diff mbox series

[v5,20/23] target/ppc: Implement cfuged instruction

Message ID 20210517205025.3777947-21-matheus.ferst@eldorado.org.br (mailing list archive)
State New, archived
Headers show
Series Base for adding PowerPC 64-bit instructions | expand

Commit Message

Matheus K. Ferst May 17, 2021, 8:50 p.m. UTC
From: Matheus Ferst <matheus.ferst@eldorado.org.br>

Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
---
v5:
- Comments to explain helper_cfuged implementation.
---
 target/ppc/helper.h                        |  1 +
 target/ppc/insn32.decode                   |  4 ++
 target/ppc/int_helper.c                    | 62 ++++++++++++++++++++++
 target/ppc/translate/fixedpoint-impl.c.inc | 12 +++++
 4 files changed, 79 insertions(+)

Comments

David Gibson May 18, 2021, 12:51 a.m. UTC | #1
On Mon, May 17, 2021 at 05:50:22PM -0300, matheus.ferst@eldorado.org.br wrote:
> From: Matheus Ferst <matheus.ferst@eldorado.org.br>
> 
> Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>

Applied to ppc-for-6.1, thanks.

> ---
> v5:
> - Comments to explain helper_cfuged implementation.
> ---
>  target/ppc/helper.h                        |  1 +
>  target/ppc/insn32.decode                   |  4 ++
>  target/ppc/int_helper.c                    | 62 ++++++++++++++++++++++
>  target/ppc/translate/fixedpoint-impl.c.inc | 12 +++++
>  4 files changed, 79 insertions(+)
> 
> diff --git a/target/ppc/helper.h b/target/ppc/helper.h
> index ea9f2a236c..c517b9f025 100644
> --- a/target/ppc/helper.h
> +++ b/target/ppc/helper.h
> @@ -46,6 +46,7 @@ DEF_HELPER_4(divwe, tl, env, tl, tl, i32)
>  DEF_HELPER_FLAGS_1(popcntb, TCG_CALL_NO_RWG_SE, tl, tl)
>  DEF_HELPER_FLAGS_2(cmpb, TCG_CALL_NO_RWG_SE, tl, tl, tl)
>  DEF_HELPER_3(sraw, tl, env, tl, tl)
> +DEF_HELPER_FLAGS_2(cfuged, TCG_CALL_NO_RWG_SE, i64, i64, i64)
>  #if defined(TARGET_PPC64)
>  DEF_HELPER_FLAGS_2(cmpeqb, TCG_CALL_NO_RWG_SE, i32, tl, tl)
>  DEF_HELPER_FLAGS_1(popcntw, TCG_CALL_NO_RWG_SE, tl, tl)
> diff --git a/target/ppc/insn32.decode b/target/ppc/insn32.decode
> index bc69c70493..d4044d9069 100644
> --- a/target/ppc/insn32.decode
> +++ b/target/ppc/insn32.decode
> @@ -87,6 +87,10 @@ STDUX           011111 ..... ..... ..... 0010110101 -   @X
>  ADDI            001110 ..... ..... ................     @D
>  ADDIS           001111 ..... ..... ................     @D
>  
> +## Fixed-Point Logical Instructions
> +
> +CFUGED          011111 ..... ..... ..... 0011011100 -   @X
> +
>  ### Move To/From System Register Instructions
>  
>  SETBC           011111 ..... ..... ----- 0110000000 -   @X_bi
> diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
> index 41f8477d4b..efa833ef64 100644
> --- a/target/ppc/int_helper.c
> +++ b/target/ppc/int_helper.c
> @@ -320,6 +320,68 @@ target_ulong helper_popcntb(target_ulong val)
>  }
>  #endif
>  
> +uint64_t helper_cfuged(uint64_t src, uint64_t mask)
> +{
> +    /*
> +     * Instead of processing the mask bit-by-bit from the most significant to
> +     * the least significant bit, as described in PowerISA, we'll handle it in
> +     * blocks of 'n' zeros/ones from LSB to MSB. To avoid the decision to use
> +     * ctz or cto, we negate the mask at the end of the loop.
> +     */
> +    target_ulong m, left = 0, right = 0;
> +    unsigned int n, i = 64;
> +    bool bit = false; /* tracks if we are processing zeros or ones */
> +
> +    if (mask == 0 || mask == -1) {
> +        return src;
> +    }
> +
> +    /* Processes the mask in blocks, from LSB to MSB */
> +    while (i) {
> +        /* Find how many bits we should take */
> +        n = ctz64(mask);
> +        if (n > i) {
> +            n = i;
> +        }
> +
> +        /*
> +         * Extracts 'n' trailing bits of src and put them on the leading 'n'
> +         * bits of 'right' or 'left', pushing down the previously extracted
> +         * values.
> +         */
> +        m = (1ll << n) - 1;
> +        if (bit) {
> +            right = ror64(right | (src & m), n);
> +        } else {
> +            left = ror64(left | (src & m), n);
> +        }
> +
> +        /*
> +         * Discards the processed bits from 'src' and 'mask'. Note that we are
> +         * removing 'n' trailing zeros from 'mask', but the logical shift will
> +         * add 'n' leading zeros back, so the population count of 'mask' is kept
> +         * the same.
> +         */
> +        src >>= n;
> +        mask >>= n;
> +        i -= n;
> +        bit = !bit;
> +        mask = ~mask;
> +    }
> +
> +    /*
> +     * At the end, right was ror'ed ctpop(mask) times. To put it back in place,
> +     * we'll shift it more 64-ctpop(mask) times.
> +     */
> +    if (bit) {
> +        n = ctpop64(mask);
> +    } else {
> +        n = 64 - ctpop64(mask);
> +    }
> +
> +    return left | (right >> n);
> +}
> +
>  /*****************************************************************************/
>  /* PowerPC 601 specific instructions (POWER bridge) */
>  target_ulong helper_div(CPUPPCState *env, target_ulong arg1, target_ulong arg2)
> diff --git a/target/ppc/translate/fixedpoint-impl.c.inc b/target/ppc/translate/fixedpoint-impl.c.inc
> index 204848d017..4038143efb 100644
> --- a/target/ppc/translate/fixedpoint-impl.c.inc
> +++ b/target/ppc/translate/fixedpoint-impl.c.inc
> @@ -227,3 +227,15 @@ TRANS(SETBC, do_set_bool_cond, false, false)
>  TRANS(SETBCR, do_set_bool_cond, false, true)
>  TRANS(SETNBC, do_set_bool_cond, true, false)
>  TRANS(SETNBCR, do_set_bool_cond, true, true)
> +
> +static bool trans_CFUGED(DisasContext *ctx, arg_X *a)
> +{
> +    REQUIRE_64BIT(ctx);
> +    REQUIRE_INSNS_FLAGS2(ctx, ISA310);
> +#if defined(TARGET_PPC64)
> +    gen_helper_cfuged(cpu_gpr[a->ra], cpu_gpr[a->rt], cpu_gpr[a->rb]);
> +#else
> +    qemu_build_not_reached();
> +#endif
> +    return true;
> +}
diff mbox series

Patch

diff --git a/target/ppc/helper.h b/target/ppc/helper.h
index ea9f2a236c..c517b9f025 100644
--- a/target/ppc/helper.h
+++ b/target/ppc/helper.h
@@ -46,6 +46,7 @@  DEF_HELPER_4(divwe, tl, env, tl, tl, i32)
 DEF_HELPER_FLAGS_1(popcntb, TCG_CALL_NO_RWG_SE, tl, tl)
 DEF_HELPER_FLAGS_2(cmpb, TCG_CALL_NO_RWG_SE, tl, tl, tl)
 DEF_HELPER_3(sraw, tl, env, tl, tl)
+DEF_HELPER_FLAGS_2(cfuged, TCG_CALL_NO_RWG_SE, i64, i64, i64)
 #if defined(TARGET_PPC64)
 DEF_HELPER_FLAGS_2(cmpeqb, TCG_CALL_NO_RWG_SE, i32, tl, tl)
 DEF_HELPER_FLAGS_1(popcntw, TCG_CALL_NO_RWG_SE, tl, tl)
diff --git a/target/ppc/insn32.decode b/target/ppc/insn32.decode
index bc69c70493..d4044d9069 100644
--- a/target/ppc/insn32.decode
+++ b/target/ppc/insn32.decode
@@ -87,6 +87,10 @@  STDUX           011111 ..... ..... ..... 0010110101 -   @X
 ADDI            001110 ..... ..... ................     @D
 ADDIS           001111 ..... ..... ................     @D
 
+## Fixed-Point Logical Instructions
+
+CFUGED          011111 ..... ..... ..... 0011011100 -   @X
+
 ### Move To/From System Register Instructions
 
 SETBC           011111 ..... ..... ----- 0110000000 -   @X_bi
diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
index 41f8477d4b..efa833ef64 100644
--- a/target/ppc/int_helper.c
+++ b/target/ppc/int_helper.c
@@ -320,6 +320,68 @@  target_ulong helper_popcntb(target_ulong val)
 }
 #endif
 
+uint64_t helper_cfuged(uint64_t src, uint64_t mask)
+{
+    /*
+     * Instead of processing the mask bit-by-bit from the most significant to
+     * the least significant bit, as described in PowerISA, we'll handle it in
+     * blocks of 'n' zeros/ones from LSB to MSB. To avoid the decision to use
+     * ctz or cto, we negate the mask at the end of the loop.
+     */
+    target_ulong m, left = 0, right = 0;
+    unsigned int n, i = 64;
+    bool bit = false; /* tracks if we are processing zeros or ones */
+
+    if (mask == 0 || mask == -1) {
+        return src;
+    }
+
+    /* Processes the mask in blocks, from LSB to MSB */
+    while (i) {
+        /* Find how many bits we should take */
+        n = ctz64(mask);
+        if (n > i) {
+            n = i;
+        }
+
+        /*
+         * Extracts 'n' trailing bits of src and put them on the leading 'n'
+         * bits of 'right' or 'left', pushing down the previously extracted
+         * values.
+         */
+        m = (1ll << n) - 1;
+        if (bit) {
+            right = ror64(right | (src & m), n);
+        } else {
+            left = ror64(left | (src & m), n);
+        }
+
+        /*
+         * Discards the processed bits from 'src' and 'mask'. Note that we are
+         * removing 'n' trailing zeros from 'mask', but the logical shift will
+         * add 'n' leading zeros back, so the population count of 'mask' is kept
+         * the same.
+         */
+        src >>= n;
+        mask >>= n;
+        i -= n;
+        bit = !bit;
+        mask = ~mask;
+    }
+
+    /*
+     * At the end, right was ror'ed ctpop(mask) times. To put it back in place,
+     * we'll shift it more 64-ctpop(mask) times.
+     */
+    if (bit) {
+        n = ctpop64(mask);
+    } else {
+        n = 64 - ctpop64(mask);
+    }
+
+    return left | (right >> n);
+}
+
 /*****************************************************************************/
 /* PowerPC 601 specific instructions (POWER bridge) */
 target_ulong helper_div(CPUPPCState *env, target_ulong arg1, target_ulong arg2)
diff --git a/target/ppc/translate/fixedpoint-impl.c.inc b/target/ppc/translate/fixedpoint-impl.c.inc
index 204848d017..4038143efb 100644
--- a/target/ppc/translate/fixedpoint-impl.c.inc
+++ b/target/ppc/translate/fixedpoint-impl.c.inc
@@ -227,3 +227,15 @@  TRANS(SETBC, do_set_bool_cond, false, false)
 TRANS(SETBCR, do_set_bool_cond, false, true)
 TRANS(SETNBC, do_set_bool_cond, true, false)
 TRANS(SETNBCR, do_set_bool_cond, true, true)
+
+static bool trans_CFUGED(DisasContext *ctx, arg_X *a)
+{
+    REQUIRE_64BIT(ctx);
+    REQUIRE_INSNS_FLAGS2(ctx, ISA310);
+#if defined(TARGET_PPC64)
+    gen_helper_cfuged(cpu_gpr[a->ra], cpu_gpr[a->rt], cpu_gpr[a->rb]);
+#else
+    qemu_build_not_reached();
+#endif
+    return true;
+}