All of lore.kernel.org
 help / color / mirror / Atom feed
From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: peter.maydell@linaro.org
Subject: [PATCH 01/11] target/arm: Introduce neon_full_reg_offset
Date: Tue, 27 Oct 2020 20:26:53 -0700	[thread overview]
Message-ID: <20201028032703.201526-2-richard.henderson@linaro.org> (raw)
In-Reply-To: <20201028032703.201526-1-richard.henderson@linaro.org>

This function makes it clear that we're talking about the whole
register, and not the 32-bit piece at index 0.  This fixes a bug
when running on a big-endian host.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/translate.c          |  8 ++++++
 target/arm/translate-neon.c.inc | 44 ++++++++++++++++-----------------
 target/arm/translate-vfp.c.inc  |  2 +-
 3 files changed, 31 insertions(+), 23 deletions(-)

diff --git a/target/arm/translate.c b/target/arm/translate.c
index 38371db540..1b61e50f9c 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -1094,6 +1094,14 @@ static inline void gen_hlt(DisasContext *s, int imm)
     unallocated_encoding(s);
 }
 
+/*
+ * Return the offset of a "full" NEON Dreg.
+ */
+static long neon_full_reg_offset(unsigned reg)
+{
+    return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]);
+}
+
 static inline long vfp_reg_offset(bool dp, unsigned reg)
 {
     if (dp) {
diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc
index 4d1a292981..e259e24c05 100644
--- a/target/arm/translate-neon.c.inc
+++ b/target/arm/translate-neon.c.inc
@@ -76,7 +76,7 @@ neon_element_offset(int reg, int element, MemOp size)
         ofs ^= 8 - element_size;
     }
 #endif
-    return neon_reg_offset(reg, 0) + ofs;
+    return neon_full_reg_offset(reg) + ofs;
 }
 
 static void neon_load_element(TCGv_i32 var, int reg, int ele, MemOp mop)
@@ -585,12 +585,12 @@ static bool trans_VLD_all_lanes(DisasContext *s, arg_VLD_all_lanes *a)
              * We cannot write 16 bytes at once because the
              * destination is unaligned.
              */
-            tcg_gen_gvec_dup_i32(size, neon_reg_offset(vd, 0),
+            tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd),
                                  8, 8, tmp);
-            tcg_gen_gvec_mov(0, neon_reg_offset(vd + 1, 0),
-                             neon_reg_offset(vd, 0), 8, 8);
+            tcg_gen_gvec_mov(0, neon_full_reg_offset(vd + 1),
+                             neon_full_reg_offset(vd), 8, 8);
         } else {
-            tcg_gen_gvec_dup_i32(size, neon_reg_offset(vd, 0),
+            tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd),
                                  vec_size, vec_size, tmp);
         }
         tcg_gen_addi_i32(addr, addr, 1 << size);
@@ -691,9 +691,9 @@ static bool trans_VLDST_single(DisasContext *s, arg_VLDST_single *a)
 static bool do_3same(DisasContext *s, arg_3same *a, GVecGen3Fn fn)
 {
     int vec_size = a->q ? 16 : 8;
-    int rd_ofs = neon_reg_offset(a->vd, 0);
-    int rn_ofs = neon_reg_offset(a->vn, 0);
-    int rm_ofs = neon_reg_offset(a->vm, 0);
+    int rd_ofs = neon_full_reg_offset(a->vd);
+    int rn_ofs = neon_full_reg_offset(a->vn);
+    int rm_ofs = neon_full_reg_offset(a->vm);
 
     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
         return false;
@@ -1177,8 +1177,8 @@ static bool do_vector_2sh(DisasContext *s, arg_2reg_shift *a, GVecGen2iFn *fn)
 {
     /* Handle a 2-reg-shift insn which can be vectorized. */
     int vec_size = a->q ? 16 : 8;
-    int rd_ofs = neon_reg_offset(a->vd, 0);
-    int rm_ofs = neon_reg_offset(a->vm, 0);
+    int rd_ofs = neon_full_reg_offset(a->vd);
+    int rm_ofs = neon_full_reg_offset(a->vm);
 
     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
         return false;
@@ -1620,8 +1620,8 @@ static bool do_fp_2sh(DisasContext *s, arg_2reg_shift *a,
 {
     /* FP operations in 2-reg-and-shift group */
     int vec_size = a->q ? 16 : 8;
-    int rd_ofs = neon_reg_offset(a->vd, 0);
-    int rm_ofs = neon_reg_offset(a->vm, 0);
+    int rd_ofs = neon_full_reg_offset(a->vd);
+    int rm_ofs = neon_full_reg_offset(a->vm);
     TCGv_ptr fpst;
 
     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
@@ -1756,7 +1756,7 @@ static bool do_1reg_imm(DisasContext *s, arg_1reg_imm *a,
         return true;
     }
 
-    reg_ofs = neon_reg_offset(a->vd, 0);
+    reg_ofs = neon_full_reg_offset(a->vd);
     vec_size = a->q ? 16 : 8;
     imm = asimd_imm_const(a->imm, a->cmode, a->op);
 
@@ -2300,9 +2300,9 @@ static bool trans_VMULL_P_3d(DisasContext *s, arg_3diff *a)
         return true;
     }
 
-    tcg_gen_gvec_3_ool(neon_reg_offset(a->vd, 0),
-                       neon_reg_offset(a->vn, 0),
-                       neon_reg_offset(a->vm, 0),
+    tcg_gen_gvec_3_ool(neon_full_reg_offset(a->vd),
+                       neon_full_reg_offset(a->vn),
+                       neon_full_reg_offset(a->vm),
                        16, 16, 0, fn_gvec);
     return true;
 }
@@ -2445,8 +2445,8 @@ static bool do_2scalar_fp_vec(DisasContext *s, arg_2scalar *a,
 {
     /* Two registers and a scalar, using gvec */
     int vec_size = a->q ? 16 : 8;
-    int rd_ofs = neon_reg_offset(a->vd, 0);
-    int rn_ofs = neon_reg_offset(a->vn, 0);
+    int rd_ofs = neon_full_reg_offset(a->vd);
+    int rn_ofs = neon_full_reg_offset(a->vn);
     int rm_ofs;
     int idx;
     TCGv_ptr fpstatus;
@@ -2477,7 +2477,7 @@ static bool do_2scalar_fp_vec(DisasContext *s, arg_2scalar *a,
     /* a->vm is M:Vm, which encodes both register and index */
     idx = extract32(a->vm, a->size + 2, 2);
     a->vm = extract32(a->vm, 0, a->size + 2);
-    rm_ofs = neon_reg_offset(a->vm, 0);
+    rm_ofs = neon_full_reg_offset(a->vm);
 
     fpstatus = fpstatus_ptr(a->size == 1 ? FPST_STD_F16 : FPST_STD);
     tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, fpstatus,
@@ -2923,7 +2923,7 @@ static bool trans_VDUP_scalar(DisasContext *s, arg_VDUP_scalar *a)
         return true;
     }
 
-    tcg_gen_gvec_dup_mem(a->size, neon_reg_offset(a->vd, 0),
+    tcg_gen_gvec_dup_mem(a->size, neon_full_reg_offset(a->vd),
                          neon_element_offset(a->vm, a->index, a->size),
                          a->q ? 16 : 8, a->q ? 16 : 8);
     return true;
@@ -3412,8 +3412,8 @@ static bool trans_VCVT_F32_F16(DisasContext *s, arg_2misc *a)
 static bool do_2misc_vec(DisasContext *s, arg_2misc *a, GVecGen2Fn *fn)
 {
     int vec_size = a->q ? 16 : 8;
-    int rd_ofs = neon_reg_offset(a->vd, 0);
-    int rm_ofs = neon_reg_offset(a->vm, 0);
+    int rd_ofs = neon_full_reg_offset(a->vd);
+    int rm_ofs = neon_full_reg_offset(a->vm);
 
     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
         return false;
diff --git a/target/arm/translate-vfp.c.inc b/target/arm/translate-vfp.c.inc
index a7ed9bc81b..368bae0a73 100644
--- a/target/arm/translate-vfp.c.inc
+++ b/target/arm/translate-vfp.c.inc
@@ -653,7 +653,7 @@ static bool trans_VDUP(DisasContext *s, arg_VDUP *a)
     }
 
     tmp = load_reg(s, a->rt);
-    tcg_gen_gvec_dup_i32(size, neon_reg_offset(a->vn, 0),
+    tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(a->vn),
                          vec_size, vec_size, tmp);
     tcg_temp_free_i32(tmp);
 
-- 
2.25.1



  reply	other threads:[~2020-10-28  3:31 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-10-28  3:26 [PATCH 00/11] target/arm: Fix neon reg offsets Richard Henderson
2020-10-28  3:26 ` Richard Henderson [this message]
2020-10-28  3:26 ` [PATCH 02/11] target/arm: Move neon_element_offset to translate.c Richard Henderson
2020-10-28  3:26 ` [PATCH 03/11] target/arm: Use neon_element_offset in neon_load/store_reg Richard Henderson
2020-10-28  3:26 ` [PATCH 04/11] target/arm: Use neon_element_offset in vfp_reg_offset Richard Henderson
2020-10-28  3:26 ` [PATCH 05/11] target/arm: Add read/write_neon_element32 Richard Henderson
2020-10-28 20:22   ` Richard Henderson
2020-10-29 10:15     ` Peter Maydell
2020-10-28  3:26 ` [PATCH 06/11] target/arm: Expand read/write_neon_element32 to all MemOp Richard Henderson
2020-10-28  3:26 ` [PATCH 07/11] target/arm: Rename neon_load_reg32 to vfp_load_reg32 Richard Henderson
2020-10-28  3:27 ` [PATCH 08/11] target/arm: Add read/write_neon_element64 Richard Henderson
2020-10-28  3:27 ` [PATCH 09/11] target/arm: Rename neon_load_reg64 to vfp_load_reg64 Richard Henderson
2020-10-28  3:27 ` [PATCH 10/11] target/arm: Simplify do_long_3d and do_2scalar_long Richard Henderson
2020-10-28  3:27 ` [PATCH 11/11] target/arm: Improve do_prewiden_3d Richard Henderson
2020-10-28 16:48 ` [PATCH 00/11] target/arm: Fix neon reg offsets Peter Maydell
2020-10-28 18:31   ` Peter Maydell
2020-10-28 19:32   ` Richard Henderson
2020-10-28 20:03 ` Peter Maydell
2020-10-29 11:04   ` 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=20201028032703.201526-2-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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.