From: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>

Add support for probing RISC-V vector extension availability in
the backend. This information will be used when deciding whether
to use vector instructions in code generation.

Cache lg2(vlenb) for the backend. The storing of lg2(vlenb) means
we can convert all of the division into subtraction.

While the compiler doesn't support RISCV_HWPROBE_EXT_ZVE64X,
we use RISCV_HWPROBE_IMA_V instead. RISCV_HWPROBE_IMA_V is more
strictly constrainted than RISCV_HWPROBE_EXT_ZVE64X. At least in
current QEMU implemenation, the V vector extension depends on the
zve64d extension.

Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
Message-ID: <20241007025700.47259-2-zhiwei_liu@linux.alibaba.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 host/include/riscv/host/cpuinfo.h |  2 ++
 util/cpuinfo-riscv.c              | 34 ++++++++++++++++++++++++++++++-
 2 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/host/include/riscv/host/cpuinfo.h b/host/include/riscv/host/cpuinfo.h
index 2b00660e36..cdc784e7b6 100644
--- a/host/include/riscv/host/cpuinfo.h
+++ b/host/include/riscv/host/cpuinfo.h
@@ -10,9 +10,11 @@
 #define CPUINFO_ZBA             (1u << 1)
 #define CPUINFO_ZBB             (1u << 2)
 #define CPUINFO_ZICOND          (1u << 3)
+#define CPUINFO_ZVE64X          (1u << 4)
 
 /* Initialized with a constructor. */
 extern unsigned cpuinfo;
+extern unsigned riscv_lg2_vlenb;
 
 /*
  * We cannot rely on constructor ordering, so other constructors must
diff --git a/util/cpuinfo-riscv.c b/util/cpuinfo-riscv.c
index 8cacc67645..971c924012 100644
--- a/util/cpuinfo-riscv.c
+++ b/util/cpuinfo-riscv.c
@@ -4,6 +4,7 @@
  */
 
 #include "qemu/osdep.h"
+#include "qemu/host-utils.h"
 #include "host/cpuinfo.h"
 
 #ifdef CONFIG_ASM_HWPROBE_H
@@ -13,6 +14,7 @@
 #endif
 
 unsigned cpuinfo;
+unsigned riscv_lg2_vlenb;
 static volatile sig_atomic_t got_sigill;
 
 static void sigill_handler(int signo, siginfo_t *si, void *data)
@@ -34,7 +36,7 @@ static void sigill_handler(int signo, siginfo_t *si, void *data)
 /* Called both as constructor and (possibly) via other constructors. */
 unsigned __attribute__((constructor)) cpuinfo_init(void)
 {
-    unsigned left = CPUINFO_ZBA | CPUINFO_ZBB | CPUINFO_ZICOND;
+    unsigned left = CPUINFO_ZBA | CPUINFO_ZBB | CPUINFO_ZICOND | CPUINFO_ZVE64X;
     unsigned info = cpuinfo;
 
     if (info) {
@@ -50,6 +52,10 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
 #endif
 #if defined(__riscv_arch_test) && defined(__riscv_zicond)
     info |= CPUINFO_ZICOND;
+#endif
+#if defined(__riscv_arch_test) && \
+    (defined(__riscv_vector) || defined(__riscv_zve64x))
+    info |= CPUINFO_ZVE64X;
 #endif
     left &= ~info;
 
@@ -69,11 +75,22 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
 #ifdef RISCV_HWPROBE_EXT_ZICOND
             info |= pair.value & RISCV_HWPROBE_EXT_ZICOND ? CPUINFO_ZICOND : 0;
             left &= ~CPUINFO_ZICOND;
+#endif
+            /* For rv64, V is Zve64d, a superset of Zve64x. */
+            info |= pair.value & RISCV_HWPROBE_IMA_V ? CPUINFO_ZVE64X : 0;
+#ifdef RISCV_HWPROBE_EXT_ZVE64X
+            info |= pair.value & RISCV_HWPROBE_EXT_ZVE64X ? CPUINFO_ZVE64X : 0;
 #endif
         }
     }
 #endif /* CONFIG_ASM_HWPROBE_H */
 
+    /*
+     * We only detect support for vectors with hwprobe.  All kernels with
+     * support for vectors in userspace also support the hwprobe syscall.
+     */
+    left &= ~CPUINFO_ZVE64X;
+
     if (left) {
         struct sigaction sa_old, sa_new;
 
@@ -113,6 +130,21 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
         assert(left == 0);
     }
 
+    if (info & CPUINFO_ZVE64X) {
+        /*
+         * We are guaranteed by RVV-1.0 that VLEN is a power of 2.
+         * We are guaranteed by Zve64x that VLEN >= 64, and that
+         * EEW of {8,16,32,64} are supported.
+         */
+        unsigned long vlenb;
+        /* csrr %0, vlenb */
+        asm volatile(".insn i 0x73, 0x2, %0, zero, -990" : "=r"(vlenb));
+        assert(vlenb >= 8);
+        assert(is_power_of_2(vlenb));
+        /* Cache VLEN in a convenient form. */
+        riscv_lg2_vlenb = ctz32(vlenb);
+    }
+
     info |= CPUINFO_ALWAYS;
     cpuinfo = info;
     return info;
-- 
2.43.0

On 10/21/24 9:11 PM, Richard Henderson wrote:
> From: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
> 
> Add support for probing RISC-V vector extension availability in
> the backend. This information will be used when deciding whether
> to use vector instructions in code generation.
> 
> Cache lg2(vlenb) for the backend. The storing of lg2(vlenb) means
> we can convert all of the division into subtraction.
> 
> While the compiler doesn't support RISCV_HWPROBE_EXT_ZVE64X,
> we use RISCV_HWPROBE_IMA_V instead. RISCV_HWPROBE_IMA_V is more
> strictly constrainted than RISCV_HWPROBE_EXT_ZVE64X. At least in
> current QEMU implemenation, the V vector extension depends on the
> zve64d extension.
> 
> Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
> Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
> Message-ID: <20241007025700.47259-2-zhiwei_liu@linux.alibaba.com>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---

Tested with a KVM guest running in an emulated RV host.


Tested-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>

>   host/include/riscv/host/cpuinfo.h |  2 ++
>   util/cpuinfo-riscv.c              | 34 ++++++++++++++++++++++++++++++-
>   2 files changed, 35 insertions(+), 1 deletion(-)
> 
> diff --git a/host/include/riscv/host/cpuinfo.h b/host/include/riscv/host/cpuinfo.h
> index 2b00660e36..cdc784e7b6 100644
> --- a/host/include/riscv/host/cpuinfo.h
> +++ b/host/include/riscv/host/cpuinfo.h
> @@ -10,9 +10,11 @@
>   #define CPUINFO_ZBA             (1u << 1)
>   #define CPUINFO_ZBB             (1u << 2)
>   #define CPUINFO_ZICOND          (1u << 3)
> +#define CPUINFO_ZVE64X          (1u << 4)
>   
>   /* Initialized with a constructor. */
>   extern unsigned cpuinfo;
> +extern unsigned riscv_lg2_vlenb;
>   
>   /*
>    * We cannot rely on constructor ordering, so other constructors must
> diff --git a/util/cpuinfo-riscv.c b/util/cpuinfo-riscv.c
> index 8cacc67645..971c924012 100644
> --- a/util/cpuinfo-riscv.c
> +++ b/util/cpuinfo-riscv.c
> @@ -4,6 +4,7 @@
>    */
>   
>   #include "qemu/osdep.h"
> +#include "qemu/host-utils.h"
>   #include "host/cpuinfo.h"
>   
>   #ifdef CONFIG_ASM_HWPROBE_H
> @@ -13,6 +14,7 @@
>   #endif
>   
>   unsigned cpuinfo;
> +unsigned riscv_lg2_vlenb;
>   static volatile sig_atomic_t got_sigill;
>   
>   static void sigill_handler(int signo, siginfo_t *si, void *data)
> @@ -34,7 +36,7 @@ static void sigill_handler(int signo, siginfo_t *si, void *data)
>   /* Called both as constructor and (possibly) via other constructors. */
>   unsigned __attribute__((constructor)) cpuinfo_init(void)
>   {
> -    unsigned left = CPUINFO_ZBA | CPUINFO_ZBB | CPUINFO_ZICOND;
> +    unsigned left = CPUINFO_ZBA | CPUINFO_ZBB | CPUINFO_ZICOND | CPUINFO_ZVE64X;
>       unsigned info = cpuinfo;
>   
>       if (info) {
> @@ -50,6 +52,10 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
>   #endif
>   #if defined(__riscv_arch_test) && defined(__riscv_zicond)
>       info |= CPUINFO_ZICOND;
> +#endif
> +#if defined(__riscv_arch_test) && \
> +    (defined(__riscv_vector) || defined(__riscv_zve64x))
> +    info |= CPUINFO_ZVE64X;
>   #endif
>       left &= ~info;
>   
> @@ -69,11 +75,22 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
>   #ifdef RISCV_HWPROBE_EXT_ZICOND
>               info |= pair.value & RISCV_HWPROBE_EXT_ZICOND ? CPUINFO_ZICOND : 0;
>               left &= ~CPUINFO_ZICOND;
> +#endif
> +            /* For rv64, V is Zve64d, a superset of Zve64x. */
> +            info |= pair.value & RISCV_HWPROBE_IMA_V ? CPUINFO_ZVE64X : 0;
> +#ifdef RISCV_HWPROBE_EXT_ZVE64X
> +            info |= pair.value & RISCV_HWPROBE_EXT_ZVE64X ? CPUINFO_ZVE64X : 0;
>   #endif
>           }
>       }
>   #endif /* CONFIG_ASM_HWPROBE_H */
>   
> +    /*
> +     * We only detect support for vectors with hwprobe.  All kernels with
> +     * support for vectors in userspace also support the hwprobe syscall.
> +     */
> +    left &= ~CPUINFO_ZVE64X;
> +
>       if (left) {
>           struct sigaction sa_old, sa_new;
>   
> @@ -113,6 +130,21 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
>           assert(left == 0);
>       }
>   
> +    if (info & CPUINFO_ZVE64X) {
> +        /*
> +         * We are guaranteed by RVV-1.0 that VLEN is a power of 2.
> +         * We are guaranteed by Zve64x that VLEN >= 64, and that
> +         * EEW of {8,16,32,64} are supported.
> +         */
> +        unsigned long vlenb;
> +        /* csrr %0, vlenb */
> +        asm volatile(".insn i 0x73, 0x2, %0, zero, -990" : "=r"(vlenb));
> +        assert(vlenb >= 8);
> +        assert(is_power_of_2(vlenb));
> +        /* Cache VLEN in a convenient form. */
> +        riscv_lg2_vlenb = ctz32(vlenb);
> +    }
> +
>       info |= CPUINFO_ALWAYS;
>       cpuinfo = info;
>       return info;