qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: qemu-arm@nongnu.org, Peter Maydell <peter.maydell@linaro.org>
Subject: [PATCH v2 11/29] target/arm: Convert FMAXNMV, FMINNMV, FMAXV, FMINV to decodetree
Date: Mon,  9 Sep 2024 09:22:21 -0700	[thread overview]
Message-ID: <20240909162240.647173-12-richard.henderson@linaro.org> (raw)
In-Reply-To: <20240909162240.647173-1-richard.henderson@linaro.org>

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/tcg/translate-a64.c | 176 ++++++++++-----------------------
 target/arm/tcg/a64.decode      |  14 +++
 2 files changed, 67 insertions(+), 123 deletions(-)

diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c
index 593a1774d8..aec2f6a542 100644
--- a/target/arm/tcg/translate-a64.c
+++ b/target/arm/tcg/translate-a64.c
@@ -6835,6 +6835,59 @@ TRANS(UMAXV, do_int_reduction, a, false, 0, tcg_gen_umax_i64)
 TRANS(SMINV, do_int_reduction, a, false, MO_SIGN, tcg_gen_smin_i64)
 TRANS(UMINV, do_int_reduction, a, false, 0, tcg_gen_umin_i64)
 
+/*
+ * do_fp_reduction helper
+ *
+ * This mirrors the Reduce() pseudocode in the ARM ARM. It is
+ * important for correct NaN propagation that we do these
+ * operations in exactly the order specified by the pseudocode.
+ *
+ * This is a recursive function.
+ */
+static TCGv_i32 do_reduction_op(DisasContext *s, int rn, MemOp esz,
+                                int ebase, int ecount, TCGv_ptr fpst,
+                                NeonGenTwoSingleOpFn *fn)
+{
+    if (ecount == 1) {
+        TCGv_i32 tcg_elem = tcg_temp_new_i32();
+        read_vec_element_i32(s, tcg_elem, rn, ebase, esz);
+        return tcg_elem;
+    } else {
+        int half = ecount >> 1;
+        TCGv_i32 tcg_hi, tcg_lo, tcg_res;
+
+        tcg_hi = do_reduction_op(s, rn, esz, ebase + half, half, fpst, fn);
+        tcg_lo = do_reduction_op(s, rn, esz, ebase, half, fpst, fn);
+        tcg_res = tcg_temp_new_i32();
+
+        fn(tcg_res, tcg_lo, tcg_hi, fpst);
+        return tcg_res;
+    }
+}
+
+static bool do_fp_reduction(DisasContext *s, arg_qrr_e *a,
+                              NeonGenTwoSingleOpFn *fn)
+{
+    if (fp_access_check(s)) {
+        MemOp esz = a->esz;
+        int elts = (a->q ? 16 : 8) >> esz;
+        TCGv_ptr fpst = fpstatus_ptr(esz == MO_16 ? FPST_FPCR_F16 : FPST_FPCR);
+        TCGv_i32 res = do_reduction_op(s, a->rn, esz, 0, elts, fpst, fn);
+        write_fp_sreg(s, a->rd, res);
+    }
+    return true;
+}
+
+TRANS_FEAT(FMAXNMV_h, aa64_fp16, do_fp_reduction, a, gen_helper_advsimd_maxnumh)
+TRANS_FEAT(FMINNMV_h, aa64_fp16, do_fp_reduction, a, gen_helper_advsimd_minnumh)
+TRANS_FEAT(FMAXV_h, aa64_fp16, do_fp_reduction, a, gen_helper_advsimd_maxh)
+TRANS_FEAT(FMINV_h, aa64_fp16, do_fp_reduction, a, gen_helper_advsimd_minh)
+
+TRANS(FMAXNMV_s, do_fp_reduction, a, gen_helper_vfp_maxnums)
+TRANS(FMINNMV_s, do_fp_reduction, a, gen_helper_vfp_minnums)
+TRANS(FMAXV_s, do_fp_reduction, a, gen_helper_vfp_maxs)
+TRANS(FMINV_s, do_fp_reduction, a, gen_helper_vfp_mins)
+
 /* Shift a TCGv src by TCGv shift_amount, put result in dst.
  * Note that it is the caller's responsibility to ensure that the
  * shift amount is in range (ie 0..31 or 0..63) and provide the ARM
@@ -9061,128 +9114,6 @@ static void disas_data_proc_fp(DisasContext *s, uint32_t insn)
     }
 }
 
-/*
- * do_reduction_op helper
- *
- * This mirrors the Reduce() pseudocode in the ARM ARM. It is
- * important for correct NaN propagation that we do these
- * operations in exactly the order specified by the pseudocode.
- *
- * This is a recursive function.
- */
-static TCGv_i32 do_reduction_op(DisasContext *s, int fpopcode, int rn,
-                                MemOp esz, int ebase, int ecount, TCGv_ptr fpst)
-{
-    if (ecount == 1) {
-        TCGv_i32 tcg_elem = tcg_temp_new_i32();
-        read_vec_element_i32(s, tcg_elem, rn, ebase, esz);
-        return tcg_elem;
-    } else {
-        int half = ecount >> 1;
-        TCGv_i32 tcg_hi, tcg_lo, tcg_res;
-
-        tcg_hi = do_reduction_op(s, fpopcode, rn, esz,
-                                 ebase + half, half, fpst);
-        tcg_lo = do_reduction_op(s, fpopcode, rn, esz,
-                                 ebase, half, fpst);
-        tcg_res = tcg_temp_new_i32();
-
-        switch (fpopcode) {
-        case 0x0c: /* fmaxnmv half-precision */
-            gen_helper_advsimd_maxnumh(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        case 0x0f: /* fmaxv half-precision */
-            gen_helper_advsimd_maxh(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        case 0x1c: /* fminnmv half-precision */
-            gen_helper_advsimd_minnumh(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        case 0x1f: /* fminv half-precision */
-            gen_helper_advsimd_minh(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        case 0x2c: /* fmaxnmv */
-            gen_helper_vfp_maxnums(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        case 0x2f: /* fmaxv */
-            gen_helper_vfp_maxs(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        case 0x3c: /* fminnmv */
-            gen_helper_vfp_minnums(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        case 0x3f: /* fminv */
-            gen_helper_vfp_mins(tcg_res, tcg_lo, tcg_hi, fpst);
-            break;
-        default:
-            g_assert_not_reached();
-        }
-        return tcg_res;
-    }
-}
-
-/* AdvSIMD across lanes
- *   31  30  29 28       24 23  22 21       17 16    12 11 10 9    5 4    0
- * +---+---+---+-----------+------+-----------+--------+-----+------+------+
- * | 0 | Q | U | 0 1 1 1 0 | size | 1 1 0 0 0 | opcode | 1 0 |  Rn  |  Rd  |
- * +---+---+---+-----------+------+-----------+--------+-----+------+------+
- */
-static void disas_simd_across_lanes(DisasContext *s, uint32_t insn)
-{
-    int rd = extract32(insn, 0, 5);
-    int rn = extract32(insn, 5, 5);
-    int size = extract32(insn, 22, 2);
-    int opcode = extract32(insn, 12, 5);
-    bool is_q = extract32(insn, 30, 1);
-    bool is_u = extract32(insn, 29, 1);
-    bool is_min = false;
-    int elements;
-
-    switch (opcode) {
-    case 0xc: /* FMAXNMV, FMINNMV */
-    case 0xf: /* FMAXV, FMINV */
-        /* Bit 1 of size field encodes min vs max and the actual size
-         * depends on the encoding of the U bit. If not set (and FP16
-         * enabled) then we do half-precision float instead of single
-         * precision.
-         */
-        is_min = extract32(size, 1, 1);
-        if (!is_u && dc_isar_feature(aa64_fp16, s)) {
-            size = 1;
-        } else if (!is_u || !is_q || extract32(size, 0, 1)) {
-            unallocated_encoding(s);
-            return;
-        } else {
-            size = 2;
-        }
-        break;
-    default:
-    case 0x3: /* SADDLV, UADDLV */
-    case 0xa: /* SMAXV, UMAXV */
-    case 0x1a: /* SMINV, UMINV */
-    case 0x1b: /* ADDV */
-        unallocated_encoding(s);
-        return;
-    }
-
-    if (!fp_access_check(s)) {
-        return;
-    }
-
-    elements = (is_q ? 16 : 8) >> size;
-
-    {
-        /* Floating point vector reduction ops which work across 32
-         * bit (single) or 16 bit (half-precision) intermediates.
-         * Note that correct NaN propagation requires that we do these
-         * operations in exactly the order specified by the pseudocode.
-         */
-        TCGv_ptr fpst = fpstatus_ptr(size == MO_16 ? FPST_FPCR_F16 : FPST_FPCR);
-        int fpopcode = opcode | is_min << 4 | is_u << 5;
-        TCGv_i32 tcg_res = do_reduction_op(s, fpopcode, rn, size,
-                                           0, elements, fpst);
-        write_fp_sreg(s, rd, tcg_res);
-    }
-}
-
 /* AdvSIMD modified immediate
  *  31  30   29  28                 19 18 16 15   12  11  10  9     5 4    0
  * +---+---+----+---------------------+-----+-------+----+---+-------+------+
@@ -11735,7 +11666,6 @@ static void disas_simd_two_reg_misc_fp16(DisasContext *s, uint32_t insn)
 static const AArch64DecodeTable data_proc_simd[] = {
     /* pattern  ,  mask     ,  fn                        */
     { 0x0e200800, 0x9f3e0c00, disas_simd_two_reg_misc },
-    { 0x0e300800, 0x9f3e0c00, disas_simd_across_lanes },
     /* simd_mod_imm decode is a subset of simd_shift_imm, so must precede it */
     { 0x0f000400, 0x9ff80400, disas_simd_mod_imm },
     { 0x0f000400, 0x9f800400, disas_simd_shift_imm },
diff --git a/target/arm/tcg/a64.decode b/target/arm/tcg/a64.decode
index 5ab4b11781..c77f9fc987 100644
--- a/target/arm/tcg/a64.decode
+++ b/target/arm/tcg/a64.decode
@@ -54,11 +54,13 @@
 @rrx_d          ........ .. . rm:5  .... idx:1 . rn:5 rd:5  &rrx_e esz=3
 
 @rr_q1e0        ........ ........ ...... rn:5 rd:5      &qrr_e q=1 esz=0
+@rr_q1e2        ........ ........ ...... rn:5 rd:5      &qrr_e q=1 esz=2
 @r2r_q1e0       ........ ........ ...... rm:5 rd:5      &qrrr_e rn=%rd q=1 esz=0
 @rrr_q1e0       ........ ... rm:5 ...... rn:5 rd:5      &qrrr_e q=1 esz=0
 @rrr_q1e3       ........ ... rm:5 ...... rn:5 rd:5      &qrrr_e q=1 esz=3
 @rrrr_q1e3      ........ ... rm:5 . ra:5 rn:5 rd:5      &qrrrr_e q=1 esz=3
 
+@qrr_h          . q:1 ...... .. ...... ...... rn:5 rd:5  &qrr_e esz=1
 @qrr_e          . q:1 ...... esz:2 ...... ...... rn:5 rd:5  &qrr_e
 
 @qrrr_b         . q:1 ...... ... rm:5 ...... rn:5 rd:5  &qrrr_e esz=0
@@ -1166,3 +1168,15 @@ SMAXV           0.00 1110 .. 11000 01010 10 ..... .....     @qrr_e
 UMAXV           0.10 1110 .. 11000 01010 10 ..... .....     @qrr_e
 SMINV           0.00 1110 .. 11000 11010 10 ..... .....     @qrr_e
 UMINV           0.10 1110 .. 11000 11010 10 ..... .....     @qrr_e
+
+FMAXNMV_h       0.00 1110 00 11000 01100 10 ..... .....     @qrr_h
+FMAXNMV_s       0110 1110 00 11000 01100 10 ..... .....     @rr_q1e2
+
+FMINNMV_h       0.00 1110 10 11000 01100 10 ..... .....     @qrr_h
+FMINNMV_s       0110 1110 10 11000 01100 10 ..... .....     @rr_q1e2
+
+FMAXV_h         0.00 1110 00 11000 01111 10 ..... .....     @qrr_h
+FMAXV_s         0110 1110 00 11000 01111 10 ..... .....     @rr_q1e2
+
+FMINV_h         0.00 1110 10 11000 01111 10 ..... .....     @qrr_h
+FMINV_s         0110 1110 10 11000 01111 10 ..... .....     @rr_q1e2
-- 
2.43.0



  parent reply	other threads:[~2024-09-09 16:25 UTC|newest]

Thread overview: 40+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-09-09 16:22 [PATCH v2 00/29] target/arm: AdvSIMD decodetree conversion, part 4 Richard Henderson
2024-09-09 16:22 ` [PATCH v2 01/29] target/arm: Replace tcg_gen_dupi_vec with constants in gengvec.c Richard Henderson
2024-09-09 16:22 ` [PATCH v2 02/29] target/arm: Replace tcg_gen_dupi_vec with constants in translate-sve.c Richard Henderson
2024-09-09 16:22 ` [PATCH v2 03/29] target/arm: Use cmpsel in gen_ushl_vec Richard Henderson
2024-09-09 16:22 ` [PATCH v2 04/29] target/arm: Use cmpsel in gen_sshl_vec Richard Henderson
2024-09-09 16:22 ` [PATCH v2 05/29] target/arm: Use tcg_gen_extract2_i64 for EXT Richard Henderson
2024-09-09 22:22   ` Philippe Mathieu-Daudé
2024-09-09 16:22 ` [PATCH v2 06/29] target/arm: Convert EXT to decodetree Richard Henderson
2024-09-09 22:22   ` Philippe Mathieu-Daudé
2024-09-09 16:22 ` [PATCH v2 07/29] target/arm: Convert TBL, TBX " Richard Henderson
2024-09-09 22:23   ` Philippe Mathieu-Daudé
2024-09-09 16:22 ` [PATCH v2 08/29] target/arm: Convert UZP, TRN, ZIP " Richard Henderson
2024-09-09 16:22 ` [PATCH v2 09/29] target/arm: Simplify do_reduction_op Richard Henderson
2024-09-09 16:22 ` [PATCH v2 10/29] target/arm: Convert ADDV, *ADDLV, *MAXV, *MINV to decodetree Richard Henderson
2024-09-09 16:22 ` Richard Henderson [this message]
2024-09-09 16:22 ` [PATCH v2 12/29] target/arm: Convert FMOVI (scalar, immediate) " Richard Henderson
2024-09-09 22:24   ` Philippe Mathieu-Daudé
2024-09-10 12:27     ` Peter Maydell
2024-09-10 14:35       ` Philippe Mathieu-Daudé
2024-09-09 16:22 ` [PATCH v2 13/29] target/arm: Convert MOVI, FMOV, ORR, BIC (vector " Richard Henderson
2024-09-09 16:22 ` [PATCH v2 14/29] target/arm: Introduce gen_gvec_sshr, gen_gvec_ushr Richard Henderson
2024-09-09 16:22 ` [PATCH v2 15/29] target/arm: Fix whitespace near gen_srshr64_i64 Richard Henderson
2024-09-10 14:37   ` Philippe Mathieu-Daudé
2024-09-09 16:22 ` [PATCH v2 16/29] target/arm: Convert handle_vec_simd_shri to decodetree Richard Henderson
2024-09-09 16:22 ` [PATCH v2 17/29] target/arm: Convert handle_vec_simd_shli " Richard Henderson
2024-09-09 16:22 ` [PATCH v2 18/29] target/arm: Use {, s}extract in handle_vec_simd_wshli Richard Henderson
2024-09-09 16:22 ` [PATCH v2 19/29] target/arm: Convert SSHLL, USHLL to decodetree Richard Henderson
2024-09-09 16:22 ` [PATCH v2 20/29] target/arm: Push tcg_rnd into handle_shri_with_rndacc Richard Henderson
2024-09-10 14:37   ` Philippe Mathieu-Daudé
2024-09-09 16:22 ` [PATCH v2 21/29] target/arm: Split out subroutines of handle_shri_with_rndacc Richard Henderson
2024-09-09 16:22 ` [PATCH v2 22/29] target/arm: Convert SHRN, RSHRN to decodetree Richard Henderson
2024-09-09 16:22 ` [PATCH v2 23/29] target/arm: Convert handle_scalar_simd_shri " Richard Henderson
2024-09-09 16:22 ` [PATCH v2 24/29] target/arm: Convert handle_scalar_simd_shli " Richard Henderson
2024-09-09 16:22 ` [PATCH v2 25/29] target/arm: Convert VQSHL, VQSHLU to gvec Richard Henderson
2024-09-09 16:22 ` [PATCH v2 26/29] target/arm: Widen NeonGenNarrowEnvFn return to 64 bits Richard Henderson
2024-09-10 14:19   ` Peter Maydell
2024-09-09 16:22 ` [PATCH v2 27/29] target/arm: Convert SQSHL, UQSHL, SQSHLU (immediate) to decodetree Richard Henderson
2024-09-09 16:22 ` [PATCH v2 28/29] target/arm: Convert vector [US]QSHRN, [US]QRSHRN, SQSHRUN " Richard Henderson
2024-09-09 16:22 ` [PATCH v2 29/29] target/arm: Convert scalar " Richard Henderson
2024-09-10 15:30 ` [PATCH v2 00/29] target/arm: AdvSIMD decodetree conversion, part 4 Peter Maydell

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20240909162240.647173-12-richard.henderson@linaro.org \
    --to=richard.henderson@linaro.org \
    --cc=peter.maydell@linaro.org \
    --cc=qemu-arm@nongnu.org \
    --cc=qemu-devel@nongnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).