[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PULL 27/35] target/arm: Merge helper_sve_clr_* and helper_sve_movz_*
From: |
Peter Maydell |
Subject: |
[PULL 27/35] target/arm: Merge helper_sve_clr_* and helper_sve_movz_* |
Date: |
Fri, 28 Aug 2020 10:24:05 +0100 |
From: Richard Henderson <richard.henderson@linaro.org>
The existing clr functions have only one vector argument, and so
can only clear in place. The existing movz functions have two
vector arguments, and so can clear while moving. Merge them, with
a flag that controls the sense of active vs inactive elements
being cleared.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20200815013145.539409-10-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
---
target/arm/helper-sve.h | 5 ---
target/arm/sve_helper.c | 70 ++++++++------------------------------
target/arm/translate-sve.c | 53 +++++++++++------------------
3 files changed, 34 insertions(+), 94 deletions(-)
diff --git a/target/arm/helper-sve.h b/target/arm/helper-sve.h
index 63c4a087caa..4411c47120d 100644
--- a/target/arm/helper-sve.h
+++ b/target/arm/helper-sve.h
@@ -269,11 +269,6 @@ DEF_HELPER_FLAGS_3(sve_uminv_h, TCG_CALL_NO_RWG, i64, ptr,
ptr, i32)
DEF_HELPER_FLAGS_3(sve_uminv_s, TCG_CALL_NO_RWG, i64, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(sve_uminv_d, TCG_CALL_NO_RWG, i64, ptr, ptr, i32)
-DEF_HELPER_FLAGS_3(sve_clr_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
-DEF_HELPER_FLAGS_3(sve_clr_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
-DEF_HELPER_FLAGS_3(sve_clr_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
-DEF_HELPER_FLAGS_3(sve_clr_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
-
DEF_HELPER_FLAGS_4(sve_movz_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(sve_movz_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(sve_movz_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
diff --git a/target/arm/sve_helper.c b/target/arm/sve_helper.c
index 382fa82bc8a..4758d46f34d 100644
--- a/target/arm/sve_helper.c
+++ b/target/arm/sve_helper.c
@@ -956,85 +956,43 @@ uint32_t HELPER(sve_pnext)(void *vd, void *vg, uint32_t
pred_desc)
return flags;
}
-/* Store zero into every active element of Zd. We will use this for two
- * and three-operand predicated instructions for which logic dictates a
- * zero result. In particular, logical shift by element size, which is
- * otherwise undefined on the host.
- *
- * For element sizes smaller than uint64_t, we use tables to expand
- * the N bits of the controlling predicate to a byte mask, and clear
- * those bytes.
+/*
+ * Copy Zn into Zd, and store zero into inactive elements.
+ * If inv, store zeros into the active elements.
*/
-void HELPER(sve_clr_b)(void *vd, void *vg, uint32_t desc)
-{
- intptr_t i, opr_sz = simd_oprsz(desc) / 8;
- uint64_t *d = vd;
- uint8_t *pg = vg;
- for (i = 0; i < opr_sz; i += 1) {
- d[i] &= ~expand_pred_b(pg[H1(i)]);
- }
-}
-
-void HELPER(sve_clr_h)(void *vd, void *vg, uint32_t desc)
-{
- intptr_t i, opr_sz = simd_oprsz(desc) / 8;
- uint64_t *d = vd;
- uint8_t *pg = vg;
- for (i = 0; i < opr_sz; i += 1) {
- d[i] &= ~expand_pred_h(pg[H1(i)]);
- }
-}
-
-void HELPER(sve_clr_s)(void *vd, void *vg, uint32_t desc)
-{
- intptr_t i, opr_sz = simd_oprsz(desc) / 8;
- uint64_t *d = vd;
- uint8_t *pg = vg;
- for (i = 0; i < opr_sz; i += 1) {
- d[i] &= ~expand_pred_s(pg[H1(i)]);
- }
-}
-
-void HELPER(sve_clr_d)(void *vd, void *vg, uint32_t desc)
-{
- intptr_t i, opr_sz = simd_oprsz(desc) / 8;
- uint64_t *d = vd;
- uint8_t *pg = vg;
- for (i = 0; i < opr_sz; i += 1) {
- if (pg[H1(i)] & 1) {
- d[i] = 0;
- }
- }
-}
-
-/* Copy Zn into Zd, and store zero into inactive elements. */
void HELPER(sve_movz_b)(void *vd, void *vn, void *vg, uint32_t desc)
{
intptr_t i, opr_sz = simd_oprsz(desc) / 8;
+ uint64_t inv = -(uint64_t)(simd_data(desc) & 1);
uint64_t *d = vd, *n = vn;
uint8_t *pg = vg;
+
for (i = 0; i < opr_sz; i += 1) {
- d[i] = n[i] & expand_pred_b(pg[H1(i)]);
+ d[i] = n[i] & (expand_pred_b(pg[H1(i)]) ^ inv);
}
}
void HELPER(sve_movz_h)(void *vd, void *vn, void *vg, uint32_t desc)
{
intptr_t i, opr_sz = simd_oprsz(desc) / 8;
+ uint64_t inv = -(uint64_t)(simd_data(desc) & 1);
uint64_t *d = vd, *n = vn;
uint8_t *pg = vg;
+
for (i = 0; i < opr_sz; i += 1) {
- d[i] = n[i] & expand_pred_h(pg[H1(i)]);
+ d[i] = n[i] & (expand_pred_h(pg[H1(i)]) ^ inv);
}
}
void HELPER(sve_movz_s)(void *vd, void *vn, void *vg, uint32_t desc)
{
intptr_t i, opr_sz = simd_oprsz(desc) / 8;
+ uint64_t inv = -(uint64_t)(simd_data(desc) & 1);
uint64_t *d = vd, *n = vn;
uint8_t *pg = vg;
+
for (i = 0; i < opr_sz; i += 1) {
- d[i] = n[i] & expand_pred_s(pg[H1(i)]);
+ d[i] = n[i] & (expand_pred_s(pg[H1(i)]) ^ inv);
}
}
@@ -1043,8 +1001,10 @@ void HELPER(sve_movz_d)(void *vd, void *vn, void *vg,
uint32_t desc)
intptr_t i, opr_sz = simd_oprsz(desc) / 8;
uint64_t *d = vd, *n = vn;
uint8_t *pg = vg;
+ uint8_t inv = simd_data(desc);
+
for (i = 0; i < opr_sz; i += 1) {
- d[i] = n[i] & -(uint64_t)(pg[H1(i)] & 1);
+ d[i] = n[i] & -(uint64_t)((pg[H1(i)] ^ inv) & 1);
}
}
diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index 96bfd32bbec..aba156d5ab6 100644
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -590,37 +590,26 @@ static bool trans_SADDV(DisasContext *s, arg_rpr_esz *a)
*** SVE Shift by Immediate - Predicated Group
*/
-/* Store zero into every active element of Zd. We will use this for two
- * and three-operand predicated instructions for which logic dictates a
- * zero result.
+/*
+ * Copy Zn into Zd, storing zeros into inactive elements.
+ * If invert, store zeros into the active elements.
*/
-static bool do_clr_zp(DisasContext *s, int rd, int pg, int esz)
-{
- static gen_helper_gvec_2 * const fns[4] = {
- gen_helper_sve_clr_b, gen_helper_sve_clr_h,
- gen_helper_sve_clr_s, gen_helper_sve_clr_d,
- };
- if (sve_access_check(s)) {
- unsigned vsz = vec_full_reg_size(s);
- tcg_gen_gvec_2_ool(vec_full_reg_offset(s, rd),
- pred_full_reg_offset(s, pg),
- vsz, vsz, 0, fns[esz]);
- }
- return true;
-}
-
-/* Copy Zn into Zd, storing zeros into inactive elements. */
-static void do_movz_zpz(DisasContext *s, int rd, int rn, int pg, int esz)
+static bool do_movz_zpz(DisasContext *s, int rd, int rn, int pg,
+ int esz, bool invert)
{
static gen_helper_gvec_3 * const fns[4] = {
gen_helper_sve_movz_b, gen_helper_sve_movz_h,
gen_helper_sve_movz_s, gen_helper_sve_movz_d,
};
- unsigned vsz = vec_full_reg_size(s);
- tcg_gen_gvec_3_ool(vec_full_reg_offset(s, rd),
- vec_full_reg_offset(s, rn),
- pred_full_reg_offset(s, pg),
- vsz, vsz, 0, fns[esz]);
+
+ if (sve_access_check(s)) {
+ unsigned vsz = vec_full_reg_size(s);
+ tcg_gen_gvec_3_ool(vec_full_reg_offset(s, rd),
+ vec_full_reg_offset(s, rn),
+ pred_full_reg_offset(s, pg),
+ vsz, vsz, invert, fns[esz]);
+ }
+ return true;
}
static bool do_zpzi_ool(DisasContext *s, arg_rpri_esz *a,
@@ -664,7 +653,7 @@ static bool trans_LSR_zpzi(DisasContext *s, arg_rpri_esz *a)
/* Shift by element size is architecturally valid.
For logical shifts, it is a zeroing operation. */
if (a->imm >= (8 << a->esz)) {
- return do_clr_zp(s, a->rd, a->pg, a->esz);
+ return do_movz_zpz(s, a->rd, a->rd, a->pg, a->esz, true);
} else {
return do_zpzi_ool(s, a, fns[a->esz]);
}
@@ -682,7 +671,7 @@ static bool trans_LSL_zpzi(DisasContext *s, arg_rpri_esz *a)
/* Shift by element size is architecturally valid.
For logical shifts, it is a zeroing operation. */
if (a->imm >= (8 << a->esz)) {
- return do_clr_zp(s, a->rd, a->pg, a->esz);
+ return do_movz_zpz(s, a->rd, a->rd, a->pg, a->esz, true);
} else {
return do_zpzi_ool(s, a, fns[a->esz]);
}
@@ -700,7 +689,7 @@ static bool trans_ASRD(DisasContext *s, arg_rpri_esz *a)
/* Shift by element size is architecturally valid. For arithmetic
right shift for division, it is a zeroing operation. */
if (a->imm >= (8 << a->esz)) {
- return do_clr_zp(s, a->rd, a->pg, a->esz);
+ return do_movz_zpz(s, a->rd, a->rd, a->pg, a->esz, true);
} else {
return do_zpzi_ool(s, a, fns[a->esz]);
}
@@ -5049,8 +5038,7 @@ static bool trans_LD1R_zpri(DisasContext *s,
arg_rpri_load *a)
/* Zero the inactive elements. */
gen_set_label(over);
- do_movz_zpz(s, a->rd, a->rd, a->pg, esz);
- return true;
+ return do_movz_zpz(s, a->rd, a->rd, a->pg, esz, false);
}
static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr,
@@ -5833,8 +5821,5 @@ static bool trans_MOVPRFX_m(DisasContext *s, arg_rpr_esz
*a)
static bool trans_MOVPRFX_z(DisasContext *s, arg_rpr_esz *a)
{
- if (sve_access_check(s)) {
- do_movz_zpz(s, a->rd, a->rn, a->pg, a->esz);
- }
- return true;
+ return do_movz_zpz(s, a->rd, a->rn, a->pg, a->esz, false);
}
--
2.20.1
- [PULL 18/35] armsse: Define ARMSSEClass correctly, (continued)
- [PULL 18/35] armsse: Define ARMSSEClass correctly, Peter Maydell, 2020/08/28
- [PULL 19/35] qemu/int128: Add int128_lshift, Peter Maydell, 2020/08/28
- [PULL 21/35] target/arm: Split out gen_gvec_fn_zzz, do_zzz_fn, Peter Maydell, 2020/08/28
- [PULL 22/35] target/arm: Rearrange {sve,fp}_check_access assert, Peter Maydell, 2020/08/28
- [PULL 20/35] target/arm: Split out gen_gvec_fn_zz, Peter Maydell, 2020/08/28
- [PULL 23/35] target/arm: Merge do_vector2_p into do_mov_p, Peter Maydell, 2020/08/28
- [PULL 24/35] target/arm: Clean up 4-operand predicate expansion, Peter Maydell, 2020/08/28
- [PULL 26/35] target/arm: Split out gen_gvec_ool_zzzp, Peter Maydell, 2020/08/28
- [PULL 25/35] target/arm: Use tcg_gen_gvec_bitsel for trans_SEL_pppp, Peter Maydell, 2020/08/28
- [PULL 28/35] target/arm: Split out gen_gvec_ool_zzp, Peter Maydell, 2020/08/28
- [PULL 27/35] target/arm: Merge helper_sve_clr_* and helper_sve_movz_*,
Peter Maydell <=
- [PULL 29/35] target/arm: Split out gen_gvec_ool_zzz, Peter Maydell, 2020/08/28
- [PULL 30/35] target/arm: Split out gen_gvec_ool_zz, Peter Maydell, 2020/08/28
- [PULL 32/35] target/arm: Generalize inl_qrdmlah_* helper functions, Peter Maydell, 2020/08/28
- [PULL 31/35] target/arm: Tidy SVE tszimm shift formats, Peter Maydell, 2020/08/28
- [PULL 33/35] target/arm: Convert integer multiply (indexed) to gvec for aa64 advsimd, Peter Maydell, 2020/08/28
- [PULL 34/35] target/arm: Convert integer multiply-add (indexed) to gvec for aa64 advsimd, Peter Maydell, 2020/08/28
- [PULL 35/35] target/arm: Convert sq{, r}dmulh to gvec for aa64 advsimd, Peter Maydell, 2020/08/28
- Re: [PULL 00/35] target-arm queue, Peter Maydell, 2020/08/28