qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: peter.maydell@linaro.org
Subject: [PULL 22/24] tcg/ppc: Use tcg_constant_vec with tcg vec expanders
Date: Wed, 13 Jan 2021 16:16:52 -1000	[thread overview]
Message-ID: <20210114021654.647242-23-richard.henderson@linaro.org> (raw)
In-Reply-To: <20210114021654.647242-1-richard.henderson@linaro.org>

Improve expand_vec_shi to use sign-extraction for MO_32.
This allows a single VSPLTISB instruction to load all of
the valid shift constants.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/ppc/tcg-target.c.inc | 44 ++++++++++++++++++++++++----------------
 1 file changed, 27 insertions(+), 17 deletions(-)

diff --git a/tcg/ppc/tcg-target.c.inc b/tcg/ppc/tcg-target.c.inc
index 1fbb1b6db0..cf64892295 100644
--- a/tcg/ppc/tcg-target.c.inc
+++ b/tcg/ppc/tcg-target.c.inc
@@ -3336,13 +3336,22 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
 static void expand_vec_shi(TCGType type, unsigned vece, TCGv_vec v0,
                            TCGv_vec v1, TCGArg imm, TCGOpcode opci)
 {
-    TCGv_vec t1 = tcg_temp_new_vec(type);
+    TCGv_vec t1;
 
-    /* Splat w/bytes for xxspltib.  */
-    tcg_gen_dupi_vec(MO_8, t1, imm & ((8 << vece) - 1));
+    if (vece == MO_32) {
+        /*
+         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
+         * So using negative numbers gets us the 4th bit easily.
+         */
+        imm = sextract32(imm, 0, 5);
+    } else {
+        imm &= (8 << vece) - 1;
+    }
+
+    /* Splat w/bytes for xxspltib when 2.07 allows MO_64. */
+    t1 = tcg_constant_vec(type, MO_8, imm);
     vec_gen_3(opci, type, vece, tcgv_vec_arg(v0),
               tcgv_vec_arg(v1), tcgv_vec_arg(t1));
-    tcg_temp_free_vec(t1);
 }
 
 static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
@@ -3400,7 +3409,7 @@ static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
 {
     TCGv_vec t1 = tcg_temp_new_vec(type);
     TCGv_vec t2 = tcg_temp_new_vec(type);
-    TCGv_vec t3, t4;
+    TCGv_vec c0, c16;
 
     switch (vece) {
     case MO_8:
@@ -3419,21 +3428,22 @@ static void expand_vec_mul(TCGType type, unsigned vece, TCGv_vec v0,
 
     case MO_32:
         tcg_debug_assert(!have_isa_2_07);
-        t3 = tcg_temp_new_vec(type);
-        t4 = tcg_temp_new_vec(type);
-        tcg_gen_dupi_vec(MO_8, t4, -16);
+        /*
+         * Only 5 bits are significant, and VSPLTISB can represent -16..15.
+         * So using -16 is a quick way to represent 16.
+         */
+        c16 = tcg_constant_vec(type, MO_8, -16);
+        c0 = tcg_constant_vec(type, MO_8, 0);
+
         vec_gen_3(INDEX_op_rotlv_vec, type, MO_32, tcgv_vec_arg(t1),
-                  tcgv_vec_arg(v2), tcgv_vec_arg(t4));
+                  tcgv_vec_arg(v2), tcgv_vec_arg(c16));
         vec_gen_3(INDEX_op_ppc_mulou_vec, type, MO_16, tcgv_vec_arg(t2),
                   tcgv_vec_arg(v1), tcgv_vec_arg(v2));
-        tcg_gen_dupi_vec(MO_8, t3, 0);
-        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t3),
-                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(t3));
-        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t3),
-                  tcgv_vec_arg(t3), tcgv_vec_arg(t4));
-        tcg_gen_add_vec(MO_32, v0, t2, t3);
-        tcg_temp_free_vec(t3);
-        tcg_temp_free_vec(t4);
+        vec_gen_4(INDEX_op_ppc_msum_vec, type, MO_16, tcgv_vec_arg(t1),
+                  tcgv_vec_arg(v1), tcgv_vec_arg(t1), tcgv_vec_arg(c0));
+        vec_gen_3(INDEX_op_shlv_vec, type, MO_32, tcgv_vec_arg(t1),
+                  tcgv_vec_arg(t1), tcgv_vec_arg(c16));
+        tcg_gen_add_vec(MO_32, v0, t1, t2);
         break;
 
     default:
-- 
2.25.1



  parent reply	other threads:[~2021-01-14  2:37 UTC|newest]

Thread overview: 50+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-01-14  2:16 [PULL 00/24] tcg patch queue Richard Henderson
2021-01-14  2:16 ` [PULL 01/24] tcg: Use tcg_out_dupi_vec from temp_load Richard Henderson
2021-01-14  2:16 ` [PULL 02/24] tcg: Increase tcg_out_dupi_vec immediate to int64_t Richard Henderson
2021-01-14  2:16 ` [PULL 03/24] tcg: Consolidate 3 bits into enum TCGTempKind Richard Henderson
2021-01-14  2:16 ` [PULL 04/24] tcg: Add temp_readonly Richard Henderson
2021-01-14  2:16 ` [PULL 05/24] tcg: Expand TCGTemp.val to 64-bits Richard Henderson
2021-01-14  2:16 ` [PULL 06/24] tcg: Rename struct tcg_temp_info to TempOptInfo Richard Henderson
2021-01-14  2:16 ` [PULL 07/24] tcg: Expand TempOptInfo to 64-bits Richard Henderson
2021-01-14  2:16 ` [PULL 08/24] tcg: Introduce TYPE_CONST temporaries Richard Henderson
2021-01-14  2:16 ` [PULL 09/24] tcg/optimize: Improve find_better_copy Richard Henderson
2021-01-14  2:16 ` [PULL 10/24] tcg/optimize: Adjust TempOptInfo allocation Richard Henderson
2021-01-14  2:16 ` [PULL 11/24] tcg/optimize: Use tcg_constant_internal with constant folding Richard Henderson
2021-01-15 23:03   ` Alistair Francis
2021-01-16 17:24     ` Richard Henderson
2021-01-18 20:17       ` Laurent Vivier
2021-01-18 21:03         ` Richard Henderson
2021-02-01 20:45   ` Richard W.M. Jones
2021-02-01 20:53     ` Richard W.M. Jones
2021-02-04  2:22     ` Richard Henderson
2021-02-04  6:41       ` David Hildenbrand
2021-02-04  7:55         ` David Hildenbrand
2021-02-04  8:38           ` David Hildenbrand
2021-02-04  9:03             ` Richard W.M. Jones
2021-02-04  9:12               ` David Hildenbrand
2021-02-04  9:29               ` Richard W.M. Jones
2021-02-04  9:30                 ` Richard W.M. Jones
2021-02-04  9:37                 ` David Hildenbrand
2021-02-04 16:04                   ` Philippe Mathieu-Daudé
2021-02-04 16:08                     ` Philippe Mathieu-Daudé
2021-02-04 16:12                     ` David Hildenbrand
2021-02-04 16:29                     ` David Hildenbrand
2021-02-04 16:57                       ` Eric Blake
2021-02-04 17:33                       ` Richard Henderson
2021-02-04 16:48                     ` Richard W.M. Jones
2021-02-04 16:52                     ` Eric Blake
2021-01-14  2:16 ` [PULL 12/24] tcg: Convert tcg_gen_dupi_vec to TCG_CONST Richard Henderson
2021-01-14  2:16 ` [PULL 13/24] tcg: Use tcg_constant_i32 with icount expander Richard Henderson
2021-01-14  2:16 ` [PULL 14/24] tcg: Use tcg_constant_{i32,i64} with tcg int expanders Richard Henderson
2021-02-10 10:38   ` Alex Bennée
2021-01-14  2:16 ` [PULL 15/24] tcg: Use tcg_constant_{i32,i64} with tcg plugins Richard Henderson
2021-01-14  2:16 ` [PULL 16/24] tcg: Use tcg_constant_{i32,i64,vec} with gvec expanders Richard Henderson
2021-01-14  2:16 ` [PULL 17/24] tcg/tci: Add special tci_movi_{i32,i64} opcodes Richard Henderson
2021-01-14  2:16 ` [PULL 18/24] tcg: Remove movi and dupi opcodes Richard Henderson
2021-01-14  2:16 ` [PULL 19/24] tcg: Add tcg_reg_alloc_dup2 Richard Henderson
2021-01-14  2:16 ` [PULL 20/24] tcg/i386: Use tcg_constant_vec with tcg vec expanders Richard Henderson
2021-01-14  2:16 ` [PULL 21/24] tcg: Remove tcg_gen_dup{8,16,32,64}i_vec Richard Henderson
2021-01-14  2:16 ` Richard Henderson [this message]
2021-01-14  2:16 ` [PULL 23/24] tcg/aarch64: Use tcg_constant_vec with tcg vec expanders Richard Henderson
2021-01-14  2:16 ` [PULL 24/24] decodetree: Open files with encoding='utf-8' Richard Henderson
2021-01-14 13:13 ` [PULL 00/24] tcg patch queue 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=20210114021654.647242-23-richard.henderson@linaro.org \
    --to=richard.henderson@linaro.org \
    --cc=peter.maydell@linaro.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).