2024-05-07 10:43:25

by Wang, Xiao W

[permalink] [raw]
Subject: [PATCH] riscv, bpf: Optimize zextw insn with Zba extension

The Zba extension provides add.uw insn which can be used to implement
zext.w with rs2 set as ZERO.

Signed-off-by: Xiao Wang <[email protected]>
---
arch/riscv/Kconfig | 19 +++++++++++++++++++
arch/riscv/net/bpf_jit.h | 18 ++++++++++++++++++
2 files changed, 37 insertions(+)

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index 6bec1bce6586..0679127cc0ea 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -586,6 +586,14 @@ config RISCV_ISA_V_PREEMPTIVE
preemption. Enabling this config will result in higher memory
consumption due to the allocation of per-task's kernel Vector context.

+config TOOLCHAIN_HAS_ZBA
+ bool
+ default y
+ depends on !64BIT || $(cc-option,-mabi=lp64 -march=rv64ima_zba)
+ depends on !32BIT || $(cc-option,-mabi=ilp32 -march=rv32ima_zba)
+ depends on LLD_VERSION >= 150000 || LD_VERSION >= 23900
+ depends on AS_HAS_OPTION_ARCH
+
config TOOLCHAIN_HAS_ZBB
bool
default y
@@ -601,6 +609,17 @@ config TOOLCHAIN_HAS_VECTOR_CRYPTO
def_bool $(as-instr, .option arch$(comma) +v$(comma) +zvkb)
depends on AS_HAS_OPTION_ARCH

+config RISCV_ISA_ZBA
+ bool "Zba extension support for bit manipulation instructions"
+ depends on TOOLCHAIN_HAS_ZBA
+ depends on RISCV_ALTERNATIVE
+ default y
+ help
+ Adds support to dynamically detect the presence of the ZBA
+ extension (address generation acceleration) and enable its usage.
+
+ If you don't know what to do here, say Y.
+
config RISCV_ISA_ZBB
bool "Zbb extension support for bit manipulation instructions"
depends on TOOLCHAIN_HAS_ZBB
diff --git a/arch/riscv/net/bpf_jit.h b/arch/riscv/net/bpf_jit.h
index f4b6b3b9edda..18a7885ba95e 100644
--- a/arch/riscv/net/bpf_jit.h
+++ b/arch/riscv/net/bpf_jit.h
@@ -18,6 +18,11 @@ static inline bool rvc_enabled(void)
return IS_ENABLED(CONFIG_RISCV_ISA_C);
}

+static inline bool rvzba_enabled(void)
+{
+ return IS_ENABLED(CONFIG_RISCV_ISA_ZBA) && riscv_has_extension_likely(RISCV_ISA_EXT_ZBA);
+}
+
static inline bool rvzbb_enabled(void)
{
return IS_ENABLED(CONFIG_RISCV_ISA_ZBB) && riscv_has_extension_likely(RISCV_ISA_EXT_ZBB);
@@ -937,6 +942,14 @@ static inline u16 rvc_sdsp(u32 imm9, u8 rs2)
return rv_css_insn(0x7, imm, rs2, 0x2);
}

+/* RV64-only ZBA instructions. */
+
+static inline u32 rvzba_zextw(u8 rd, u8 rs1)
+{
+ /* add.uw rd, rs1, ZERO */
+ return rv_r_insn(0x04, RV_REG_ZERO, rs1, 0, rd, 0x3b);
+}
+
#endif /* __riscv_xlen == 64 */

/* Helper functions that emit RVC instructions when possible. */
@@ -1159,6 +1172,11 @@ static inline void emit_zexth(u8 rd, u8 rs, struct rv_jit_context *ctx)

static inline void emit_zextw(u8 rd, u8 rs, struct rv_jit_context *ctx)
{
+ if (rvzba_enabled()) {
+ emit(rvzba_zextw(rd, rs), ctx);
+ return;
+ }
+
emit_slli(rd, rs, 32, ctx);
emit_srli(rd, rd, 32, ctx);
}
--
2.25.1



2024-05-08 11:21:13

by Wang, Xiao W

[permalink] [raw]
Subject: RE: [PATCH] riscv, bpf: Optimize zextw insn with Zba extension



> -----Original Message-----
> From: Pu Lehui <[email protected]>
> Sent: Tuesday, May 7, 2024 8:47 PM
> To: Wang, Xiao W <[email protected]>; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]
> Cc: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected]; linux-
> [email protected]; [email protected]; [email protected];
> Li, Haicheng <[email protected]>
> Subject: Re: [PATCH] riscv, bpf: Optimize zextw insn with Zba extension
>
>
> On 2024/5/7 18:45, Xiao Wang wrote:
> > The Zba extension provides add.uw insn which can be used to implement
> > zext.w with rs2 set as ZERO.
> >
> > Signed-off-by: Xiao Wang <[email protected]>
> > ---
> > arch/riscv/Kconfig | 19 +++++++++++++++++++
> > arch/riscv/net/bpf_jit.h | 18 ++++++++++++++++++
> > 2 files changed, 37 insertions(+)
> >
> > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> > index 6bec1bce6586..0679127cc0ea 100644
> > --- a/arch/riscv/Kconfig
> > +++ b/arch/riscv/Kconfig
> > @@ -586,6 +586,14 @@ config RISCV_ISA_V_PREEMPTIVE
> > preemption. Enabling this config will result in higher memory
> > consumption due to the allocation of per-task's kernel Vector
> context.
> >
> > +config TOOLCHAIN_HAS_ZBA
> > + bool
> > + default y
> > + depends on !64BIT || $(cc-option,-mabi=lp64 -march=rv64ima_zba)
> > + depends on !32BIT || $(cc-option,-mabi=ilp32 -march=rv32ima_zba)
> > + depends on LLD_VERSION >= 150000 || LD_VERSION >= 23900
> > + depends on AS_HAS_OPTION_ARCH
> > +
> > config TOOLCHAIN_HAS_ZBB
> > bool
> > default y
> > @@ -601,6 +609,17 @@ config TOOLCHAIN_HAS_VECTOR_CRYPTO
> > def_bool $(as-instr, .option arch$(comma) +v$(comma) +zvkb)
> > depends on AS_HAS_OPTION_ARCH
> >
> > +config RISCV_ISA_ZBA
> > + bool "Zba extension support for bit manipulation instructions"
> > + depends on TOOLCHAIN_HAS_ZBA
> > + depends on RISCV_ALTERNATIVE
> > + default y
> > + help
> > + Adds support to dynamically detect the presence of the ZBA
> > + extension (address generation acceleration) and enable its usage.
>
> It would be better to add Zba's function description like Zbb.

OK, would add some description on next version.

Thanks,
Xiao

>
> > +
> > + If you don't know what to do here, say Y.
> > +
> > config RISCV_ISA_ZBB
> > bool "Zbb extension support for bit manipulation instructions"
> > depends on TOOLCHAIN_HAS_ZBB
> > diff --git a/arch/riscv/net/bpf_jit.h b/arch/riscv/net/bpf_jit.h
> > index f4b6b3b9edda..18a7885ba95e 100644
> > --- a/arch/riscv/net/bpf_jit.h
> > +++ b/arch/riscv/net/bpf_jit.h
> > @@ -18,6 +18,11 @@ static inline bool rvc_enabled(void)
> > return IS_ENABLED(CONFIG_RISCV_ISA_C);
> > }
> >
> > +static inline bool rvzba_enabled(void)
> > +{
> > + return IS_ENABLED(CONFIG_RISCV_ISA_ZBA) &&
> riscv_has_extension_likely(RISCV_ISA_EXT_ZBA);
> > +}
> > +
> > static inline bool rvzbb_enabled(void)
> > {
> > return IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
> riscv_has_extension_likely(RISCV_ISA_EXT_ZBB);
> > @@ -937,6 +942,14 @@ static inline u16 rvc_sdsp(u32 imm9, u8 rs2)
> > return rv_css_insn(0x7, imm, rs2, 0x2);
> > }
> >
> > +/* RV64-only ZBA instructions. */
> > +
> > +static inline u32 rvzba_zextw(u8 rd, u8 rs1)
> > +{
> > + /* add.uw rd, rs1, ZERO */
> > + return rv_r_insn(0x04, RV_REG_ZERO, rs1, 0, rd, 0x3b);
> > +}
> > +
> > #endif /* __riscv_xlen == 64 */
> >
> > /* Helper functions that emit RVC instructions when possible. */
> > @@ -1159,6 +1172,11 @@ static inline void emit_zexth(u8 rd, u8 rs, struct
> rv_jit_context *ctx)
> >
> > static inline void emit_zextw(u8 rd, u8 rs, struct rv_jit_context *ctx)
> > {
> > + if (rvzba_enabled()) {
> > + emit(rvzba_zextw(rd, rs), ctx);
> > + return;
> > + }
>
> Looks good to me. It seems that Zba has fewer uses in rv64 bpf jit.
>
> > +
> > emit_slli(rd, rs, 32, ctx);
> > emit_srli(rd, rd, 32, ctx);
> > }

2024-05-09 09:31:48

by Wang, Xiao W

[permalink] [raw]
Subject: RE: [PATCH] riscv, bpf: Optimize zextw insn with Zba extension

Hi Ben,

> -----Original Message-----
> From: [email protected] <[email protected]>
> Sent: Tuesday, May 7, 2024 10:00 PM
> To: Pu Lehui <[email protected]>; Wang, Xiao W
> <[email protected]>; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]
> Cc: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected]; linux-
> [email protected]; [email protected]; [email protected];
> Li, Haicheng <[email protected]>
> Subject: Re: [PATCH] riscv, bpf: Optimize zextw insn with Zba extension
>
> On 07/05/2024 13:47, Pu Lehui wrote:
> >
> > On 2024/5/7 18:45, Xiao Wang wrote:
> >> The Zba extension provides add.uw insn which can be used to implement
> >> zext.w with rs2 set as ZERO.
> >>
> >> Signed-off-by: Xiao Wang <[email protected]>
> >> ---
> >>   arch/riscv/Kconfig       | 19 +++++++++++++++++++
> >>   arch/riscv/net/bpf_jit.h | 18 ++++++++++++++++++
> >>   2 files changed, 37 insertions(+)
> >>
> >> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> >> index 6bec1bce6586..0679127cc0ea 100644
> >> --- a/arch/riscv/Kconfig
> >> +++ b/arch/riscv/Kconfig
> >> @@ -586,6 +586,14 @@ config RISCV_ISA_V_PREEMPTIVE
> >>         preemption. Enabling this config will result in higher memory
> >>         consumption due to the allocation of per-task's kernel Vector
> >> context.
> >> +config TOOLCHAIN_HAS_ZBA
> >> +    bool
> >> +    default y
> >> +    depends on !64BIT || $(cc-option,-mabi=lp64 -march=rv64ima_zba)
> >> +    depends on !32BIT || $(cc-option,-mabi=ilp32 -march=rv32ima_zba)
> >> +    depends on LLD_VERSION >= 150000 || LD_VERSION >= 23900
> >> +    depends on AS_HAS_OPTION_ARCH
> >> +
> >>   config TOOLCHAIN_HAS_ZBB
>
> At this point would it be easier to ask the toolchain what's enabled
> and put into kconfig via some sort of script?

You mean to use some sort of script to automatically detect whether LD linker
can support a certain riscv extension?

I just went through the help guide of riscv64-linux-gnu-ld, and did some trial,
the "--architecture" argument usage is different from the CC's "-march". I can't find
other relevant args. It looks currently there's no direct method to dump the
supported riscv extensions of a LD.

Then I tried below test to see if we can check it via building from /dev/null, just like
the cc-option does.
Note: here CC is Zbb capable
CC -march=rv64gc_zbb -x c -c /dev/null -o tmp.o && LD tmp.o -o tmp.elf

I find that whether LD is Zbb capable or not, the above command just outputs
"warning: cannot find entry symbol _start ..", no error comes with older LD.
When we provide some Zbb-arch-tagged non-empty *.o file to the LD, then
the older LD would error out "unsupported ISA subset ..".
But I'm afraid it's not a good idea to add a source file for each extension detection.

Maybe in future there would be an easy way to ask the LD what riscv extensions
are supported.

BRs,
Xiao

>
> --
> Ben Dooks http://www.codethink.co.uk/
> Senior Engineer Codethink - Providing Genius
>
> https://www.codethink.co.uk/privacy.html