From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([209.51.188.92]:36338) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1hJ0Gq-00076e-1T for qemu-devel@nongnu.org; Tue, 23 Apr 2019 14:35:57 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1hJ0Go-00075e-AA for qemu-devel@nongnu.org; Tue, 23 Apr 2019 14:35:56 -0400 Received: from mx1.redhat.com ([209.132.183.28]:38756) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1hJ0Go-00072H-0o for qemu-devel@nongnu.org; Tue, 23 Apr 2019 14:35:54 -0400 References: <20190420073442.7488-1-richard.henderson@linaro.org> <20190420073442.7488-21-richard.henderson@linaro.org> From: David Hildenbrand Message-ID: <2c311907-8954-fe86-536e-880d39307c97@redhat.com> Date: Tue, 23 Apr 2019 20:35:50 +0200 MIME-Version: 1.0 In-Reply-To: <20190420073442.7488-21-richard.henderson@linaro.org> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH 20/38] tcg: Add support for vector absolute value List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Richard Henderson , qemu-devel@nongnu.org On 20.04.19 09:34, Richard Henderson wrote: > Signed-off-by: Richard Henderson > --- > accel/tcg/tcg-runtime.h | 5 +++ > tcg/aarch64/tcg-target.h | 1 + > tcg/i386/tcg-target.h | 1 + > tcg/tcg-op-gvec.h | 2 + > tcg/tcg-opc.h | 1 + > tcg/tcg.h | 1 + > accel/tcg/tcg-runtime-gvec.c | 48 ++++++++++++++++++++++++ > tcg/tcg-op-gvec.c | 71 ++++++++++++++++++++++++++++++++++++ > tcg/tcg-op-vec.c | 31 ++++++++++++++++ > tcg/tcg.c | 2 + > tcg/README | 4 ++ > 11 files changed, 167 insertions(+) > > diff --git a/accel/tcg/tcg-runtime.h b/accel/tcg/tcg-runtime.h > index ed3ce5fd91..6d73dc2d65 100644 > --- a/accel/tcg/tcg-runtime.h > +++ b/accel/tcg/tcg-runtime.h > @@ -225,6 +225,11 @@ DEF_HELPER_FLAGS_3(gvec_neg16, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > DEF_HELPER_FLAGS_3(gvec_neg32, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > DEF_HELPER_FLAGS_3(gvec_neg64, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > > +DEF_HELPER_FLAGS_3(gvec_abs8, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > +DEF_HELPER_FLAGS_3(gvec_abs16, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > +DEF_HELPER_FLAGS_3(gvec_abs32, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > +DEF_HELPER_FLAGS_3(gvec_abs64, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > + > DEF_HELPER_FLAGS_3(gvec_not, TCG_CALL_NO_RWG, void, ptr, ptr, i32) > DEF_HELPER_FLAGS_4(gvec_and, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) > DEF_HELPER_FLAGS_4(gvec_or, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) > diff --git a/tcg/aarch64/tcg-target.h b/tcg/aarch64/tcg-target.h > index f5640a229b..21d06d928c 100644 > --- a/tcg/aarch64/tcg-target.h > +++ b/tcg/aarch64/tcg-target.h > @@ -132,6 +132,7 @@ typedef enum { > #define TCG_TARGET_HAS_orc_vec 1 > #define TCG_TARGET_HAS_not_vec 1 > #define TCG_TARGET_HAS_neg_vec 1 > +#define TCG_TARGET_HAS_abs_vec 0 > #define TCG_TARGET_HAS_shi_vec 1 > #define TCG_TARGET_HAS_shs_vec 0 > #define TCG_TARGET_HAS_shv_vec 1 > diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h > index 618aa520d2..7445f05885 100644 > --- a/tcg/i386/tcg-target.h > +++ b/tcg/i386/tcg-target.h > @@ -182,6 +182,7 @@ extern bool have_avx2; > #define TCG_TARGET_HAS_orc_vec 0 > #define TCG_TARGET_HAS_not_vec 0 > #define TCG_TARGET_HAS_neg_vec 0 > +#define TCG_TARGET_HAS_abs_vec 0 > #define TCG_TARGET_HAS_shi_vec 1 > #define TCG_TARGET_HAS_shs_vec 1 > #define TCG_TARGET_HAS_shv_vec have_avx2 > diff --git a/tcg/tcg-op-gvec.h b/tcg/tcg-op-gvec.h > index f9c6058e92..46f58febbf 100644 > --- a/tcg/tcg-op-gvec.h > +++ b/tcg/tcg-op-gvec.h > @@ -228,6 +228,8 @@ void tcg_gen_gvec_not(unsigned vece, uint32_t dofs, uint32_t aofs, > uint32_t oprsz, uint32_t maxsz); > void tcg_gen_gvec_neg(unsigned vece, uint32_t dofs, uint32_t aofs, > uint32_t oprsz, uint32_t maxsz); > +void tcg_gen_gvec_abs(unsigned vece, uint32_t dofs, uint32_t aofs, > + uint32_t oprsz, uint32_t maxsz); > > void tcg_gen_gvec_add(unsigned vece, uint32_t dofs, uint32_t aofs, > uint32_t bofs, uint32_t oprsz, uint32_t maxsz); > diff --git a/tcg/tcg-opc.h b/tcg/tcg-opc.h > index 4bf71f261f..4a2dd116eb 100644 > --- a/tcg/tcg-opc.h > +++ b/tcg/tcg-opc.h > @@ -225,6 +225,7 @@ DEF(add_vec, 1, 2, 0, IMPLVEC) > DEF(sub_vec, 1, 2, 0, IMPLVEC) > DEF(mul_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_mul_vec)) > DEF(neg_vec, 1, 1, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_neg_vec)) > +DEF(abs_vec, 1, 1, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_abs_vec)) > DEF(ssadd_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_sat_vec)) > DEF(usadd_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_sat_vec)) > DEF(sssub_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_sat_vec)) > diff --git a/tcg/tcg.h b/tcg/tcg.h > index 48d4d2e03e..986055fdfa 100644 > --- a/tcg/tcg.h > +++ b/tcg/tcg.h > @@ -176,6 +176,7 @@ typedef uint64_t TCGRegSet; > && !defined(TCG_TARGET_HAS_v128) \ > && !defined(TCG_TARGET_HAS_v256) > #define TCG_TARGET_MAYBE_vec 0 > +#define TCG_TARGET_HAS_abs_vec 0 > #define TCG_TARGET_HAS_neg_vec 0 > #define TCG_TARGET_HAS_not_vec 0 > #define TCG_TARGET_HAS_andc_vec 0 > diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c > index 7b88f5590c..dd08095773 100644 > --- a/accel/tcg/tcg-runtime-gvec.c > +++ b/accel/tcg/tcg-runtime-gvec.c > @@ -398,6 +398,54 @@ void HELPER(gvec_neg64)(void *d, void *a, uint32_t desc) > clear_high(d, oprsz, desc); > } > > +void HELPER(gvec_abs8)(void *d, void *a, uint32_t desc) > +{ > + intptr_t oprsz = simd_oprsz(desc); > + intptr_t i; > + > + for (i = 0; i < oprsz; i += sizeof(int8_t)) { > + int8_t aa = *(int8_t *)(a + i); > + *(int8_t *)(d + i) = aa < 0 ? -aa : aa; > + } > + clear_high(d, oprsz, desc); > +} > + > +void HELPER(gvec_abs16)(void *d, void *a, uint32_t desc) > +{ > + intptr_t oprsz = simd_oprsz(desc); > + intptr_t i; > + > + for (i = 0; i < oprsz; i += sizeof(int16_t)) { > + int16_t aa = *(int16_t *)(a + i); > + *(int16_t *)(d + i) = aa < 0 ? -aa : aa; > + } > + clear_high(d, oprsz, desc); > +} > + > +void HELPER(gvec_abs32)(void *d, void *a, uint32_t desc) > +{ > + intptr_t oprsz = simd_oprsz(desc); > + intptr_t i; > + > + for (i = 0; i < oprsz; i += sizeof(int32_t)) { > + int32_t aa = *(int32_t *)(a + i); > + *(int32_t *)(d + i) = aa < 0 ? -aa : aa; > + } > + clear_high(d, oprsz, desc); > +} > + > +void HELPER(gvec_abs64)(void *d, void *a, uint32_t desc) > +{ > + intptr_t oprsz = simd_oprsz(desc); > + intptr_t i; > + > + for (i = 0; i < oprsz; i += sizeof(int64_t)) { > + int64_t aa = *(int64_t *)(a + i); > + *(int64_t *)(d + i) = aa < 0 ? -aa : aa; > + } > + clear_high(d, oprsz, desc); > +} > + > void HELPER(gvec_mov)(void *d, void *a, uint32_t desc) > { > intptr_t oprsz = simd_oprsz(desc); > diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c > index 4eb0747ddd..87d5a01cc9 100644 > --- a/tcg/tcg-op-gvec.c > +++ b/tcg/tcg-op-gvec.c > @@ -88,6 +88,14 @@ static bool tcg_can_emit_vecop_list(const TCGOpcode *list, > continue; > } > break; > + case INDEX_op_abs_vec: > + if (tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece) > + && (tcg_can_emit_vec_op(INDEX_op_smax_vec, type, vece) > 0 > + || tcg_can_emit_vec_op(INDEX_op_sari_vec, type, vece) > 0 > + || tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece))) { > + continue; > + } > + break; > default: > break; > } > @@ -2239,6 +2247,69 @@ void tcg_gen_gvec_neg(unsigned vece, uint32_t dofs, uint32_t aofs, > tcg_gen_gvec_2(dofs, aofs, oprsz, maxsz, &g[vece]); > } > > +static void gen_absv_mask(TCGv_i64 d, TCGv_i64 b, unsigned vece) > +{ > + TCGv_i64 t = tcg_temp_new_i64(); > + int nbit = 8 << vece; > + > + /* Create -1 for each negative element. */ > + tcg_gen_shri_i64(t, b, nbit - 1); > + tcg_gen_andi_i64(t, t, dup_const(vece, 1)); > + tcg_gen_muli_i64(t, t, (1 << nbit) - 1); > + > + /* > + * Invert (via xor -1) and add one (via sub -1). > + * Because of the ordering the msb is cleared, > + * so we never have carry into the next element. > + */ > + tcg_gen_xor_i64(d, b, t); > + tcg_gen_sub_i64(d, d, t); > + > + tcg_temp_free_i64(t); > +} > + > +static void tcg_gen_vec_abs8_i64(TCGv_i64 d, TCGv_i64 b) > +{ > + gen_absv_mask(d, b, MO_8); > +} > + > +static void tcg_gen_vec_abs16_i64(TCGv_i64 d, TCGv_i64 b) > +{ > + gen_absv_mask(d, b, MO_16); > +} > + > +void tcg_gen_gvec_abs(unsigned vece, uint32_t dofs, uint32_t aofs, > + uint32_t oprsz, uint32_t maxsz) > +{ > + static const TCGOpcode vecop_list[] = { INDEX_op_abs_vec, 0 }; > + static const GVecGen2 g[4] = { > + { .fni8 = tcg_gen_vec_abs8_i64, > + .fniv = tcg_gen_abs_vec, > + .fno = gen_helper_gvec_abs8, > + .opt_opc = vecop_list, > + .vece = MO_8 }, > + { .fni8 = tcg_gen_vec_abs16_i64, > + .fniv = tcg_gen_abs_vec, > + .fno = gen_helper_gvec_abs16, > + .opt_opc = vecop_list, > + .vece = MO_16 }, > + { .fni4 = tcg_gen_abs_i32, > + .fniv = tcg_gen_abs_vec, > + .fno = gen_helper_gvec_abs32, > + .opt_opc = vecop_list, > + .vece = MO_32 }, > + { .fni8 = tcg_gen_abs_i64, > + .fniv = tcg_gen_abs_vec, > + .fno = gen_helper_gvec_abs64, > + .opt_opc = vecop_list, > + .prefer_i64 = TCG_TARGET_REG_BITS == 64, > + .vece = MO_64 }, > + }; > + > + tcg_debug_assert(vece <= MO_64); > + tcg_gen_gvec_2(dofs, aofs, oprsz, maxsz, &g[vece]); > +} > + > void tcg_gen_gvec_and(unsigned vece, uint32_t dofs, uint32_t aofs, > uint32_t bofs, uint32_t oprsz, uint32_t maxsz) > { > diff --git a/tcg/tcg-op-vec.c b/tcg/tcg-op-vec.c > index 5ec8e0b1a0..b7f21145bb 100644 > --- a/tcg/tcg-op-vec.c > +++ b/tcg/tcg-op-vec.c > @@ -345,6 +345,37 @@ void tcg_gen_neg_vec(unsigned vece, TCGv_vec r, TCGv_vec a) > tcg_swap_vecop_list(hold_list); > } > > +void tcg_gen_abs_vec(unsigned vece, TCGv_vec r, TCGv_vec a) > +{ > + const TCGOpcode *hold_list; > + > + tcg_assert_listed_vecop(INDEX_op_abs_vec); > + hold_list = tcg_swap_vecop_list(NULL); > + > + if (!do_op2(vece, r, a, INDEX_op_abs_vec)) { > + TCGType type = tcgv_vec_temp(r)->base_type; > + TCGv_vec t = tcg_temp_new_vec(type); > + > + tcg_debug_assert(tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece)); > + if (tcg_can_emit_vec_op(INDEX_op_smax_vec, type, vece) > 0) { > + tcg_gen_neg_vec(vece, t, a); > + tcg_gen_smax_vec(vece, r, a, t); > + } else { > + if (tcg_can_emit_vec_op(INDEX_op_sari_vec, type, vece) > 0) { > + tcg_gen_sari_vec(vece, t, a, (8 << vece) - 1); > + } else { > + do_dupi_vec(t, MO_REG, 0); > + tcg_gen_cmp_vec(TCG_COND_LT, vece, t, a, t); > + } > + tcg_gen_xor_vec(vece, r, a, t); > + tcg_gen_sub_vec(vece, r, r, t); > + } > + > + tcg_temp_free_vec(t); > + } > + tcg_swap_vecop_list(hold_list); > +} > + > static void do_shifti(TCGOpcode opc, unsigned vece, > TCGv_vec r, TCGv_vec a, int64_t i) > { > diff --git a/tcg/tcg.c b/tcg/tcg.c > index 0b0b228bb5..86a95a636b 100644 > --- a/tcg/tcg.c > +++ b/tcg/tcg.c > @@ -1621,6 +1621,8 @@ bool tcg_op_supported(TCGOpcode op) > return have_vec && TCG_TARGET_HAS_not_vec; > case INDEX_op_neg_vec: > return have_vec && TCG_TARGET_HAS_neg_vec; > + case INDEX_op_abs_vec: > + return have_vec && TCG_TARGET_HAS_abs_vec; > case INDEX_op_andc_vec: > return have_vec && TCG_TARGET_HAS_andc_vec; > case INDEX_op_orc_vec: > diff --git a/tcg/README b/tcg/README > index c30e5418a6..cbdfd3b6bc 100644 > --- a/tcg/README > +++ b/tcg/README > @@ -561,6 +561,10 @@ E.g. VECL=1 -> 64 << 1 -> v128, and VECE=2 -> 1 << 2 -> i32. > > Similarly, v0 = -v1. > > +* abs_vec v0, v1 > + > + Similarly, v0 = v1 < 0 ? -v1 : v1, in elements across the vector. > + > * smin_vec: > * umin_vec: > > Thanks, using this for VECTOR LOAD POSITIVE on s390x. -- Thanks, David / dhildenb