* [PATCH 0/6] target/arm: partial vector cleanup
@ 2020-05-01 21:13 Richard Henderson
2020-05-01 21:13 ` [PATCH 1/6] target/arm: Create gen_gvec_[us]sra Richard Henderson
` (5 more replies)
0 siblings, 6 replies; 7+ messages in thread
From: Richard Henderson @ 2020-05-01 21:13 UTC (permalink / raw)
To: qemu-devel; +Cc: peter.maydell
This is not complete, but shows the direction I'd like to go.
It may well help what Peter is doing with the neon decodetree
conversion. It may be helpful to apply before the conversion
in order to reduce the number of special cases. As may
continuing with the cleanup; I'll probably work on that more
next week.
r~
Richard Henderson (6):
target/arm: Create gen_gvec_[us]sra
target/arm: Create gen_gvec_{u,s}{rshr,rsra}
target/arm: Create gen_gvec_{sri,sli}
target/arm: Remove unnecessary range check for VSHL
target/arm: Tidy handle_vec_simd_shri
target/arm: Wrap vector compare zero GVecGen2 in GVecGen2Fn
target/arm/helper.h | 40 ++
target/arm/translate.h | 39 +-
target/arm/translate-a64.c | 102 +---
target/arm/translate.c | 1074 +++++++++++++++++++++++-------------
target/arm/vec_helper.c | 113 ++++
5 files changed, 903 insertions(+), 465 deletions(-)
--
2.20.1
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH 1/6] target/arm: Create gen_gvec_[us]sra
2020-05-01 21:13 [PATCH 0/6] target/arm: partial vector cleanup Richard Henderson
@ 2020-05-01 21:13 ` Richard Henderson
2020-05-01 21:13 ` [PATCH 2/6] target/arm: Create gen_gvec_{u,s}{rshr,rsra} Richard Henderson
` (4 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Richard Henderson @ 2020-05-01 21:13 UTC (permalink / raw)
To: qemu-devel; +Cc: peter.maydell
The functions eliminate duplication of the special cases for
this operation. They match up with the GVecGen2iFn typedef.
Add out-of-line helpers. We got away with only having inline
expanders because the neon vector size is only 16 bytes, and
we know that the inline expansion will always succeed.
When we reuse this for SVE, tcg-gvec-op may decide to use an
out-of-line helper due to longer vector lengths.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/helper.h | 10 +++
target/arm/translate.h | 7 +-
target/arm/translate-a64.c | 15 +---
target/arm/translate.c | 161 ++++++++++++++++++++++---------------
target/arm/vec_helper.c | 25 ++++++
5 files changed, 139 insertions(+), 79 deletions(-)
diff --git a/target/arm/helper.h b/target/arm/helper.h
index 5817626b20..9bc162345c 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -691,6 +691,16 @@ DEF_HELPER_FLAGS_4(gvec_pmull_q, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(neon_pmull_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ssra_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ssra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ssra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ssra_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_usra_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_usra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_usra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_usra_d, TCG_CALL_NO_RWG, void, 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 98b319f3f6..a39cf22666 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -285,8 +285,6 @@ extern const GVecGen3 mls_op[4];
extern const GVecGen3 cmtst_op[4];
extern const GVecGen3 sshl_op[4];
extern const GVecGen3 ushl_op[4];
-extern const GVecGen2i ssra_op[4];
-extern const GVecGen2i usra_op[4];
extern const GVecGen2i sri_op[4];
extern const GVecGen2i sli_op[4];
extern const GVecGen4 uqadd_op[4];
@@ -299,6 +297,11 @@ void gen_sshl_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b);
void gen_ushl_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b);
void gen_sshl_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b);
+void gen_gvec_ssra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_usra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, 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 010b36633e..03f4dc5805 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -10205,19 +10205,8 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
switch (opcode) {
case 0x02: /* SSRA / USRA (accumulate) */
- if (is_u) {
- /* Shift count same as element size produces zero to add. */
- if (shift == 8 << size) {
- goto done;
- }
- gen_gvec_op2i(s, is_q, rd, rn, shift, &usra_op[size]);
- } else {
- /* Shift count same as element size produces all sign to add. */
- if (shift == 8 << size) {
- shift -= 1;
- }
- gen_gvec_op2i(s, is_q, rd, rn, shift, &ssra_op[size]);
- }
+ gen_gvec_fn2i(s, is_q, rd, rn, shift,
+ is_u ? gen_gvec_usra : gen_gvec_ssra, size);
return;
case 0x08: /* SRI */
/* Shift count same as element size is valid but does nothing. */
diff --git a/target/arm/translate.c b/target/arm/translate.c
index a96899549b..04114906d7 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -4146,33 +4146,51 @@ static void gen_ssra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
tcg_gen_add_vec(vece, d, d, a);
}
-static const TCGOpcode vecop_list_ssra[] = {
- INDEX_op_sari_vec, INDEX_op_add_vec, 0
-};
+void gen_gvec_ssra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_sari_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_ssra8_i64,
+ .fniv = gen_ssra_vec,
+ .fno = gen_helper_gvec_ssra_b,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fni8 = gen_ssra16_i64,
+ .fniv = gen_ssra_vec,
+ .fno = gen_helper_gvec_ssra_h,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_ssra32_i32,
+ .fniv = gen_ssra_vec,
+ .fno = gen_helper_gvec_ssra_s,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_ssra64_i64,
+ .fniv = gen_ssra_vec,
+ .fno = gen_helper_gvec_ssra_b,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_64 },
+ };
-const GVecGen2i ssra_op[4] = {
- { .fni8 = gen_ssra8_i64,
- .fniv = gen_ssra_vec,
- .load_dest = true,
- .opt_opc = vecop_list_ssra,
- .vece = MO_8 },
- { .fni8 = gen_ssra16_i64,
- .fniv = gen_ssra_vec,
- .load_dest = true,
- .opt_opc = vecop_list_ssra,
- .vece = MO_16 },
- { .fni4 = gen_ssra32_i32,
- .fniv = gen_ssra_vec,
- .load_dest = true,
- .opt_opc = vecop_list_ssra,
- .vece = MO_32 },
- { .fni8 = gen_ssra64_i64,
- .fniv = gen_ssra_vec,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .opt_opc = vecop_list_ssra,
- .load_dest = true,
- .vece = MO_64 },
-};
+ /* tszimm encoding produces immediates in the range [1..esize]. */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Signed results in all sign bits.
+ */
+ shift = MIN(shift, (8 << vece) - 1);
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+}
static void gen_usra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
{
@@ -4204,33 +4222,55 @@ static void gen_usra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
tcg_gen_add_vec(vece, d, d, a);
}
-static const TCGOpcode vecop_list_usra[] = {
- INDEX_op_shri_vec, INDEX_op_add_vec, 0
-};
+void gen_gvec_usra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_usra8_i64,
+ .fniv = gen_usra_vec,
+ .fno = gen_helper_gvec_usra_b,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_8, },
+ { .fni8 = gen_usra16_i64,
+ .fniv = gen_usra_vec,
+ .fno = gen_helper_gvec_usra_h,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_16, },
+ { .fni4 = gen_usra32_i32,
+ .fniv = gen_usra_vec,
+ .fno = gen_helper_gvec_usra_s,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_32, },
+ { .fni8 = gen_usra64_i64,
+ .fniv = gen_usra_vec,
+ .fno = gen_helper_gvec_usra_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_64, },
+ };
-const GVecGen2i usra_op[4] = {
- { .fni8 = gen_usra8_i64,
- .fniv = gen_usra_vec,
- .load_dest = true,
- .opt_opc = vecop_list_usra,
- .vece = MO_8, },
- { .fni8 = gen_usra16_i64,
- .fniv = gen_usra_vec,
- .load_dest = true,
- .opt_opc = vecop_list_usra,
- .vece = MO_16, },
- { .fni4 = gen_usra32_i32,
- .fniv = gen_usra_vec,
- .load_dest = true,
- .opt_opc = vecop_list_usra,
- .vece = MO_32, },
- { .fni8 = gen_usra64_i64,
- .fniv = gen_usra_vec,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .load_dest = true,
- .opt_opc = vecop_list_usra,
- .vece = MO_64, },
-};
+ /* tszimm encoding produces immediates in the range [1..esize]. */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Unsigned results in all zeros as input to accumulate: nop.
+ */
+ if (shift < (8 << vece)) {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ } else {
+ /* Nop, but we do need to clear the tail. */
+ tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
+ }
+}
static void gen_shr8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
{
@@ -5596,19 +5636,12 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
case 1: /* VSRA */
/* Right shift comes here negative. */
shift = -shift;
- /* Shifts larger than the element size are architecturally
- * valid. Unsigned results in all zeros; signed results
- * in all sign bits.
- */
- if (!u) {
- tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size, vec_size,
- MIN(shift, (8 << size) - 1),
- &ssra_op[size]);
- } else if (shift >= 8 << size) {
- /* rd += 0 */
+ if (u) {
+ gen_gvec_usra(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
} else {
- tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size, vec_size,
- shift, &usra_op[size]);
+ gen_gvec_ssra(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
}
return 0;
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index 3d534188a8..230085b35e 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -899,6 +899,31 @@ void HELPER(gvec_sqsub_d)(void *vd, void *vq, void *vn,
clear_tail(d, oprsz, simd_maxsz(desc));
}
+
+#define DO_SRA(NAME, TYPE) \
+void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \
+{ \
+ intptr_t i, oprsz = simd_oprsz(desc); \
+ int shift = simd_data(desc); \
+ TYPE *d = vd, *n = vn; \
+ for (i = 0; i < oprsz / sizeof(TYPE); i++) { \
+ d[i] += n[i] >> shift; \
+ } \
+ clear_tail(d, oprsz, simd_maxsz(desc)); \
+}
+
+DO_SRA(gvec_ssra_b, int8_t)
+DO_SRA(gvec_ssra_h, int16_t)
+DO_SRA(gvec_ssra_s, int32_t)
+DO_SRA(gvec_ssra_d, int64_t)
+
+DO_SRA(gvec_usra_b, uint8_t)
+DO_SRA(gvec_usra_h, uint16_t)
+DO_SRA(gvec_usra_s, uint32_t)
+DO_SRA(gvec_usra_d, uint64_t)
+
+#undef DO_SRA
+
/*
* Convert float16 to float32, raising no exceptions and
* preserving exceptional values, including SNaN.
--
2.20.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 2/6] target/arm: Create gen_gvec_{u,s}{rshr,rsra}
2020-05-01 21:13 [PATCH 0/6] target/arm: partial vector cleanup Richard Henderson
2020-05-01 21:13 ` [PATCH 1/6] target/arm: Create gen_gvec_[us]sra Richard Henderson
@ 2020-05-01 21:13 ` Richard Henderson
2020-05-01 21:13 ` [PATCH 3/6] target/arm: Create gen_gvec_{sri,sli} Richard Henderson
` (3 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Richard Henderson @ 2020-05-01 21:13 UTC (permalink / raw)
To: qemu-devel; +Cc: peter.maydell
Create vectorized versions of handle_shri_with_rndacc
for shift+round and shift+round+accumulate. Add out-of-line
helpers in preparation for longer vector lengths from SVE.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/helper.h | 20 ++
target/arm/translate.h | 9 +
target/arm/translate-a64.c | 11 +-
target/arm/translate.c | 461 +++++++++++++++++++++++++++++++++++--
target/arm/vec_helper.c | 50 ++++
5 files changed, 525 insertions(+), 26 deletions(-)
diff --git a/target/arm/helper.h b/target/arm/helper.h
index 9bc162345c..aeb1f52455 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -701,6 +701,26 @@ DEF_HELPER_FLAGS_3(gvec_usra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_usra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_usra_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_urshr_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_urshr_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_urshr_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_urshr_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_srsra_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srsra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srsra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srsra_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_ursra_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ursra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ursra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ursra_d, TCG_CALL_NO_RWG, void, 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 a39cf22666..823821f82c 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -302,6 +302,15 @@ void gen_gvec_ssra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
void gen_gvec_usra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_srshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_urshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_srsra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_ursra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, 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 03f4dc5805..1ef05d5ce1 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -10235,10 +10235,15 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
return;
case 0x04: /* SRSHR / URSHR (rounding) */
- break;
+ gen_gvec_fn2i(s, is_q, rd, rn, shift,
+ is_u ? gen_gvec_urshr : gen_gvec_srshr, size);
+ return;
+
case 0x06: /* SRSRA / URSRA (accum + rounding) */
- accumulate = true;
- break;
+ gen_gvec_fn2i(s, is_q, rd, rn, shift,
+ is_u ? gen_gvec_ursra : gen_gvec_srsra, size);
+ return;
+
default:
g_assert_not_reached();
}
diff --git a/target/arm/translate.c b/target/arm/translate.c
index 04114906d7..d724022cb6 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -4272,6 +4272,422 @@ void gen_gvec_usra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
}
}
+/*
+ * Shift one less than the requested amount, and the low bit is
+ * the rounding bit. For the 8 and 16-bit operations, because we
+ * mask the low bit, we can perform a normal integer shift instead
+ * of a vector shift.
+ */
+static void gen_srshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_shri_i64(t, a, sh - 1);
+ tcg_gen_andi_i64(t, t, dup_const(MO_8, 1));
+ tcg_gen_vec_sar8i_i64(d, a, sh);
+ tcg_gen_vec_add8_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_shri_i64(t, a, sh - 1);
+ tcg_gen_andi_i64(t, t, dup_const(MO_16, 1));
+ tcg_gen_vec_sar16i_i64(d, a, sh);
+ tcg_gen_vec_add16_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ tcg_gen_extract_i32(t, a, sh - 1, 1);
+ tcg_gen_sari_i32(d, a, sh);
+ tcg_gen_add_i32(d, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_srshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_extract_i64(t, a, sh - 1, 1);
+ tcg_gen_sari_i64(d, a, sh);
+ tcg_gen_add_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+ TCGv_vec ones = tcg_temp_new_vec_matching(d);
+
+ tcg_gen_shri_vec(vece, t, a, sh - 1);
+ tcg_gen_dupi_vec(vece, ones, 1);
+ tcg_gen_and_vec(vece, t, t, ones);
+ tcg_gen_sari_vec(vece, d, a, sh);
+ tcg_gen_add_vec(vece, d, d, t);
+
+ tcg_temp_free_vec(t);
+ tcg_temp_free_vec(ones);
+}
+
+void gen_gvec_srshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_sari_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_srshr8_i64,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_b,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fni8 = gen_srshr16_i64,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_h,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_srshr32_i32,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_s,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_srshr64_i64,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ if (shift == (8 << vece)) {
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Signed results in all sign bits. With rounding, this produces
+ * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
+ * I.e. always zero.
+ */
+ tcg_gen_gvec_dup_imm(vece, rd_ofs, opr_sz, max_sz, 0);
+ } else {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ }
+}
+
+static void gen_srsra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ gen_srshr8_i64(t, a, sh);
+ tcg_gen_vec_add8_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srsra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ gen_srshr16_i64(t, a, sh);
+ tcg_gen_vec_add16_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srsra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ gen_srshr32_i32(t, a, sh);
+ tcg_gen_add_i32(d, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_srsra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ gen_srshr64_i64(t, a, sh);
+ tcg_gen_add_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srsra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+ gen_srshr_vec(vece, t, a, sh);
+ tcg_gen_add_vec(vece, d, d, t);
+ tcg_temp_free_vec(t);
+}
+
+void gen_gvec_srsra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_sari_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_srsra8_i64,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_b,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_8 },
+ { .fni8 = gen_srsra16_i64,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_h,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_16 },
+ { .fni4 = gen_srsra32_i32,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_s,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_32 },
+ { .fni8 = gen_srsra64_i64,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Signed results in all sign bits. With rounding, this produces
+ * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
+ * I.e. always zero. With accumulation, this leaves D unchanged.
+ */
+ if (shift == (8 << vece)) {
+ /* Nop, but we do need to clear the tail. */
+ tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
+ } else {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ }
+}
+
+static void gen_urshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_shri_i64(t, a, sh - 1);
+ tcg_gen_andi_i64(t, t, dup_const(MO_8, 1));
+ tcg_gen_vec_shr8i_i64(d, a, sh);
+ tcg_gen_vec_add8_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_urshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_shri_i64(t, a, sh - 1);
+ tcg_gen_andi_i64(t, t, dup_const(MO_16, 1));
+ tcg_gen_vec_shr16i_i64(d, a, sh);
+ tcg_gen_vec_add16_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_urshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ tcg_gen_extract_i32(t, a, sh - 1, 1);
+ tcg_gen_shri_i32(d, a, sh);
+ tcg_gen_add_i32(d, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_urshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_extract_i64(t, a, sh - 1, 1);
+ tcg_gen_shri_i64(d, a, sh);
+ tcg_gen_add_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_urshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t shift)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+ TCGv_vec ones = tcg_temp_new_vec_matching(d);
+
+ tcg_gen_shri_vec(vece, t, a, shift - 1);
+ tcg_gen_dupi_vec(vece, ones, 1);
+ tcg_gen_and_vec(vece, t, t, ones);
+ tcg_gen_shri_vec(vece, d, a, shift);
+ tcg_gen_add_vec(vece, d, d, t);
+
+ tcg_temp_free_vec(t);
+ tcg_temp_free_vec(ones);
+}
+
+void gen_gvec_urshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_urshr8_i64,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_b,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fni8 = gen_urshr16_i64,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_h,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_urshr32_i32,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_s,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_urshr64_i64,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ if (shift == (8 << vece)) {
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Unsigned results in zero. With rounding, this produces a
+ * copy of the most significant bit.
+ */
+ tcg_gen_gvec_shri(vece, rd_ofs, rm_ofs, shift - 1, opr_sz, max_sz);
+ } else {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ }
+}
+
+static void gen_ursra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ if (sh == 8) {
+ tcg_gen_vec_shr8i_i64(t, a, 7);
+ } else {
+ gen_urshr8_i64(t, a, sh);
+ }
+ tcg_gen_vec_add8_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_ursra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ if (sh == 16) {
+ tcg_gen_vec_shr16i_i64(t, a, 15);
+ } else {
+ gen_urshr16_i64(t, a, sh);
+ }
+ tcg_gen_vec_add16_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_ursra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ if (sh == 32) {
+ tcg_gen_shri_i32(t, a, 31);
+ } else {
+ gen_urshr32_i32(t, a, sh);
+ }
+ tcg_gen_add_i32(d, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_ursra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ if (sh == 64) {
+ tcg_gen_shri_i64(t, a, 63);
+ } else {
+ gen_urshr64_i64(t, a, sh);
+ }
+ tcg_gen_add_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_ursra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+ if (sh == (8 << vece)) {
+ tcg_gen_shri_vec(vece, t, a, sh - 1);
+ } else {
+ gen_urshr_vec(vece, t, a, sh);
+ }
+ tcg_gen_add_vec(vece, d, d, t);
+ tcg_temp_free_vec(t);
+}
+
+void gen_gvec_ursra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_ursra8_i64,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_b,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_8 },
+ { .fni8 = gen_ursra16_i64,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_h,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_16 },
+ { .fni4 = gen_ursra32_i32,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_s,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_32 },
+ { .fni8 = gen_ursra64_i64,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+}
+
static void gen_shr8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
{
uint64_t mask = dup_const(MO_8, 0xff >> shift);
@@ -5645,6 +6061,28 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
}
return 0;
+ case 2: /* VRSHR */
+ /* Right shift comes here negative. */
+ shift = -shift;
+ if (u) {
+ gen_gvec_urshr(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
+ } else {
+ gen_gvec_srshr(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
+ }
+ return 0;
+
+ case 3: /* VRSRA */
+ if (u) {
+ gen_gvec_ursra(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
+ } else {
+ gen_gvec_srsra(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
+ }
+ return 0;
+
case 4: /* VSRI */
if (!u) {
return 1;
@@ -5696,13 +6134,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
neon_load_reg64(cpu_V0, rm + pass);
tcg_gen_movi_i64(cpu_V1, imm);
switch (op) {
- case 2: /* VRSHR */
- case 3: /* VRSRA */
- if (u)
- gen_helper_neon_rshl_u64(cpu_V0, cpu_V0, cpu_V1);
- else
- gen_helper_neon_rshl_s64(cpu_V0, cpu_V0, cpu_V1);
- break;
case 6: /* VQSHLU */
gen_helper_neon_qshlu_s64(cpu_V0, cpu_env,
cpu_V0, cpu_V1);
@@ -5719,11 +6150,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
default:
g_assert_not_reached();
}
- if (op == 3) {
- /* Accumulate. */
- neon_load_reg64(cpu_V1, rd + pass);
- tcg_gen_add_i64(cpu_V0, cpu_V0, cpu_V1);
- }
neon_store_reg64(cpu_V0, rd + pass);
} else { /* size < 3 */
/* Operands in T0 and T1. */
@@ -5731,10 +6157,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
tmp2 = tcg_temp_new_i32();
tcg_gen_movi_i32(tmp2, imm);
switch (op) {
- case 2: /* VRSHR */
- case 3: /* VRSRA */
- GEN_NEON_INTEGER_OP(rshl);
- break;
case 6: /* VQSHLU */
switch (size) {
case 0:
@@ -5760,13 +6182,6 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
g_assert_not_reached();
}
tcg_temp_free_i32(tmp2);
-
- if (op == 3) {
- /* Accumulate. */
- tmp2 = neon_load_reg(rd, pass);
- gen_neon_add(size, tmp, tmp2);
- tcg_temp_free_i32(tmp2);
- }
neon_store_reg(rd, pass, tmp);
}
} /* for pass */
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index 230085b35e..fd8b2bff49 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -924,6 +924,56 @@ DO_SRA(gvec_usra_d, uint64_t)
#undef DO_SRA
+#define DO_RSHR(NAME, TYPE) \
+void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \
+{ \
+ intptr_t i, oprsz = simd_oprsz(desc); \
+ int shift = simd_data(desc); \
+ TYPE *d = vd, *n = vn; \
+ for (i = 0; i < oprsz / sizeof(TYPE); i++) { \
+ TYPE tmp = n[i] >> (shift - 1); \
+ d[i] = (tmp >> 1) + (tmp & 1); \
+ } \
+ clear_tail(d, oprsz, simd_maxsz(desc)); \
+}
+
+DO_RSHR(gvec_srshr_b, int8_t)
+DO_RSHR(gvec_srshr_h, int16_t)
+DO_RSHR(gvec_srshr_s, int32_t)
+DO_RSHR(gvec_srshr_d, int64_t)
+
+DO_RSHR(gvec_urshr_b, uint8_t)
+DO_RSHR(gvec_urshr_h, uint16_t)
+DO_RSHR(gvec_urshr_s, uint32_t)
+DO_RSHR(gvec_urshr_d, uint64_t)
+
+#undef DO_RSHR
+
+#define DO_RSRA(NAME, TYPE) \
+void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \
+{ \
+ intptr_t i, oprsz = simd_oprsz(desc); \
+ int shift = simd_data(desc); \
+ TYPE *d = vd, *n = vn; \
+ for (i = 0; i < oprsz / sizeof(TYPE); i++) { \
+ TYPE tmp = n[i] >> (shift - 1); \
+ d[i] += (tmp >> 1) + (tmp & 1); \
+ } \
+ clear_tail(d, oprsz, simd_maxsz(desc)); \
+}
+
+DO_RSRA(gvec_srsra_b, int8_t)
+DO_RSRA(gvec_srsra_h, int16_t)
+DO_RSRA(gvec_srsra_s, int32_t)
+DO_RSRA(gvec_srsra_d, int64_t)
+
+DO_RSRA(gvec_ursra_b, uint8_t)
+DO_RSRA(gvec_ursra_h, uint16_t)
+DO_RSRA(gvec_ursra_s, uint32_t)
+DO_RSRA(gvec_ursra_d, uint64_t)
+
+#undef DO_RSRA
+
/*
* Convert float16 to float32, raising no exceptions and
* preserving exceptional values, including SNaN.
--
2.20.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 3/6] target/arm: Create gen_gvec_{sri,sli}
2020-05-01 21:13 [PATCH 0/6] target/arm: partial vector cleanup Richard Henderson
2020-05-01 21:13 ` [PATCH 1/6] target/arm: Create gen_gvec_[us]sra Richard Henderson
2020-05-01 21:13 ` [PATCH 2/6] target/arm: Create gen_gvec_{u,s}{rshr,rsra} Richard Henderson
@ 2020-05-01 21:13 ` Richard Henderson
2020-05-01 21:13 ` [PATCH 4/6] target/arm: Remove unnecessary range check for VSHL Richard Henderson
` (2 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Richard Henderson @ 2020-05-01 21:13 UTC (permalink / raw)
To: qemu-devel; +Cc: peter.maydell
The functions eliminate duplication of the special cases for
this operation. They match up with the GVecGen2iFn typedef.
Add out-of-line helpers. We got away with only having inline
expanders because the neon vector size is only 16 bytes, and
we know that the inline expansion will always succeed.
When we reuse this for SVE, tcg-gvec-op may decide to use an
out-of-line helper due to longer vector lengths.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/helper.h | 10 ++
target/arm/translate.h | 7 +-
target/arm/translate-a64.c | 20 +---
target/arm/translate.c | 186 +++++++++++++++++++++----------------
target/arm/vec_helper.c | 38 ++++++++
5 files changed, 160 insertions(+), 101 deletions(-)
diff --git a/target/arm/helper.h b/target/arm/helper.h
index aeb1f52455..33c76192d2 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -721,6 +721,16 @@ DEF_HELPER_FLAGS_3(gvec_ursra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_ursra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_ursra_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_sri_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_sri_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_sri_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_sri_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_sli_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+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)
+
#ifdef TARGET_AARCH64
#include "helper-a64.h"
#include "helper-sve.h"
diff --git a/target/arm/translate.h b/target/arm/translate.h
index 823821f82c..7a2008f0dd 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -285,8 +285,6 @@ extern const GVecGen3 mls_op[4];
extern const GVecGen3 cmtst_op[4];
extern const GVecGen3 sshl_op[4];
extern const GVecGen3 ushl_op[4];
-extern const GVecGen2i sri_op[4];
-extern const GVecGen2i sli_op[4];
extern const GVecGen4 uqadd_op[4];
extern const GVecGen4 sqadd_op[4];
extern const GVecGen4 uqsub_op[4];
@@ -311,6 +309,11 @@ void gen_gvec_srsra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
void gen_gvec_ursra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_sri(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_sli(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, 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 1ef05d5ce1..bc326dadda 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -602,16 +602,6 @@ static void gen_gvec_op2(DisasContext *s, bool is_q, int rd,
is_q ? 16 : 8, vec_full_reg_size(s), gvec_op);
}
-/* Expand a 2-operand + immediate AdvSIMD vector operation using
- * an op descriptor.
- */
-static void gen_gvec_op2i(DisasContext *s, bool is_q, int rd,
- int rn, int64_t imm, const GVecGen2i *gvec_op)
-{
- tcg_gen_gvec_2i(vec_full_reg_offset(s, rd), vec_full_reg_offset(s, rn),
- is_q ? 16 : 8, vec_full_reg_size(s), imm, gvec_op);
-}
-
/* Expand a 3-operand AdvSIMD vector operation using an op descriptor. */
static void gen_gvec_op3(DisasContext *s, bool is_q, int rd,
int rn, int rm, const GVecGen3 *gvec_op)
@@ -10208,12 +10198,9 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
gen_gvec_fn2i(s, is_q, rd, rn, shift,
is_u ? gen_gvec_usra : gen_gvec_ssra, size);
return;
+
case 0x08: /* SRI */
- /* Shift count same as element size is valid but does nothing. */
- if (shift == 8 << size) {
- goto done;
- }
- gen_gvec_op2i(s, is_q, rd, rn, shift, &sri_op[size]);
+ gen_gvec_fn2i(s, is_q, rd, rn, shift, gen_gvec_sri, size);
return;
case 0x00: /* SSHR / USHR */
@@ -10264,7 +10251,6 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
}
tcg_temp_free_i64(tcg_round);
- done:
clear_vec_high(s, is_q, rd);
}
@@ -10289,7 +10275,7 @@ static void handle_vec_simd_shli(DisasContext *s, bool is_q, bool insert,
}
if (insert) {
- gen_gvec_op2i(s, is_q, rd, rn, shift, &sli_op[size]);
+ gen_gvec_fn2i(s, is_q, rd, rn, shift, gen_gvec_sli, size);
} else {
gen_gvec_fn2i(s, is_q, rd, rn, shift, tcg_gen_gvec_shli, size);
}
diff --git a/target/arm/translate.c b/target/arm/translate.c
index d724022cb6..f730eb5b75 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -4726,47 +4726,62 @@ static void gen_shr64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
static void gen_shr_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
{
- if (sh == 0) {
- tcg_gen_mov_vec(d, a);
- } else {
- TCGv_vec t = tcg_temp_new_vec_matching(d);
- TCGv_vec m = tcg_temp_new_vec_matching(d);
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+ TCGv_vec m = tcg_temp_new_vec_matching(d);
- tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK((8 << vece) - sh, sh));
- tcg_gen_shri_vec(vece, t, a, sh);
- tcg_gen_and_vec(vece, d, d, m);
- tcg_gen_or_vec(vece, d, d, t);
+ tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK((8 << vece) - sh, sh));
+ tcg_gen_shri_vec(vece, t, a, sh);
+ tcg_gen_and_vec(vece, d, d, m);
+ tcg_gen_or_vec(vece, d, d, t);
- tcg_temp_free_vec(t);
- tcg_temp_free_vec(m);
- }
+ tcg_temp_free_vec(t);
+ tcg_temp_free_vec(m);
}
-static const TCGOpcode vecop_list_sri[] = { INDEX_op_shri_vec, 0 };
+void gen_gvec_sri(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = { INDEX_op_shri_vec, 0 };
+ const GVecGen2i ops[4] = {
+ { .fni8 = gen_shr8_ins_i64,
+ .fniv = gen_shr_ins_vec,
+ .fno = gen_helper_gvec_sri_b,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fni8 = gen_shr16_ins_i64,
+ .fniv = gen_shr_ins_vec,
+ .fno = gen_helper_gvec_sri_h,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_shr32_ins_i32,
+ .fniv = gen_shr_ins_vec,
+ .fno = gen_helper_gvec_sri_s,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_shr64_ins_i64,
+ .fniv = gen_shr_ins_vec,
+ .fno = gen_helper_gvec_sri_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
-const GVecGen2i sri_op[4] = {
- { .fni8 = gen_shr8_ins_i64,
- .fniv = gen_shr_ins_vec,
- .load_dest = true,
- .opt_opc = vecop_list_sri,
- .vece = MO_8 },
- { .fni8 = gen_shr16_ins_i64,
- .fniv = gen_shr_ins_vec,
- .load_dest = true,
- .opt_opc = vecop_list_sri,
- .vece = MO_16 },
- { .fni4 = gen_shr32_ins_i32,
- .fniv = gen_shr_ins_vec,
- .load_dest = true,
- .opt_opc = vecop_list_sri,
- .vece = MO_32 },
- { .fni8 = gen_shr64_ins_i64,
- .fniv = gen_shr_ins_vec,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .load_dest = true,
- .opt_opc = vecop_list_sri,
- .vece = MO_64 },
-};
+ /* tszimm encoding produces immediates in the range [1..esize]. */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ /* Shift of esize leaves destination unchanged. */
+ if (shift < (8 << vece)) {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ } else {
+ /* Nop, but we do need to clear the tail. */
+ tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
+ }
+}
static void gen_shl8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
{
@@ -4804,47 +4819,60 @@ static void gen_shl64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
static void gen_shl_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
{
- if (sh == 0) {
- tcg_gen_mov_vec(d, a);
- } else {
- TCGv_vec t = tcg_temp_new_vec_matching(d);
- TCGv_vec m = tcg_temp_new_vec_matching(d);
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+ TCGv_vec m = tcg_temp_new_vec_matching(d);
- tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK(0, sh));
- tcg_gen_shli_vec(vece, t, a, sh);
- tcg_gen_and_vec(vece, d, d, m);
- tcg_gen_or_vec(vece, d, d, t);
+ tcg_gen_shli_vec(vece, t, a, sh);
+ tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK(0, sh));
+ tcg_gen_and_vec(vece, d, d, m);
+ tcg_gen_or_vec(vece, d, d, t);
- tcg_temp_free_vec(t);
- tcg_temp_free_vec(m);
- }
+ tcg_temp_free_vec(t);
+ tcg_temp_free_vec(m);
}
-static const TCGOpcode vecop_list_sli[] = { INDEX_op_shli_vec, 0 };
+void gen_gvec_sli(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
+ const GVecGen2i ops[4] = {
+ { .fni8 = gen_shl8_ins_i64,
+ .fniv = gen_shl_ins_vec,
+ .fno = gen_helper_gvec_sli_b,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fni8 = gen_shl16_ins_i64,
+ .fniv = gen_shl_ins_vec,
+ .fno = gen_helper_gvec_sli_h,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_shl32_ins_i32,
+ .fniv = gen_shl_ins_vec,
+ .fno = gen_helper_gvec_sli_s,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_shl64_ins_i64,
+ .fniv = gen_shl_ins_vec,
+ .fno = gen_helper_gvec_sli_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .load_dest = true,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
-const GVecGen2i sli_op[4] = {
- { .fni8 = gen_shl8_ins_i64,
- .fniv = gen_shl_ins_vec,
- .load_dest = true,
- .opt_opc = vecop_list_sli,
- .vece = MO_8 },
- { .fni8 = gen_shl16_ins_i64,
- .fniv = gen_shl_ins_vec,
- .load_dest = true,
- .opt_opc = vecop_list_sli,
- .vece = MO_16 },
- { .fni4 = gen_shl32_ins_i32,
- .fniv = gen_shl_ins_vec,
- .load_dest = true,
- .opt_opc = vecop_list_sli,
- .vece = MO_32 },
- { .fni8 = gen_shl64_ins_i64,
- .fniv = gen_shl_ins_vec,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .load_dest = true,
- .opt_opc = vecop_list_sli,
- .vece = MO_64 },
-};
+ /* tszimm encoding produces immediates in the range [0..esize-1]. */
+ tcg_debug_assert(shift >= 0);
+ tcg_debug_assert(shift < (8 << vece));
+
+ if (shift == 0) {
+ tcg_gen_gvec_mov(vece, rd_ofs, rm_ofs, opr_sz, max_sz);
+ } else {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ }
+}
static void gen_mla8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
{
@@ -6089,20 +6117,14 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
}
/* Right shift comes here negative. */
shift = -shift;
- /* Shift out of range leaves destination unchanged. */
- if (shift < 8 << size) {
- tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size, vec_size,
- shift, &sri_op[size]);
- }
+ gen_gvec_sri(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
return 0;
case 5: /* VSHL, VSLI */
if (u) { /* VSLI */
- /* Shift out of range leaves destination unchanged. */
- if (shift < 8 << size) {
- tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size,
- vec_size, shift, &sli_op[size]);
- }
+ gen_gvec_sli(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
} else { /* VSHL */
/* Shifts larger than the element size are
* architecturally valid and results in zero.
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index fd8b2bff49..096fea67ef 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -974,6 +974,44 @@ DO_RSRA(gvec_ursra_d, uint64_t)
#undef DO_RSRA
+#define DO_SRI(NAME, TYPE) \
+void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \
+{ \
+ intptr_t i, oprsz = simd_oprsz(desc); \
+ int shift = simd_data(desc); \
+ TYPE *d = vd, *n = vn; \
+ for (i = 0; i < oprsz / sizeof(TYPE); i++) { \
+ d[i] = deposit64(d[i], 0, sizeof(TYPE) * 8 - shift, n[i] >> shift); \
+ } \
+ clear_tail(d, oprsz, simd_maxsz(desc)); \
+}
+
+DO_SRI(gvec_sri_b, uint8_t)
+DO_SRI(gvec_sri_h, uint16_t)
+DO_SRI(gvec_sri_s, uint32_t)
+DO_SRI(gvec_sri_d, uint64_t)
+
+#undef DO_SRI
+
+#define DO_SLI(NAME, TYPE) \
+void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \
+{ \
+ intptr_t i, oprsz = simd_oprsz(desc); \
+ int shift = simd_data(desc); \
+ TYPE *d = vd, *n = vn; \
+ for (i = 0; i < oprsz / sizeof(TYPE); i++) { \
+ d[i] = deposit64(d[i], shift, sizeof(TYPE) * 8 - shift, n[i]); \
+ } \
+ clear_tail(d, oprsz, simd_maxsz(desc)); \
+}
+
+DO_SLI(gvec_sli_b, uint8_t)
+DO_SLI(gvec_sli_h, uint16_t)
+DO_SLI(gvec_sli_s, uint32_t)
+DO_SLI(gvec_sli_d, uint64_t)
+
+#undef DO_SLI
+
/*
* Convert float16 to float32, raising no exceptions and
* preserving exceptional values, including SNaN.
--
2.20.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 4/6] target/arm: Remove unnecessary range check for VSHL
2020-05-01 21:13 [PATCH 0/6] target/arm: partial vector cleanup Richard Henderson
` (2 preceding siblings ...)
2020-05-01 21:13 ` [PATCH 3/6] target/arm: Create gen_gvec_{sri,sli} Richard Henderson
@ 2020-05-01 21:13 ` Richard Henderson
2020-05-01 21:13 ` [PATCH 5/6] target/arm: Tidy handle_vec_simd_shri Richard Henderson
2020-05-01 21:13 ` [PATCH 6/6] target/arm: Wrap vector compare zero GVecGen2 in GVecGen2Fn Richard Henderson
5 siblings, 0 replies; 7+ messages in thread
From: Richard Henderson @ 2020-05-01 21:13 UTC (permalink / raw)
To: qemu-devel; +Cc: peter.maydell
In 1dc8425e551, while converting to gvec, I added an extra range check
against the shift count. This was unnecessary because the encoding of
the shift count produces 0 to the element size - 1.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/translate.c | 12 ++----------
1 file changed, 2 insertions(+), 10 deletions(-)
diff --git a/target/arm/translate.c b/target/arm/translate.c
index f730eb5b75..f082384117 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -6126,16 +6126,8 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
gen_gvec_sli(size, rd_ofs, rm_ofs, shift,
vec_size, vec_size);
} else { /* VSHL */
- /* Shifts larger than the element size are
- * architecturally valid and results in zero.
- */
- if (shift >= 8 << size) {
- tcg_gen_gvec_dup_imm(size, rd_ofs,
- vec_size, vec_size, 0);
- } else {
- tcg_gen_gvec_shli(size, rd_ofs, rm_ofs, shift,
- vec_size, vec_size);
- }
+ tcg_gen_gvec_shli(size, rd_ofs, rm_ofs, shift,
+ vec_size, vec_size);
}
return 0;
}
--
2.20.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 5/6] target/arm: Tidy handle_vec_simd_shri
2020-05-01 21:13 [PATCH 0/6] target/arm: partial vector cleanup Richard Henderson
` (3 preceding siblings ...)
2020-05-01 21:13 ` [PATCH 4/6] target/arm: Remove unnecessary range check for VSHL Richard Henderson
@ 2020-05-01 21:13 ` Richard Henderson
2020-05-01 21:13 ` [PATCH 6/6] target/arm: Wrap vector compare zero GVecGen2 in GVecGen2Fn Richard Henderson
5 siblings, 0 replies; 7+ messages in thread
From: Richard Henderson @ 2020-05-01 21:13 UTC (permalink / raw)
To: qemu-devel; +Cc: peter.maydell
Now that we've converted all cases to gvec, there is quite a bit
of dead code at the end of the function. Remove it.
Sink the call to gen_gvec_fn2i to the end, loading a function
pointer within the switch statement.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/translate-a64.c | 56 ++++++++++----------------------------
1 file changed, 14 insertions(+), 42 deletions(-)
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index bc326dadda..5937069992 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -10172,16 +10172,7 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
int size = 32 - clz32(immh) - 1;
int immhb = immh << 3 | immb;
int shift = 2 * (8 << size) - immhb;
- bool accumulate = false;
- int dsize = is_q ? 128 : 64;
- int esize = 8 << size;
- int elements = dsize/esize;
- MemOp memop = size | (is_u ? 0 : MO_SIGN);
- TCGv_i64 tcg_rn = new_tmp_a64(s);
- TCGv_i64 tcg_rd = new_tmp_a64(s);
- TCGv_i64 tcg_round;
- uint64_t round_const;
- int i;
+ GVecGen2iFn *gvec_fn;
if (extract32(immh, 3, 1) && !is_q) {
unallocated_encoding(s);
@@ -10195,13 +10186,12 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
switch (opcode) {
case 0x02: /* SSRA / USRA (accumulate) */
- gen_gvec_fn2i(s, is_q, rd, rn, shift,
- is_u ? gen_gvec_usra : gen_gvec_ssra, size);
- return;
+ gvec_fn = is_u ? gen_gvec_usra : gen_gvec_ssra;
+ break;
case 0x08: /* SRI */
- gen_gvec_fn2i(s, is_q, rd, rn, shift, gen_gvec_sri, size);
- return;
+ gvec_fn = gen_gvec_sri;
+ break;
case 0x00: /* SSHR / USHR */
if (is_u) {
@@ -10209,49 +10199,31 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
/* Shift count the same size as element size produces zero. */
tcg_gen_gvec_dup_imm(size, vec_full_reg_offset(s, rd),
is_q ? 16 : 8, vec_full_reg_size(s), 0);
- } else {
- gen_gvec_fn2i(s, is_q, rd, rn, shift, tcg_gen_gvec_shri, size);
+ return;
}
+ gvec_fn = tcg_gen_gvec_shri;
} else {
/* Shift count the same size as element size produces all sign. */
if (shift == 8 << size) {
shift -= 1;
}
- gen_gvec_fn2i(s, is_q, rd, rn, shift, tcg_gen_gvec_sari, size);
+ gvec_fn = tcg_gen_gvec_sari;
}
- return;
+ break;
case 0x04: /* SRSHR / URSHR (rounding) */
- gen_gvec_fn2i(s, is_q, rd, rn, shift,
- is_u ? gen_gvec_urshr : gen_gvec_srshr, size);
- return;
+ gvec_fn = is_u ? gen_gvec_urshr : gen_gvec_srshr;
+ break;
case 0x06: /* SRSRA / URSRA (accum + rounding) */
- gen_gvec_fn2i(s, is_q, rd, rn, shift,
- is_u ? gen_gvec_ursra : gen_gvec_srsra, size);
- return;
+ gvec_fn = is_u ? gen_gvec_ursra : gen_gvec_srsra;
+ break;
default:
g_assert_not_reached();
}
- round_const = 1ULL << (shift - 1);
- tcg_round = tcg_const_i64(round_const);
-
- for (i = 0; i < elements; i++) {
- read_vec_element(s, tcg_rn, rn, i, memop);
- if (accumulate) {
- read_vec_element(s, tcg_rd, rd, i, memop);
- }
-
- handle_shri_with_rndacc(tcg_rd, tcg_rn, tcg_round,
- accumulate, is_u, size, shift);
-
- write_vec_element(s, tcg_rd, rd, i, size);
- }
- tcg_temp_free_i64(tcg_round);
-
- clear_vec_high(s, is_q, rd);
+ gen_gvec_fn2i(s, is_q, rd, rn, shift, gvec_fn, size);
}
/* SHL/SLI - Vector shift left */
--
2.20.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 6/6] target/arm: Wrap vector compare zero GVecGen2 in GVecGen2Fn
2020-05-01 21:13 [PATCH 0/6] target/arm: partial vector cleanup Richard Henderson
` (4 preceding siblings ...)
2020-05-01 21:13 ` [PATCH 5/6] target/arm: Tidy handle_vec_simd_shri Richard Henderson
@ 2020-05-01 21:13 ` Richard Henderson
5 siblings, 0 replies; 7+ messages in thread
From: Richard Henderson @ 2020-05-01 21:13 UTC (permalink / raw)
To: qemu-devel; +Cc: peter.maydell
Provide a functional interface for the vector expansion.
This fits better with the existing set of helpers that
we provide for other operations.
Macro-ize the 5 nearly identical comparisons.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/arm/translate.h | 16 ++-
target/arm/translate-a64.c | 22 ++--
target/arm/translate.c | 254 ++++++++-----------------------------
3 files changed, 74 insertions(+), 218 deletions(-)
diff --git a/target/arm/translate.h b/target/arm/translate.h
index 7a2008f0dd..20ec9cedd7 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -275,11 +275,17 @@ static inline void gen_swstep_exception(DisasContext *s, int isv, int ex)
uint64_t vfp_expand_imm(int size, uint8_t imm8);
/* Vector operations shared between ARM and AArch64. */
-extern const GVecGen2 ceq0_op[4];
-extern const GVecGen2 clt0_op[4];
-extern const GVecGen2 cgt0_op[4];
-extern const GVecGen2 cle0_op[4];
-extern const GVecGen2 cge0_op[4];
+void gen_gvec_ceq0(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_clt0(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_cgt0(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_cle0(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ uint32_t opr_sz, uint32_t max_sz);
+void gen_gvec_cge0(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ uint32_t opr_sz, uint32_t max_sz);
+
extern const GVecGen3 mla_op[4];
extern const GVecGen3 mls_op[4];
extern const GVecGen3 cmtst_op[4];
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index 5937069992..8208651394 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -594,14 +594,6 @@ static void gen_gvec_fn4(DisasContext *s, bool is_q, int rd, int rn, int rm,
is_q ? 16 : 8, vec_full_reg_size(s));
}
-/* Expand a 2-operand AdvSIMD vector operation using an op descriptor. */
-static void gen_gvec_op2(DisasContext *s, bool is_q, int rd,
- int rn, const GVecGen2 *gvec_op)
-{
- tcg_gen_gvec_2(vec_full_reg_offset(s, rd), vec_full_reg_offset(s, rn),
- is_q ? 16 : 8, vec_full_reg_size(s), gvec_op);
-}
-
/* Expand a 3-operand AdvSIMD vector operation using an op descriptor. */
static void gen_gvec_op3(DisasContext *s, bool is_q, int rd,
int rn, int rm, const GVecGen3 *gvec_op)
@@ -12327,13 +12319,21 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
}
break;
case 0x8: /* CMGT, CMGE */
- gen_gvec_op2(s, is_q, rd, rn, u ? &cge0_op[size] : &cgt0_op[size]);
+ if (u) {
+ gen_gvec_fn2(s, is_q, rd, rn, gen_gvec_cge0, size);
+ } else {
+ gen_gvec_fn2(s, is_q, rd, rn, gen_gvec_cgt0, size);
+ }
return;
case 0x9: /* CMEQ, CMLE */
- gen_gvec_op2(s, is_q, rd, rn, u ? &cle0_op[size] : &ceq0_op[size]);
+ if (u) {
+ gen_gvec_fn2(s, is_q, rd, rn, gen_gvec_cle0, size);
+ } else {
+ gen_gvec_fn2(s, is_q, rd, rn, gen_gvec_ceq0, size);
+ }
return;
case 0xa: /* CMLT */
- gen_gvec_op2(s, is_q, rd, rn, &clt0_op[size]);
+ gen_gvec_fn2(s, is_q, rd, rn, gen_gvec_clt0, size);
return;
case 0xb:
if (u) { /* ABS, NEG */
diff --git a/target/arm/translate.c b/target/arm/translate.c
index f082384117..b08c4a2527 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -3917,204 +3917,59 @@ static int do_v81_helper(DisasContext *s, gen_helper_gvec_3_ptr *fn,
return 1;
}
-static void gen_ceq0_i32(TCGv_i32 d, TCGv_i32 a)
-{
- tcg_gen_setcondi_i32(TCG_COND_EQ, d, a, 0);
- tcg_gen_neg_i32(d, d);
-}
-
-static void gen_ceq0_i64(TCGv_i64 d, TCGv_i64 a)
-{
- tcg_gen_setcondi_i64(TCG_COND_EQ, d, a, 0);
- tcg_gen_neg_i64(d, d);
-}
-
-static void gen_ceq0_vec(unsigned vece, TCGv_vec d, TCGv_vec a)
-{
- TCGv_vec zero = tcg_const_zeros_vec_matching(d);
- tcg_gen_cmp_vec(TCG_COND_EQ, vece, d, a, zero);
- tcg_temp_free_vec(zero);
-}
+#define GEN_CMP0(NAME, COND) \
+ static void gen_##NAME##0_i32(TCGv_i32 d, TCGv_i32 a) \
+ { \
+ tcg_gen_setcondi_i32(COND, d, a, 0); \
+ tcg_gen_neg_i32(d, d); \
+ } \
+ static void gen_##NAME##0_i64(TCGv_i64 d, TCGv_i64 a) \
+ { \
+ tcg_gen_setcondi_i64(COND, d, a, 0); \
+ tcg_gen_neg_i64(d, d); \
+ } \
+ static void gen_##NAME##0_vec(unsigned vece, TCGv_vec d, TCGv_vec a) \
+ { \
+ TCGv_vec zero = tcg_const_zeros_vec_matching(d); \
+ tcg_gen_cmp_vec(COND, vece, d, a, zero); \
+ tcg_temp_free_vec(zero); \
+ } \
+ void gen_gvec_##NAME##0(unsigned vece, uint32_t d, uint32_t m, \
+ uint32_t opr_sz, uint32_t max_sz) \
+ { \
+ const GVecGen2 op[4] = { \
+ { .fno = gen_helper_gvec_##NAME##0_b, \
+ .fniv = gen_##NAME##0_vec, \
+ .opt_opc = vecop_list_cmp, \
+ .vece = MO_8 }, \
+ { .fno = gen_helper_gvec_##NAME##0_h, \
+ .fniv = gen_##NAME##0_vec, \
+ .opt_opc = vecop_list_cmp, \
+ .vece = MO_16 }, \
+ { .fni4 = gen_##NAME##0_i32, \
+ .fniv = gen_##NAME##0_vec, \
+ .opt_opc = vecop_list_cmp, \
+ .vece = MO_32 }, \
+ { .fni8 = gen_##NAME##0_i64, \
+ .fniv = gen_##NAME##0_vec, \
+ .opt_opc = vecop_list_cmp, \
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64, \
+ .vece = MO_64 }, \
+ }; \
+ tcg_gen_gvec_2(d, m, opr_sz, max_sz, &op[vece]); \
+ }
static const TCGOpcode vecop_list_cmp[] = {
INDEX_op_cmp_vec, 0
};
-const GVecGen2 ceq0_op[4] = {
- { .fno = gen_helper_gvec_ceq0_b,
- .fniv = gen_ceq0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_8 },
- { .fno = gen_helper_gvec_ceq0_h,
- .fniv = gen_ceq0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_16 },
- { .fni4 = gen_ceq0_i32,
- .fniv = gen_ceq0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_32 },
- { .fni8 = gen_ceq0_i64,
- .fniv = gen_ceq0_vec,
- .opt_opc = vecop_list_cmp,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .vece = MO_64 },
-};
+GEN_CMP0(ceq, TCG_COND_EQ)
+GEN_CMP0(cle, TCG_COND_LE)
+GEN_CMP0(cge, TCG_COND_GE)
+GEN_CMP0(clt, TCG_COND_LT)
+GEN_CMP0(cgt, TCG_COND_GT)
-static void gen_cle0_i32(TCGv_i32 d, TCGv_i32 a)
-{
- tcg_gen_setcondi_i32(TCG_COND_LE, d, a, 0);
- tcg_gen_neg_i32(d, d);
-}
-
-static void gen_cle0_i64(TCGv_i64 d, TCGv_i64 a)
-{
- tcg_gen_setcondi_i64(TCG_COND_LE, d, a, 0);
- tcg_gen_neg_i64(d, d);
-}
-
-static void gen_cle0_vec(unsigned vece, TCGv_vec d, TCGv_vec a)
-{
- TCGv_vec zero = tcg_const_zeros_vec_matching(d);
- tcg_gen_cmp_vec(TCG_COND_LE, vece, d, a, zero);
- tcg_temp_free_vec(zero);
-}
-
-const GVecGen2 cle0_op[4] = {
- { .fno = gen_helper_gvec_cle0_b,
- .fniv = gen_cle0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_8 },
- { .fno = gen_helper_gvec_cle0_h,
- .fniv = gen_cle0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_16 },
- { .fni4 = gen_cle0_i32,
- .fniv = gen_cle0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_32 },
- { .fni8 = gen_cle0_i64,
- .fniv = gen_cle0_vec,
- .opt_opc = vecop_list_cmp,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .vece = MO_64 },
-};
-
-static void gen_cge0_i32(TCGv_i32 d, TCGv_i32 a)
-{
- tcg_gen_setcondi_i32(TCG_COND_GE, d, a, 0);
- tcg_gen_neg_i32(d, d);
-}
-
-static void gen_cge0_i64(TCGv_i64 d, TCGv_i64 a)
-{
- tcg_gen_setcondi_i64(TCG_COND_GE, d, a, 0);
- tcg_gen_neg_i64(d, d);
-}
-
-static void gen_cge0_vec(unsigned vece, TCGv_vec d, TCGv_vec a)
-{
- TCGv_vec zero = tcg_const_zeros_vec_matching(d);
- tcg_gen_cmp_vec(TCG_COND_GE, vece, d, a, zero);
- tcg_temp_free_vec(zero);
-}
-
-const GVecGen2 cge0_op[4] = {
- { .fno = gen_helper_gvec_cge0_b,
- .fniv = gen_cge0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_8 },
- { .fno = gen_helper_gvec_cge0_h,
- .fniv = gen_cge0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_16 },
- { .fni4 = gen_cge0_i32,
- .fniv = gen_cge0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_32 },
- { .fni8 = gen_cge0_i64,
- .fniv = gen_cge0_vec,
- .opt_opc = vecop_list_cmp,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .vece = MO_64 },
-};
-
-static void gen_clt0_i32(TCGv_i32 d, TCGv_i32 a)
-{
- tcg_gen_setcondi_i32(TCG_COND_LT, d, a, 0);
- tcg_gen_neg_i32(d, d);
-}
-
-static void gen_clt0_i64(TCGv_i64 d, TCGv_i64 a)
-{
- tcg_gen_setcondi_i64(TCG_COND_LT, d, a, 0);
- tcg_gen_neg_i64(d, d);
-}
-
-static void gen_clt0_vec(unsigned vece, TCGv_vec d, TCGv_vec a)
-{
- TCGv_vec zero = tcg_const_zeros_vec_matching(d);
- tcg_gen_cmp_vec(TCG_COND_LT, vece, d, a, zero);
- tcg_temp_free_vec(zero);
-}
-
-const GVecGen2 clt0_op[4] = {
- { .fno = gen_helper_gvec_clt0_b,
- .fniv = gen_clt0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_8 },
- { .fno = gen_helper_gvec_clt0_h,
- .fniv = gen_clt0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_16 },
- { .fni4 = gen_clt0_i32,
- .fniv = gen_clt0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_32 },
- { .fni8 = gen_clt0_i64,
- .fniv = gen_clt0_vec,
- .opt_opc = vecop_list_cmp,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .vece = MO_64 },
-};
-
-static void gen_cgt0_i32(TCGv_i32 d, TCGv_i32 a)
-{
- tcg_gen_setcondi_i32(TCG_COND_GT, d, a, 0);
- tcg_gen_neg_i32(d, d);
-}
-
-static void gen_cgt0_i64(TCGv_i64 d, TCGv_i64 a)
-{
- tcg_gen_setcondi_i64(TCG_COND_GT, d, a, 0);
- tcg_gen_neg_i64(d, d);
-}
-
-static void gen_cgt0_vec(unsigned vece, TCGv_vec d, TCGv_vec a)
-{
- TCGv_vec zero = tcg_const_zeros_vec_matching(d);
- tcg_gen_cmp_vec(TCG_COND_GT, vece, d, a, zero);
- tcg_temp_free_vec(zero);
-}
-
-const GVecGen2 cgt0_op[4] = {
- { .fno = gen_helper_gvec_cgt0_b,
- .fniv = gen_cgt0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_8 },
- { .fno = gen_helper_gvec_cgt0_h,
- .fniv = gen_cgt0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_16 },
- { .fni4 = gen_cgt0_i32,
- .fniv = gen_cgt0_vec,
- .opt_opc = vecop_list_cmp,
- .vece = MO_32 },
- { .fni8 = gen_cgt0_i64,
- .fniv = gen_cgt0_vec,
- .opt_opc = vecop_list_cmp,
- .prefer_i64 = TCG_TARGET_REG_BITS == 64,
- .vece = MO_64 },
-};
+#undef GEN_CMP0
static void gen_ssra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
{
@@ -7146,24 +7001,19 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
break;
case NEON_2RM_VCEQ0:
- tcg_gen_gvec_2(rd_ofs, rm_ofs, vec_size,
- vec_size, &ceq0_op[size]);
+ gen_gvec_ceq0(size, rd_ofs, rm_ofs, vec_size, vec_size);
break;
case NEON_2RM_VCGT0:
- tcg_gen_gvec_2(rd_ofs, rm_ofs, vec_size,
- vec_size, &cgt0_op[size]);
+ gen_gvec_cgt0(size, rd_ofs, rm_ofs, vec_size, vec_size);
break;
case NEON_2RM_VCLE0:
- tcg_gen_gvec_2(rd_ofs, rm_ofs, vec_size,
- vec_size, &cle0_op[size]);
+ gen_gvec_cle0(size, rd_ofs, rm_ofs, vec_size, vec_size);
break;
case NEON_2RM_VCGE0:
- tcg_gen_gvec_2(rd_ofs, rm_ofs, vec_size,
- vec_size, &cge0_op[size]);
+ gen_gvec_cge0(size, rd_ofs, rm_ofs, vec_size, vec_size);
break;
case NEON_2RM_VCLT0:
- tcg_gen_gvec_2(rd_ofs, rm_ofs, vec_size,
- vec_size, &clt0_op[size]);
+ gen_gvec_clt0(size, rd_ofs, rm_ofs, vec_size, vec_size);
break;
default:
--
2.20.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
end of thread, other threads:[~2020-05-01 21:19 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-05-01 21:13 [PATCH 0/6] target/arm: partial vector cleanup Richard Henderson
2020-05-01 21:13 ` [PATCH 1/6] target/arm: Create gen_gvec_[us]sra Richard Henderson
2020-05-01 21:13 ` [PATCH 2/6] target/arm: Create gen_gvec_{u,s}{rshr,rsra} Richard Henderson
2020-05-01 21:13 ` [PATCH 3/6] target/arm: Create gen_gvec_{sri,sli} Richard Henderson
2020-05-01 21:13 ` [PATCH 4/6] target/arm: Remove unnecessary range check for VSHL Richard Henderson
2020-05-01 21:13 ` [PATCH 5/6] target/arm: Tidy handle_vec_simd_shri Richard Henderson
2020-05-01 21:13 ` [PATCH 6/6] target/arm: Wrap vector compare zero GVecGen2 in GVecGen2Fn Richard Henderson
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.