From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: rajav@quicinc.com, qemu-arm@nongnu.org, apazos@quicinc.com
Subject: [PATCH 29/31] target/arm: Vectorize SABD/UABD
Date: Thu, 26 Mar 2020 16:08:36 -0700 [thread overview]
Message-ID: <20200326230838.31112-30-richard.henderson@linaro.org> (raw)
In-Reply-To: <20200326230838.31112-1-richard.henderson@linaro.org>
Include 64-bit element size in preparation for SVE.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/helper.h | 10 +++
target/arm/translate.h | 5 ++
target/arm/translate-a64.c | 8 ++-
target/arm/translate.c | 133 ++++++++++++++++++++++++++++++++++++-
target/arm/vec_helper.c | 88 ++++++++++++++++++++++++
5 files changed, 240 insertions(+), 4 deletions(-)
diff --git a/target/arm/helper.h b/target/arm/helper.h
index 5ef7bb158f..97ccbd70c6 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -748,6 +748,16 @@ DEF_HELPER_FLAGS_3(gvec_sli_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_sli_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_sli_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sabd_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sabd_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sabd_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sabd_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(gvec_uabd_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_uabd_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_uabd_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_uabd_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
#ifdef TARGET_AARCH64
#include "helper-a64.h"
#include "helper-sve.h"
diff --git a/target/arm/translate.h b/target/arm/translate.h
index 843ecc1472..c453aa1c47 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -311,6 +311,11 @@ void arm_gen_gvec_sri(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
void arm_gen_gvec_sli(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void arm_gen_gvec_sabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+ uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz);
+void arm_gen_gvec_uabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+ uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz);
+
/*
* Forward to the isar_feature_* tests given a DisasContext pointer.
*/
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index fc156a217a..1791c26a39 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -12159,6 +12159,13 @@ static void disas_simd_3same_int(DisasContext *s, uint32_t insn)
gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_smin, size);
}
return;
+ case 0xe: /* SABD, UABD */
+ if (u) {
+ gen_gvec_fn3(s, is_q, rd, rn, rm, arm_gen_gvec_uabd, size);
+ } else {
+ gen_gvec_fn3(s, is_q, rd, rn, rm, arm_gen_gvec_sabd, size);
+ }
+ return;
case 0x10: /* ADD, SUB */
if (u) {
gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_sub, size);
@@ -12291,7 +12298,6 @@ static void disas_simd_3same_int(DisasContext *s, uint32_t insn)
genenvfn = fns[size][u];
break;
}
- case 0xe: /* SABD, UABD */
case 0xf: /* SABA, UABA */
{
static NeonGenTwoOpFn * const fns[3][2] = {
diff --git a/target/arm/translate.c b/target/arm/translate.c
index bb6db53598..a29868976a 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -4849,6 +4849,126 @@ const GVecGen4 sqsub_op[4] = {
.vece = MO_64 },
};
+static void gen_sabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ tcg_gen_sub_i32(t, a, b);
+ tcg_gen_sub_i32(d, b, a);
+ tcg_gen_movcond_i32(TCG_COND_LT, d, a, b, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_sabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_sub_i64(t, a, b);
+ tcg_gen_sub_i64(d, b, a);
+ tcg_gen_movcond_i64(TCG_COND_LT, d, a, b, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_sabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+ tcg_gen_smin_vec(vece, t, a, b);
+ tcg_gen_smax_vec(vece, d, a, b);
+ tcg_gen_sub_vec(vece, d, d, t);
+ tcg_temp_free_vec(t);
+}
+
+void arm_gen_gvec_sabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+ uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_sub_vec, INDEX_op_smin_vec, INDEX_op_smax_vec, 0
+ };
+ static const GVecGen3 ops[4] = {
+ { .fniv = gen_sabd_vec,
+ .fno = gen_helper_gvec_sabd_b,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fniv = gen_sabd_vec,
+ .fno = gen_helper_gvec_sabd_h,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_sabd_i32,
+ .fniv = gen_sabd_vec,
+ .fno = gen_helper_gvec_sabd_s,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_sabd_i64,
+ .fniv = gen_sabd_vec,
+ .fno = gen_helper_gvec_sabd_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
+ tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
+}
+
+static void gen_uabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ tcg_gen_sub_i32(t, a, b);
+ tcg_gen_sub_i32(d, b, a);
+ tcg_gen_movcond_i32(TCG_COND_LTU, d, a, b, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_uabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_sub_i64(t, a, b);
+ tcg_gen_sub_i64(d, b, a);
+ tcg_gen_movcond_i64(TCG_COND_LTU, d, a, b, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_uabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+ tcg_gen_umin_vec(vece, t, a, b);
+ tcg_gen_umax_vec(vece, d, a, b);
+ tcg_gen_sub_vec(vece, d, d, t);
+ tcg_temp_free_vec(t);
+}
+
+void arm_gen_gvec_uabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
+ uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_sub_vec, INDEX_op_umin_vec, INDEX_op_umax_vec, 0
+ };
+ static const GVecGen3 ops[4] = {
+ { .fniv = gen_uabd_vec,
+ .fno = gen_helper_gvec_uabd_b,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fniv = gen_uabd_vec,
+ .fno = gen_helper_gvec_uabd_h,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_uabd_i32,
+ .fniv = gen_uabd_vec,
+ .fno = gen_helper_gvec_uabd_s,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_uabd_i64,
+ .fniv = gen_uabd_vec,
+ .fno = gen_helper_gvec_uabd_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
+ tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
+}
+
/* Translate a NEON data processing instruction. Return nonzero if the
instruction is invalid.
We process data in a mixture of 32-bit and 64-bit chunks.
@@ -5107,6 +5227,16 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
tcg_gen_gvec_3(rd_ofs, rm_ofs, rn_ofs, vec_size, vec_size,
u ? &ushl_op[size] : &sshl_op[size]);
return 0;
+
+ case NEON_3R_VABD:
+ if (u) {
+ arm_gen_gvec_uabd(size, rd_ofs, rn_ofs, rm_ofs,
+ vec_size, vec_size);
+ } else {
+ arm_gen_gvec_sabd(size, rd_ofs, rn_ofs, rm_ofs,
+ vec_size, vec_size);
+ }
+ return 0;
}
if (size == 3) {
@@ -5237,9 +5367,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
case NEON_3R_VQRSHL:
GEN_NEON_INTEGER_OP_ENV(qrshl);
break;
- case NEON_3R_VABD:
- GEN_NEON_INTEGER_OP(abd);
- break;
case NEON_3R_VABA:
GEN_NEON_INTEGER_OP(abd);
tcg_temp_free_i32(tmp2);
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index 27035a8a42..e0694c16f4 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -1492,3 +1492,91 @@ void HELPER(gvec_umulh_d)(void *vd, void *vn, void *vm, uint32_t desc)
}
clear_tail(d, opr_sz, simd_maxsz(desc));
}
+
+void HELPER(gvec_sabd_b)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ int8_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
+
+void HELPER(gvec_sabd_h)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ int16_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz / 2; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
+
+void HELPER(gvec_sabd_s)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ int32_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz / 4; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
+
+void HELPER(gvec_sabd_d)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ int64_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz / 8; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
+
+void HELPER(gvec_uabd_b)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ uint8_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
+
+void HELPER(gvec_uabd_h)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ uint16_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz / 2; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
+
+void HELPER(gvec_uabd_s)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ uint32_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz / 4; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
+
+void HELPER(gvec_uabd_d)(void *vd, void *vn, void *vm, uint32_t desc)
+{
+ intptr_t i, opr_sz = simd_oprsz(desc);
+ uint64_t *d = vd, *n = vn, *m = vm;
+
+ for (i = 0; i < opr_sz / 8; ++i) {
+ d[i] = n[i] < m[i] ? m[i] - n[i] : n[i] - m[i];
+ }
+ clear_tail(d, opr_sz, simd_maxsz(desc));
+}
--
2.20.1
next prev parent reply other threads:[~2020-03-26 23:29 UTC|newest]
Thread overview: 44+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-03-26 23:08 [PATCH for-5.1 00/31] target/arm: SVE2, part 1 Richard Henderson
2020-03-26 23:08 ` [PATCH 01/31] target/arm: Add ID_AA64ZFR0 fields and isar_feature_aa64_sve2 Richard Henderson
2020-04-11 18:45 ` Alex Bennée
2020-03-26 23:08 ` [PATCH 02/31] target/arm: Implement SVE2 Integer Multiply - Unpredicated Richard Henderson
2020-03-26 23:08 ` [PATCH 03/31] target/arm: Implement SVE2 integer pairwise add and accumulate long Richard Henderson
2020-03-26 23:08 ` [PATCH 04/31] target/arm: Remove fp_status from helper_{recpe, rsqrte}_u32 Richard Henderson
2020-03-26 23:08 ` [PATCH 05/31] target/arm: Implement SVE2 integer unary operations (predicated) Richard Henderson
2020-03-26 23:08 ` [PATCH 06/31] target/arm: Split out saturating/rounding shifts from neon Richard Henderson
2020-03-26 23:08 ` [PATCH 07/31] target/arm: Implement SVE2 saturating/rounding bitwise shift left (predicated) Richard Henderson
2020-03-26 23:08 ` [PATCH 08/31] target/arm: Implement SVE2 integer halving add/subtract (predicated) Richard Henderson
2020-03-26 23:08 ` [PATCH 09/31] target/arm: Implement SVE2 integer pairwise arithmetic Richard Henderson
2020-04-13 16:02 ` Laurent Desnogues
2020-03-26 23:08 ` [PATCH 10/31] target/arm: Implement SVE2 saturating add/subtract (predicated) Richard Henderson
2020-03-26 23:08 ` [PATCH 11/31] target/arm: Implement SVE2 integer add/subtract long Richard Henderson
2020-04-13 16:09 ` Laurent Desnogues
2020-03-26 23:08 ` [PATCH 12/31] target/arm: Implement SVE2 integer add/subtract interleaved long Richard Henderson
2020-03-26 23:08 ` [PATCH 13/31] target/arm: Implement SVE2 integer add/subtract wide Richard Henderson
2020-04-13 16:11 ` Laurent Desnogues
2020-03-26 23:08 ` [PATCH 14/31] target/arm: Implement SVE2 integer multiply long Richard Henderson
2020-03-26 23:08 ` [PATCH 15/31] target/arm: Implement PMULLB and PMULLT Richard Henderson
2020-03-26 23:08 ` [PATCH 16/31] target/arm: Tidy SVE tszimm shift formats Richard Henderson
2020-03-26 23:08 ` [PATCH 17/31] target/arm: Implement SVE2 bitwise shift left long Richard Henderson
2020-03-26 23:08 ` [PATCH 18/31] target/arm: Implement SVE2 bitwise exclusive-or interleaved Richard Henderson
2020-03-26 23:08 ` [PATCH 19/31] target/arm: Implement SVE2 bitwise permute Richard Henderson
2020-03-26 23:08 ` [PATCH 20/31] target/arm: Implement SVE2 complex integer add Richard Henderson
2020-04-13 16:20 ` Laurent Desnogues
2020-03-26 23:08 ` [PATCH 21/31] target/arm: Implement SVE2 integer absolute difference and accumulate long Richard Henderson
2020-04-13 16:15 ` Laurent Desnogues
2020-04-13 23:19 ` Richard Henderson
2020-04-14 7:04 ` Laurent Desnogues
2020-03-26 23:08 ` [PATCH 22/31] target/arm: Implement SVE2 integer add/subtract long with carry Richard Henderson
2020-04-13 16:18 ` Laurent Desnogues
2020-03-26 23:08 ` [PATCH 23/31] target/arm: Create arm_gen_gvec_[us]sra Richard Henderson
2020-03-26 23:08 ` [PATCH 24/31] target/arm: Create arm_gen_gvec_{u,s}{rshr,rsra} Richard Henderson
2020-03-26 23:08 ` [PATCH 25/31] target/arm: Implement SVE2 bitwise shift right and accumulate Richard Henderson
2020-03-26 23:08 ` [PATCH 26/31] target/arm: Create arm_gen_gvec_{sri,sli} Richard Henderson
2020-03-26 23:08 ` [PATCH 27/31] target/arm: Tidy handle_vec_simd_shri Richard Henderson
2020-03-26 23:08 ` [PATCH 28/31] target/arm: Implement SVE2 bitwise shift and insert Richard Henderson
2020-03-26 23:08 ` Richard Henderson [this message]
2020-03-26 23:08 ` [PATCH 30/31] target/arm: Vectorize SABA/UABA Richard Henderson
2020-03-26 23:08 ` [PATCH 31/31] target/arm: Implement SVE2 integer absolute difference and accumulate Richard Henderson
2020-04-01 21:17 ` [PATCH for-5.1 00/31] target/arm: SVE2, part 1 Ana Pazos
2020-04-22 2:51 ` LIU Zhiwei
2020-04-22 2:55 ` Richard Henderson
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=20200326230838.31112-30-richard.henderson@linaro.org \
--to=richard.henderson@linaro.org \
--cc=apazos@quicinc.com \
--cc=qemu-arm@nongnu.org \
--cc=qemu-devel@nongnu.org \
--cc=rajav@quicinc.com \
/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).