This removes the 256 byte parity table from the executable on
x86, s390 and RISC-V/zbb hosts.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 include/qemu/host-utils.h                | 16 ++++++++++++++++
 target/i386/tcg/helper-tcg.h             | 12 ++++++++++++
 target/i386/tcg/cc_helper_template.h.inc | 20 ++++++++++----------
 target/i386/tcg/cc_helper.c              |  2 ++
 target/i386/tcg/int_helper.c             |  4 ++--
 5 files changed, 42 insertions(+), 12 deletions(-)

diff --git a/include/qemu/host-utils.h b/include/qemu/host-utils.h
index ead97d354d6..bd4c684e5b5 100644
--- a/include/qemu/host-utils.h
+++ b/include/qemu/host-utils.h
@@ -126,6 +126,13 @@ static inline uint64_t muldiv64_round_up(uint64_t a, uint32_t b, uint32_t c)
 }
 #endif
 
+#if defined __POPCNT__ || defined __s390x__|| defined __riscv_zbb
+#define HAVE_FAST_CTPOP 1
+#endif
+#if defined __i386__ || defined __x86_64__ || defined HAVE_FAST_CTPOP
+#define HAVE_FAST_PARITY8 1
+#endif
+
 /**
  * clz8 - count leading zeros in a 8-bit value.
  * @val: The value to search
@@ -313,6 +320,15 @@ static inline int ctpop8(uint8_t val)
     return __builtin_popcount(val);
 }
 
+/*
+ * parity8 - return the parity (1 = odd) of an 8-bit value.
+ * @val: The value to search
+ */
+static inline int parity8(uint8_t val)
+{
+    return __builtin_parity(val);
+}
+
 /**
  * ctpop16 - count the population of one bits in a 16-bit value.
  * @val: The value to search
diff --git a/target/i386/tcg/helper-tcg.h b/target/i386/tcg/helper-tcg.h
index 15d6c6f8b4f..fb6354873d1 100644
--- a/target/i386/tcg/helper-tcg.h
+++ b/target/i386/tcg/helper-tcg.h
@@ -21,6 +21,7 @@
 #define I386_HELPER_TCG_H
 
 #include "exec/exec-all.h"
+#include "qemu/host-utils.h"
 
 /* Maximum instruction code size */
 #define TARGET_MAX_INSN_SIZE 16
@@ -87,7 +88,18 @@ G_NORETURN void x86_cpu_do_unaligned_access(CPUState *cs, vaddr vaddr,
 #endif
 
 /* cc_helper.c */
+#ifdef HAVE_FAST_PARITY8
+static inline unsigned int compute_pf(uint8_t x)
+{
+    return !parity8(x) * CC_P;
+}
+#else
 extern const uint8_t parity_table[256];
+static inline unsigned int compute_pf(uint8_t x)
+{
+    return parity_table[x];
+}
+#endif
 
 /* misc_helper.c */
 void cpu_load_eflags(CPUX86State *env, int eflags, int update_mask);
diff --git a/target/i386/tcg/cc_helper_template.h.inc b/target/i386/tcg/cc_helper_template.h.inc
index 4cbbc73c3cd..8af8b8539f9 100644
--- a/target/i386/tcg/cc_helper_template.h.inc
+++ b/target/i386/tcg/cc_helper_template.h.inc
@@ -45,7 +45,7 @@ static uint32_t glue(compute_all_add, SUFFIX)(DATA_TYPE dst, DATA_TYPE src1)
     DATA_TYPE src2 = dst - src1;
 
     cf = dst < src1;
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = (dst ^ src1 ^ src2) & CC_A;
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
@@ -65,7 +65,7 @@ static uint32_t glue(compute_all_adc, SUFFIX)(DATA_TYPE dst, DATA_TYPE src1,
     DATA_TYPE src2 = dst - src1 - src3;
 
     cf = (src3 ? dst <= src1 : dst < src1);
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = (dst ^ src1 ^ src2) & 0x10;
     zf = (dst == 0) << 6;
     sf = lshift(dst, 8 - DATA_BITS) & 0x80;
@@ -85,7 +85,7 @@ static uint32_t glue(compute_all_sub, SUFFIX)(DATA_TYPE dst, DATA_TYPE src2)
     DATA_TYPE src1 = dst + src2;
 
     cf = src1 < src2;
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = (dst ^ src1 ^ src2) & CC_A;
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
@@ -107,7 +107,7 @@ static uint32_t glue(compute_all_sbb, SUFFIX)(DATA_TYPE dst, DATA_TYPE src2,
     DATA_TYPE src1 = dst + src2 + src3;
 
     cf = (src3 ? src1 <= src2 : src1 < src2);
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = (dst ^ src1 ^ src2) & 0x10;
     zf = (dst == 0) << 6;
     sf = lshift(dst, 8 - DATA_BITS) & 0x80;
@@ -128,7 +128,7 @@ static uint32_t glue(compute_all_logic, SUFFIX)(DATA_TYPE dst, DATA_TYPE src1)
     uint32_t cf, pf, af, zf, sf, of;
 
     cf = 0;
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = 0;
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
@@ -144,7 +144,7 @@ static uint32_t glue(compute_all_inc, SUFFIX)(DATA_TYPE dst, DATA_TYPE src1)
     cf = src1;
     src1 = dst - 1;
     src2 = 1;
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = (dst ^ src1 ^ src2) & CC_A;
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
@@ -160,7 +160,7 @@ static uint32_t glue(compute_all_dec, SUFFIX)(DATA_TYPE dst, DATA_TYPE src1)
     cf = src1;
     src1 = dst + 1;
     src2 = 1;
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = (dst ^ src1 ^ src2) & CC_A;
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
@@ -173,7 +173,7 @@ static uint32_t glue(compute_all_shl, SUFFIX)(DATA_TYPE dst, DATA_TYPE src1)
     uint32_t cf, pf, af, zf, sf, of;
 
     cf = (src1 >> (DATA_BITS - 1)) & CC_C;
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = 0; /* undefined */
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
@@ -192,7 +192,7 @@ static uint32_t glue(compute_all_sar, SUFFIX)(DATA_TYPE dst, DATA_TYPE src1)
     uint32_t cf, pf, af, zf, sf, of;
 
     cf = src1 & 1;
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = 0; /* undefined */
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
@@ -209,7 +209,7 @@ static uint32_t glue(compute_all_mul, SUFFIX)(DATA_TYPE dst, target_long src1)
     uint32_t cf, pf, af, zf, sf, of;
 
     cf = (src1 != 0);
-    pf = parity_table[(uint8_t)dst];
+    pf = compute_pf(dst);
     af = 0; /* undefined */
     zf = (dst == 0) * CC_Z;
     sf = lshift(dst, 8 - DATA_BITS) & CC_S;
diff --git a/target/i386/tcg/cc_helper.c b/target/i386/tcg/cc_helper.c
index c24e6a14c07..d65345f059d 100644
--- a/target/i386/tcg/cc_helper.c
+++ b/target/i386/tcg/cc_helper.c
@@ -22,6 +22,7 @@
 #include "exec/helper-proto.h"
 #include "helper-tcg.h"
 
+#ifndef HAVE_FAST_PARITY8
 const uint8_t parity_table[256] = {
     CC_P, 0, 0, CC_P, 0, CC_P, CC_P, 0,
     0, CC_P, CC_P, 0, CC_P, 0, 0, CC_P,
@@ -56,6 +57,7 @@ const uint8_t parity_table[256] = {
     CC_P, 0, 0, CC_P, 0, CC_P, CC_P, 0,
     0, CC_P, CC_P, 0, CC_P, 0, 0, CC_P,
 };
+#endif
 
 #define SHIFT 0
 #include "cc_helper_template.h.inc"
diff --git a/target/i386/tcg/int_helper.c b/target/i386/tcg/int_helper.c
index e1f92405282..1a02e9d8434 100644
--- a/target/i386/tcg/int_helper.c
+++ b/target/i386/tcg/int_helper.c
@@ -237,7 +237,7 @@ void helper_daa(CPUX86State *env)
     env->regs[R_EAX] = (env->regs[R_EAX] & ~0xff) | al;
     /* well, speed is not an issue here, so we compute the flags by hand */
     eflags |= (al == 0) << 6; /* zf */
-    eflags |= parity_table[al]; /* pf */
+    eflags |= compute_pf(al);
     eflags |= (al & 0x80); /* sf */
     CC_SRC = eflags;
     CC_OP = CC_OP_EFLAGS;
@@ -269,7 +269,7 @@ void helper_das(CPUX86State *env)
     env->regs[R_EAX] = (env->regs[R_EAX] & ~0xff) | al;
     /* well, speed is not an issue here, so we compute the flags by hand */
     eflags |= (al == 0) << 6; /* zf */
-    eflags |= parity_table[al]; /* pf */
+    eflags |= compute_pf(al);
     eflags |= (al & 0x80); /* sf */
     CC_SRC = eflags;
     CC_OP = CC_OP_EFLAGS;
-- 
2.46.2

On 10/20/24 08:53, Paolo Bonzini wrote:
> This removes the 256 byte parity table from the executable on
> x86, s390 and RISC-V/zbb hosts.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>   include/qemu/host-utils.h                | 16 ++++++++++++++++
>   target/i386/tcg/helper-tcg.h             | 12 ++++++++++++
>   target/i386/tcg/cc_helper_template.h.inc | 20 ++++++++++----------
>   target/i386/tcg/cc_helper.c              |  2 ++
>   target/i386/tcg/int_helper.c             |  4 ++--
>   5 files changed, 42 insertions(+), 12 deletions(-)
> 
> diff --git a/include/qemu/host-utils.h b/include/qemu/host-utils.h
> index ead97d354d6..bd4c684e5b5 100644
> --- a/include/qemu/host-utils.h
> +++ b/include/qemu/host-utils.h
> @@ -126,6 +126,13 @@ static inline uint64_t muldiv64_round_up(uint64_t a, uint32_t b, uint32_t c)
>   }
>   #endif
>   
> +#if defined __POPCNT__ || defined __s390x__|| defined __riscv_zbb
> +#define HAVE_FAST_CTPOP 1
> +#endif
> +#if defined __i386__ || defined __x86_64__ || defined HAVE_FAST_CTPOP
> +#define HAVE_FAST_PARITY8 1
> +#endif

This misses out on a few, and is rather pointless; see below.

> +/*
> + * parity8 - return the parity (1 = odd) of an 8-bit value.
> + * @val: The value to search
> + */
> +static inline int parity8(uint8_t val)
> +{
> +    return __builtin_parity(val);
> +}

This should be pretty darn close to ideal for all hosts.

> +#ifdef HAVE_FAST_PARITY8
> +static inline unsigned int compute_pf(uint8_t x)
> +{
> +    return !parity8(x) * CC_P;
> +}
> +#else
>   extern const uint8_t parity_table[256];
> +static inline unsigned int compute_pf(uint8_t x)
> +{
> +    return parity_table[x];
> +}
> +#endif

All common hosts have either parity or popcount.  I think there's no point keeping the 
other path for some hosts (non-zbb riscv, older sparc64, ?).  They are just as well served 
by pulling in libgcc's implementations.


r~