > > > > endif > > > > @@ -24,8 +28,8 @@ ifeq ($(CONFIG_CMODEL_MEDANY),y) > > > > CMODEL = medany > > > > endif > > > > > > > > -ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_C) -mabi=$(ABI) \ > > > > - -mcmodel=$(CMODEL) > > > > +ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_F)$(ARCH_C) > > > > ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_FD)$(ARCH_C) > > > > > -mabi=$(ABI) \ > > > > + -mcmodel=$(CMODEL) > > > > > > > > PLATFORM_CPPFLAGS += $(ARCH_FLAGS) > > > > CFLAGS_EFI += $(ARCH_FLAGS) > > > Regards, > Bin Thanks for your review Eric
Hi Eric, On Fri, May 24, 2019 at 4:04 PM Eric Lin <dslin1010@gmail.com> wrote: > > Hi Bin > > Bin Meng <bmeng.cn@gmail.com> 於 2019年5月22日 週三 下午5:25寫道： > > > > Hi Eric, > > > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > > > Hi Bin, > > > > > > > -----Original Message----- > > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > > To: Eric Te-Sheng Lin(林德晟) > > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(陳建志); > > > > Greentime Ying-Han Hu(胡英漢); dslin1010@gmail.com > > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > > support > > > > > > > > Hi Eric, > > > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > > > adds > > > > > > > > > > OK I'll correct it as adds > > > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > > extension? > > > > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > > > I don't understand. What difference do these two toochains have? Isn't > > the v5d toolchain's default -march string be pre-configured to imafd? > > But even if the toolchain is pre-configured to generate fd > > instruction, I think it can be override by the compiler flags. Can you > > please share the details of the toolchain you used? I suspect you have > > to fix your toolchain, not U-Boot. > > > > It's seems the ABI issue. Because our toolchain don't support > multilib, the v5d toolchain libraries ABI is ilp64d. > If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will > get link error as below: Then please fix your compiler. Passing -mabi=lp64 to compiler is definitely correct. > ... > riscv64-linux-ld.bfd: > /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o): > can't link double-float modules with soft-float modules > riscv64-linux-ld.bfd: failed to merge target specific data of file > /nds64le-linux-glibc-v5d/bin/../lib/gcc/riscv64-linux/7.3.0/libgcc.a(save-restore.o) > examples/standalone/Makefile:62: recipe for target > 'examples/standalone/hello_world' failed > ... > > so we would like to override the compiler flag mabi to lp64d. No, adding this to U-Boot is a workaround to your compiler. Regards, Bin
Hi Bin, On Fri, May 24, 2019 at 05:40:08PM +0800, Bin Meng wrote: > Hi Eric, > > On Fri, May 24, 2019 at 4:04 PM Eric Lin <dslin1010@gmail.com> wrote: > > > > Hi Bin > > > > Bin Meng <bmeng.cn@gmail.com> ??? 2019???5???22??? ?????? ??????5:25????????? > > > > > > Hi Eric, > > > > > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > > > > > Hi Bin, > > > > > > > > > -----Original Message----- > > > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > > > To: Eric Te-Sheng Lin(?????????) > > > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(?????????); > > > > > Greentime Ying-Han Hu(?????????); dslin1010@gmail.com > > > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > > > support > > > > > > > > > > Hi Eric, > > > > > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > > > > > adds > > > > > > > > > > > > > OK I'll correct it as adds > > > > > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > > > extension? > > > > > > > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > > > > > I don't understand. What difference do these two toochains have? Isn't > > > the v5d toolchain's default -march string be pre-configured to imafd? > > > But even if the toolchain is pre-configured to generate fd > > > instruction, I think it can be override by the compiler flags. Can you > > > please share the details of the toolchain you used? I suspect you have > > > to fix your toolchain, not U-Boot. > > > > > > > It's seems the ABI issue. Because our toolchain don't support > > multilib, the v5d toolchain libraries ABI is ilp64d. > > If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will > > get link error as below: > > Then please fix your compiler. Passing -mabi=lp64 to compiler is > definitely correct. One of our customer ever used floating point in their project. They want to calculate something with floating point instructions in u-Boot before entering kernel. IMHO, we may consider this case for user to use -mabi=ilp64d. This CONFIG will be useful for them to compile their codes and linking together with u-Boot. Thanks :)
Hi Greentime, On Tue, May 28, 2019 at 3:54 PM Greentime Hu <greentime@andestech.com> wrote: > > Hi Bin, > On Fri, May 24, 2019 at 05:40:08PM +0800, Bin Meng wrote: > > Hi Eric, > > > > On Fri, May 24, 2019 at 4:04 PM Eric Lin <dslin1010@gmail.com> wrote: > > > > > > Hi Bin > > > > > > Bin Meng <bmeng.cn@gmail.com> ??? 2019???5???22??? ?????? ??????5:25????????? > > > > > > > > Hi Eric, > > > > > > > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > > > > > > > Hi Bin, > > > > > > > > > > > -----Original Message----- > > > > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > > > > To: Eric Te-Sheng Lin(?????????) > > > > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(?????????); > > > > > > Greentime Ying-Han Hu(?????????); dslin1010@gmail.com > > > > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > > > > support > > > > > > > > > > > > Hi Eric, > > > > > > > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > > > > > > > adds > > > > > > > > > > > > > > > > OK I'll correct it as adds > > > > > > > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > > > > extension? > > > > > > > > > > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > > > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > > > > > > > I don't understand. What difference do these two toochains have? Isn't > > > > the v5d toolchain's default -march string be pre-configured to imafd? > > > > But even if the toolchain is pre-configured to generate fd > > > > instruction, I think it can be override by the compiler flags. Can you > > > > please share the details of the toolchain you used? I suspect you have > > > > to fix your toolchain, not U-Boot. > > > > > > > > > > It's seems the ABI issue. Because our toolchain don't support > > > multilib, the v5d toolchain libraries ABI is ilp64d. > > > If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will > > > get link error as below: > > > > Then please fix your compiler. Passing -mabi=lp64 to compiler is > > definitely correct. > > One of our customer ever used floating point in their project. They want > to calculate something with floating point instructions in u-Boot before > entering kernel. This is a new use case. For other architectures like x86, we don't allow U-Boot to be compiled using hardware float ABI. Why can't they use software floating point instead? > > IMHO, we may consider this case for user to use -mabi=ilp64d. > This CONFIG will be useful for them to compile their codes and linking together > with u-Boot. > > Thanks :) Regards, Bin
Hi Bin, On Tue, May 28, 2019 at 05:58:47PM +0800, Bin Meng wrote: > Hi Greentime, > > On Tue, May 28, 2019 at 3:54 PM Greentime Hu <greentime@andestech.com> wrote: > > > > Hi Bin, > > On Fri, May 24, 2019 at 05:40:08PM +0800, Bin Meng wrote: > > > Hi Eric, > > > > > > On Fri, May 24, 2019 at 4:04 PM Eric Lin <dslin1010@gmail.com> wrote: > > > > > > > > Hi Bin > > > > > > > > Bin Meng <bmeng.cn@gmail.com> ??? 2019???5???22??? ?????? ??????5:25????????? > > > > > > > > > > Hi Eric, > > > > > > > > > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > > > > > > > > > Hi Bin, > > > > > > > > > > > > > -----Original Message----- > > > > > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > > > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > > > > > To: Eric Te-Sheng Lin(?????????) > > > > > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(?????????); > > > > > > > Greentime Ying-Han Hu(?????????); dslin1010@gmail.com > > > > > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > > > > > support > > > > > > > > > > > > > > Hi Eric, > > > > > > > > > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > > > > > > > > > adds > > > > > > > > > > > > > > > > > > > OK I'll correct it as adds > > > > > > > > > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > > > > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > > > > > extension? > > > > > > > > > > > > > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > > > > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > > > > > > > > > I don't understand. What difference do these two toochains have? Isn't > > > > > the v5d toolchain's default -march string be pre-configured to imafd? > > > > > But even if the toolchain is pre-configured to generate fd > > > > > instruction, I think it can be override by the compiler flags. Can you > > > > > please share the details of the toolchain you used? I suspect you have > > > > > to fix your toolchain, not U-Boot. > > > > > > > > > > > > > It's seems the ABI issue. Because our toolchain don't support > > > > multilib, the v5d toolchain libraries ABI is ilp64d. > > > > If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will > > > > get link error as below: > > > > > > Then please fix your compiler. Passing -mabi=lp64 to compiler is > > > definitely correct. > > > > One of our customer ever used floating point in their project. They want > > to calculate something with floating point instructions in u-Boot before > > entering kernel. > > This is a new use case. For other architectures like x86, we don't > allow U-Boot to be compiled using hardware float ABI. Why can't they > use software floating point instead? It's because they'd like to calculate something very quickly. It may just be done in few cycles with floating point instructions and it will be very useful for their usages instead of using soft-float. That's the reason they ask us to provide such kind of CONFIG and we think it might also be helpful for users who are trying to do some mathematics with floating point instructions in u-Boot. Thanks. :) > > > > IMHO, we may consider this case for user to use -mabi=ilp64d. > > This CONFIG will be useful for them to compile their codes and linking together > > with u-Boot. > > > > Thanks :) > > Regards, > Bin
On Tue, May 28, 2019 at 6:50 PM Greentime Hu <greentime@andestech.com> wrote: > > Hi Bin, > On Tue, May 28, 2019 at 05:58:47PM +0800, Bin Meng wrote: > > Hi Greentime, > > > > On Tue, May 28, 2019 at 3:54 PM Greentime Hu <greentime@andestech.com> wrote: > > > > > > Hi Bin, > > > On Fri, May 24, 2019 at 05:40:08PM +0800, Bin Meng wrote: > > > > Hi Eric, > > > > > > > > On Fri, May 24, 2019 at 4:04 PM Eric Lin <dslin1010@gmail.com> wrote: > > > > > > > > > > Hi Bin > > > > > > > > > > Bin Meng <bmeng.cn@gmail.com> ??? 2019???5???22??? ?????? ??????5:25????????? > > > > > > > > > > > > Hi Eric, > > > > > > > > > > > > On Wed, May 22, 2019 at 4:23 PM <tesheng@andestech.com> wrote: > > > > > > > > > > > > > > Hi Bin, > > > > > > > > > > > > > > > -----Original Message----- > > > > > > > > From: Bin Meng [mailto:bmeng.cn@gmail.com] > > > > > > > > Sent: Tuesday, May 21, 2019 3:56 PM > > > > > > > > To: Eric Te-Sheng Lin(?????????) > > > > > > > > Cc: U-Boot Mailing List; Lukas Auer; Anup Patel; Rick Jian-Zhi Chen(?????????); > > > > > > > > Greentime Ying-Han Hu(?????????); dslin1010@gmail.com > > > > > > > > Subject: Re: [PATCH] riscv: add Kconfig entries for the F and D ISA extensions > > > > > > > > support > > > > > > > > > > > > > > > > Hi Eric, > > > > > > > > > > > > > > > > On Tue, May 21, 2019 at 3:18 PM Eric Lin <tesheng@andestech.com> wrote: > > > > > > > > > > > > > > > > > > This patch add Kconfig entries for the F (Single-Precision) > > > > > > > > > > > > > > > > adds > > > > > > > > > > > > > > > > > > > > > > OK I'll correct it as adds > > > > > > > > > > > > > > > > and D (Double-Precision) floating point instruction-set extensions. > > > > > > > > > > > > > > > > > > > > > > > > > Could you please provide reason that why U-Boot has to be compiled using F/D > > > > > > > > extension? > > > > > > > > > > > > > > > > > > > > > > Cause on AE350 platform, we have two different kinds of toolchain v5d (support I/M/A/C/F/D ISA) and > > > > > > > v5 (support I/M/A/C ISA). If we use the v5d toolchain to build U-Boot it will build fail, so we would like to add F/D extension on U-Boot. > > > > > > > > > > > > I don't understand. What difference do these two toochains have? Isn't > > > > > > the v5d toolchain's default -march string be pre-configured to imafd? > > > > > > But even if the toolchain is pre-configured to generate fd > > > > > > instruction, I think it can be override by the compiler flags. Can you > > > > > > please share the details of the toolchain you used? I suspect you have > > > > > > to fix your toolchain, not U-Boot. > > > > > > > > > > > > > > > > It's seems the ABI issue. Because our toolchain don't support > > > > > multilib, the v5d toolchain libraries ABI is ilp64d. > > > > > If I use the v5d toolchain to build U-Boot with -mabi=lp64, it will > > > > > get link error as below: > > > > > > > > Then please fix your compiler. Passing -mabi=lp64 to compiler is > > > > definitely correct. > > > > > > One of our customer ever used floating point in their project. They want > > > to calculate something with floating point instructions in u-Boot before > > > entering kernel. > > > > This is a new use case. For other architectures like x86, we don't > > allow U-Boot to be compiled using hardware float ABI. Why can't they > > use software floating point instead? > > It's because they'd like to calculate something very quickly. It may just be > done in few cycles with floating point instructions and it will be very useful > for their usages instead of using soft-float. > > That's the reason they ask us to provide such kind of CONFIG and we think it > might also be helpful for users who are trying to do some mathematics with > floating point instructions in u-Boot. > I will have to leave this to other people to comment. I personally do not like this to be supported in the upstream U-Boot. However if others do not object, I am OK too. Regards, Bin
diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index 362f3cdc65..a8031fa230 100644 --- a/arch/riscv/Kconfig +++ b/arch/riscv/Kconfig @@ -91,6 +91,14 @@ config RISCV_ISA_C when building U-Boot, which results in compressed instructions in the U-Boot binary. +config RISCV_ISA_F + bool "Emit Floating-point instructions" + default n + help + Adds "F" to the ISA subsets that the toolchain is allowed to emit + when building U-Boot, which results in Single and Double-precision instructions + in the U-Boot binary. + config RISCV_ISA_A def_bool y diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile index 0b80eb8d86..87ec0ea4b5 100644 --- a/arch/riscv/Makefile +++ b/arch/riscv/Makefile @@ -5,15 +5,19 @@ ifeq ($(CONFIG_ARCH_RV64I),y) ARCH_BASE = rv64im - ABI = lp64 + ABI := lp64 endif ifeq ($(CONFIG_ARCH_RV32I),y) ARCH_BASE = rv32im - ABI = ilp32 + ABI := ilp32 endif ifeq ($(CONFIG_RISCV_ISA_A),y) ARCH_A = a endif +ifeq ($(CONFIG_RISCV_ISA_F),y) + ARCH_F = fd + ABI := $(ABI)d +endif ifeq ($(CONFIG_RISCV_ISA_C),y) ARCH_C = c endif @@ -24,8 +28,8 @@ ifeq ($(CONFIG_CMODEL_MEDANY),y) CMODEL = medany endif -ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_C) -mabi=$(ABI) \ - -mcmodel=$(CMODEL) +ARCH_FLAGS = -march=$(ARCH_BASE)$(ARCH_A)$(ARCH_F)$(ARCH_C) -mabi=$(ABI) \ + -mcmodel=$(CMODEL) PLATFORM_CPPFLAGS += $(ARCH_FLAGS) CFLAGS_EFI += $(ARCH_FLAGS)
This patch add Kconfig entries for the F (Single-Precision) and D (Double-Precision) floating point instruction-set extensions. Signed-off-by: Eric Lin <tesheng@andestech.com> --- arch/riscv/Kconfig | 8 ++++++++ arch/riscv/Makefile | 12 ++++++++---- 2 files changed, 16 insertions(+), 4 deletions(-)