[v2,25/29] target/arm: Convert VQSHL, VQSHLU to gvec

Message ID	20240909162240.647173-26-richard.henderson@linaro.org (mailing list archive)
State	New, archived
Headers	show Return-Path: <qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org> From: Richard Henderson <richard.henderson@linaro.org> To: qemu-devel@nongnu.org Cc: qemu-arm@nongnu.org Subject: [PATCH v2 25/29] target/arm: Convert VQSHL, VQSHLU to gvec Date: Mon, 9 Sep 2024 09:22:35 -0700 Message-ID: <20240909162240.647173-26-richard.henderson@linaro.org> In-Reply-To: <20240909162240.647173-1-richard.henderson@linaro.org> References: <20240909162240.647173-1-richard.henderson@linaro.org> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Received-SPF: pass client-ip=2607:f8b0:4864:20::632; envelope-from=richard.henderson@linaro.org; helo=mail-pl1-x632.google.com X-Spam_score_int: -20 X-Spam_score: -2.1 X-Spam_bar: -- X-Spam_report: (-2.1 / 5.0 requ) BAYES_00=-1.9, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_NONE=-0.0001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001 autolearn=ham autolearn_force=no X-Spam_action: no action Precedence: list Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org
Series	target/arm: AdvSIMD decodetree conversion, part 4 \| expand [v2,00/29] target/arm: AdvSIMD decodetree conversion, part 4 [v2,01/29] target/arm: Replace tcg_gen_dupi_vec with constants in gengvec.c [v2,02/29] target/arm: Replace tcg_gen_dupi_vec with constants in translate-sve.c [v2,03/29] target/arm: Use cmpsel in gen_ushl_vec [v2,04/29] target/arm: Use cmpsel in gen_sshl_vec [v2,05/29] target/arm: Use tcg_gen_extract2_i64 for EXT [v2,06/29] target/arm: Convert EXT to decodetree [v2,07/29] target/arm: Convert TBL, TBX to decodetree [v2,08/29] target/arm: Convert UZP, TRN, ZIP to decodetree [v2,09/29] target/arm: Simplify do_reduction_op [v2,10/29] target/arm: Convert ADDV, ADDLV, MAXV, *MINV to decodetree [v2,11/29] target/arm: Convert FMAXNMV, FMINNMV, FMAXV, FMINV to decodetree [v2,12/29] target/arm: Convert FMOVI (scalar, immediate) to decodetree [v2,13/29] target/arm: Convert MOVI, FMOV, ORR, BIC (vector immediate) to decodetree [v2,14/29] target/arm: Introduce gen_gvec_sshr, gen_gvec_ushr [v2,15/29] target/arm: Fix whitespace near gen_srshr64_i64 [v2,16/29] target/arm: Convert handle_vec_simd_shri to decodetree [v2,17/29] target/arm: Convert handle_vec_simd_shli to decodetree [v2,18/29] target/arm: Use {, s}extract in handle_vec_simd_wshli [v2,19/29] target/arm: Convert SSHLL, USHLL to decodetree [v2,20/29] target/arm: Push tcg_rnd into handle_shri_with_rndacc [v2,21/29] target/arm: Split out subroutines of handle_shri_with_rndacc [v2,22/29] target/arm: Convert SHRN, RSHRN to decodetree [v2,23/29] target/arm: Convert handle_scalar_simd_shri to decodetree [v2,24/29] target/arm: Convert handle_scalar_simd_shli to decodetree [v2,25/29] target/arm: Convert VQSHL, VQSHLU to gvec [v2,26/29] target/arm: Widen NeonGenNarrowEnvFn return to 64 bits [v2,27/29] target/arm: Convert SQSHL, UQSHL, SQSHLU (immediate) to decodetree [v2,28/29] target/arm: Convert vector [US]QSHRN, [US]QRSHRN, SQSHRUN to decodetree [v2,29/29] target/arm: Convert scalar [US]QSHRN, [US]QRSHRN, SQSHRUN to decodetree

diff --git a/target/arm/helper.h b/target/arm/helper.h index b463be38c5..b40589d329 100644 --- a/target/arm/helper.h +++ b/target/arm/helper.h @@ -324,6 +324,18 @@ DEF_HELPER_FLAGS_5(neon_uqrshl_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32 DEF_HELPER_FLAGS_5(neon_uqrshl_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(neon_uqrshl_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(neon_uqrshl_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshli_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshli_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshli_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshli_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_uqshli_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_uqshli_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_uqshli_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_uqshli_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshlui_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshlui_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshlui_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(neon_sqshlui_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(gvec_srshl_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(gvec_srshl_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/translate.h b/target/arm/tcg/translate.h index 45990ae292..7721c627e9 100644 --- a/target/arm/tcg/translate.h +++ b/target/arm/tcg/translate.h @@ -471,6 +471,13 @@ void gen_neon_sqrshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, void gen_neon_uqrshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz); +void gen_neon_sqshli(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + int64_t c, uint32_t opr_sz, uint32_t max_sz); +void gen_neon_uqshli(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + int64_t c, uint32_t opr_sz, uint32_t max_sz); +void gen_neon_sqshlui(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + int64_t c, uint32_t opr_sz, uint32_t max_sz); + void gen_gvec_shadd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz); void gen_gvec_uhadd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, diff --git a/target/arm/tcg/gengvec.c b/target/arm/tcg/gengvec.c index 3abdc57202..f652520b65 100644 --- a/target/arm/tcg/gengvec.c +++ b/target/arm/tcg/gengvec.c @@ -1313,6 +1313,42 @@ void gen_neon_uqrshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, opr_sz, max_sz, 0, fns[vece]); } +void gen_neon_sqshli(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + int64_t c, uint32_t opr_sz, uint32_t max_sz) +{ + static gen_helper_gvec_2_ptr * const fns[] = { + gen_helper_neon_sqshli_b, gen_helper_neon_sqshli_h, + gen_helper_neon_sqshli_s, gen_helper_neon_sqshli_d, + }; + tcg_debug_assert(vece <= MO_64); + tcg_debug_assert(c >= 0 && c <= (8 << vece)); + tcg_gen_gvec_2_ptr(rd_ofs, rn_ofs, tcg_env, opr_sz, max_sz, c, fns[vece]); +} + +void gen_neon_uqshli(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + int64_t c, uint32_t opr_sz, uint32_t max_sz) +{ + static gen_helper_gvec_2_ptr * const fns[] = { + gen_helper_neon_uqshli_b, gen_helper_neon_uqshli_h, + gen_helper_neon_uqshli_s, gen_helper_neon_uqshli_d, + }; + tcg_debug_assert(vece <= MO_64); + tcg_debug_assert(c >= 0 && c <= (8 << vece)); + tcg_gen_gvec_2_ptr(rd_ofs, rn_ofs, tcg_env, opr_sz, max_sz, c, fns[vece]); +} + +void gen_neon_sqshlui(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + int64_t c, uint32_t opr_sz, uint32_t max_sz) +{ + static gen_helper_gvec_2_ptr * const fns[] = { + gen_helper_neon_sqshlui_b, gen_helper_neon_sqshlui_h, + gen_helper_neon_sqshlui_s, gen_helper_neon_sqshlui_d, + }; + tcg_debug_assert(vece <= MO_64); + tcg_debug_assert(c >= 0 && c <= (8 << vece)); + tcg_gen_gvec_2_ptr(rd_ofs, rn_ofs, tcg_env, opr_sz, max_sz, c, fns[vece]); +} + void gen_uqadd_bhs(TCGv_i64 res, TCGv_i64 qc, TCGv_i64 a, TCGv_i64 b, MemOp esz) { uint64_t max = MAKE_64BIT_MASK(0, 8 << esz); diff --git a/target/arm/tcg/neon_helper.c b/target/arm/tcg/neon_helper.c index 082bfd88ad..739e16e441 100644 --- a/target/arm/tcg/neon_helper.c +++ b/target/arm/tcg/neon_helper.c @@ -141,6 +141,19 @@ void HELPER(name)(void *vd, void *vn, void *vm, void *venv, uint32_t desc) \ clear_tail(d, opr_sz, simd_maxsz(desc)); \ } +#define NEON_GVEC_VOP2i_ENV(name, vtype) \ +void HELPER(name)(void *vd, void *vn, void *venv, uint32_t desc) \ +{ \ + intptr_t i, opr_sz = simd_oprsz(desc); \ + int imm = simd_data(desc); \ + vtype *d = vd, *n = vn; \ + CPUARMState *env = venv; \ + for (i = 0; i < opr_sz / sizeof(vtype); i++) { \ + NEON_FN(d[i], n[i], imm); \ + } \ + clear_tail(d, opr_sz, simd_maxsz(desc)); \ +} + /* Pairwise operations. */ /* For 32-bit elements each segment only contains a single element, so the elementwise and pairwise operations are the same. */ @@ -271,22 +284,26 @@ uint64_t HELPER(neon_rshl_u64)(uint64_t val, uint64_t shift) (dest = do_uqrshl_bhs(src1, (int8_t)src2, 8, false, env->vfp.qc)) NEON_VOP_ENV(qshl_u8, neon_u8, 4) NEON_GVEC_VOP2_ENV(neon_uqshl_b, uint8_t) +NEON_GVEC_VOP2i_ENV(neon_uqshli_b, uint8_t) #undef NEON_FN #define NEON_FN(dest, src1, src2) \ (dest = do_uqrshl_bhs(src1, (int8_t)src2, 16, false, env->vfp.qc)) NEON_VOP_ENV(qshl_u16, neon_u16, 2) NEON_GVEC_VOP2_ENV(neon_uqshl_h, uint16_t) +NEON_GVEC_VOP2i_ENV(neon_uqshli_h, uint16_t) #undef NEON_FN #define NEON_FN(dest, src1, src2) \ (dest = do_uqrshl_bhs(src1, (int8_t)src2, 32, false, env->vfp.qc)) NEON_GVEC_VOP2_ENV(neon_uqshl_s, uint32_t) +NEON_GVEC_VOP2i_ENV(neon_uqshli_s, uint32_t) #undef NEON_FN #define NEON_FN(dest, src1, src2) \ (dest = do_uqrshl_d(src1, (int8_t)src2, false, env->vfp.qc)) NEON_GVEC_VOP2_ENV(neon_uqshl_d, uint64_t) +NEON_GVEC_VOP2i_ENV(neon_uqshli_d, uint64_t) #undef NEON_FN uint32_t HELPER(neon_qshl_u32)(CPUARMState *env, uint32_t val, uint32_t shift) @@ -303,22 +320,26 @@ uint64_t HELPER(neon_qshl_u64)(CPUARMState *env, uint64_t val, uint64_t shift) (dest = do_sqrshl_bhs(src1, (int8_t)src2, 8, false, env->vfp.qc)) NEON_VOP_ENV(qshl_s8, neon_s8, 4) NEON_GVEC_VOP2_ENV(neon_sqshl_b, int8_t) +NEON_GVEC_VOP2i_ENV(neon_sqshli_b, int8_t) #undef NEON_FN #define NEON_FN(dest, src1, src2) \ (dest = do_sqrshl_bhs(src1, (int8_t)src2, 16, false, env->vfp.qc)) NEON_VOP_ENV(qshl_s16, neon_s16, 2) NEON_GVEC_VOP2_ENV(neon_sqshl_h, int16_t) +NEON_GVEC_VOP2i_ENV(neon_sqshli_h, int16_t) #undef NEON_FN #define NEON_FN(dest, src1, src2) \ (dest = do_sqrshl_bhs(src1, (int8_t)src2, 32, false, env->vfp.qc)) NEON_GVEC_VOP2_ENV(neon_sqshl_s, int32_t) +NEON_GVEC_VOP2i_ENV(neon_sqshli_s, int32_t) #undef NEON_FN #define NEON_FN(dest, src1, src2) \ (dest = do_sqrshl_d(src1, (int8_t)src2, false, env->vfp.qc)) NEON_GVEC_VOP2_ENV(neon_sqshl_d, int64_t) +NEON_GVEC_VOP2i_ENV(neon_sqshli_d, int64_t) #undef NEON_FN uint32_t HELPER(neon_qshl_s32)(CPUARMState *env, uint32_t val, uint32_t shift) @@ -334,11 +355,13 @@ uint64_t HELPER(neon_qshl_s64)(CPUARMState *env, uint64_t val, uint64_t shift) #define NEON_FN(dest, src1, src2) \ (dest = do_suqrshl_bhs(src1, (int8_t)src2, 8, false, env->vfp.qc)) NEON_VOP_ENV(qshlu_s8, neon_s8, 4) +NEON_GVEC_VOP2i_ENV(neon_sqshlui_b, int8_t) #undef NEON_FN #define NEON_FN(dest, src1, src2) \ (dest = do_suqrshl_bhs(src1, (int8_t)src2, 16, false, env->vfp.qc)) NEON_VOP_ENV(qshlu_s16, neon_s16, 2) +NEON_GVEC_VOP2i_ENV(neon_sqshlui_h, int16_t) #undef NEON_FN uint32_t HELPER(neon_qshlu_s32)(CPUARMState *env, uint32_t val, uint32_t shift) @@ -351,6 +374,16 @@ uint64_t HELPER(neon_qshlu_s64)(CPUARMState *env, uint64_t val, uint64_t shift) return do_suqrshl_d(val, (int8_t)shift, false, env->vfp.qc); } +#define NEON_FN(dest, src1, src2) \ + (dest = do_suqrshl_bhs(src1, (int8_t)src2, 32, false, env->vfp.qc)) +NEON_GVEC_VOP2i_ENV(neon_sqshlui_s, int32_t) +#undef NEON_FN + +#define NEON_FN(dest, src1, src2) \ + (dest = do_suqrshl_d(src1, (int8_t)src2, false, env->vfp.qc)) +NEON_GVEC_VOP2i_ENV(neon_sqshlui_d, int64_t) +#undef NEON_FN + #define NEON_FN(dest, src1, src2) \ (dest = do_uqrshl_bhs(src1, (int8_t)src2, 8, true, env->vfp.qc)) NEON_VOP_ENV(qrshl_u8, neon_u8, 4) diff --git a/target/arm/tcg/translate-neon.c b/target/arm/tcg/translate-neon.c index a31a78c347..6dd70d1c53 100644 --- a/target/arm/tcg/translate-neon.c +++ b/target/arm/tcg/translate-neon.c @@ -1101,113 +1101,9 @@ DO_2SH(VRSRA_S, gen_gvec_srsra) DO_2SH(VRSRA_U, gen_gvec_ursra) DO_2SH(VSHR_S, gen_gvec_sshr) DO_2SH(VSHR_U, gen_gvec_ushr) - -static bool do_2shift_env_64(DisasContext *s, arg_2reg_shift *a, - NeonGenTwo64OpEnvFn *fn) -{ - /* - * 2-reg-and-shift operations, size == 3 case, where the - * function needs to be passed tcg_env. - */ - TCGv_i64 constimm; - int pass; - - if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { - return false; - } - - /* UNDEF accesses to D16-D31 if they don't exist. */ - if (!dc_isar_feature(aa32_simd_r32, s) && - ((a->vd | a->vm) & 0x10)) { - return false; - } - - if ((a->vm | a->vd) & a->q) { - return false; - } - - if (!vfp_access_check(s)) { - return true; - } - - /* - * To avoid excessive duplication of ops we implement shift - * by immediate using the variable shift operations. - */ - constimm = tcg_constant_i64(dup_const(a->size, a->shift)); - - for (pass = 0; pass < a->q + 1; pass++) { - TCGv_i64 tmp = tcg_temp_new_i64(); - - read_neon_element64(tmp, a->vm, pass, MO_64); - fn(tmp, tcg_env, tmp, constimm); - write_neon_element64(tmp, a->vd, pass, MO_64); - } - return true; -} - -static bool do_2shift_env_32(DisasContext *s, arg_2reg_shift *a, - NeonGenTwoOpEnvFn *fn) -{ - /* - * 2-reg-and-shift operations, size < 3 case, where the - * helper needs to be passed tcg_env. - */ - TCGv_i32 constimm, tmp; - int pass; - - if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { - return false; - } - - /* UNDEF accesses to D16-D31 if they don't exist. */ - if (!dc_isar_feature(aa32_simd_r32, s) && - ((a->vd | a->vm) & 0x10)) { - return false; - } - - if ((a->vm | a->vd) & a->q) { - return false; - } - - if (!vfp_access_check(s)) { - return true; - } - - /* - * To avoid excessive duplication of ops we implement shift - * by immediate using the variable shift operations. - */ - constimm = tcg_constant_i32(dup_const(a->size, a->shift)); - tmp = tcg_temp_new_i32(); - - for (pass = 0; pass < (a->q ? 4 : 2); pass++) { - read_neon_element32(tmp, a->vm, pass, MO_32); - fn(tmp, tcg_env, tmp, constimm); - write_neon_element32(tmp, a->vd, pass, MO_32); - } - return true; -} - -#define DO_2SHIFT_ENV(INSN, FUNC) \ - static bool trans_##INSN##_64_2sh(DisasContext *s, arg_2reg_shift *a) \ - { \ - return do_2shift_env_64(s, a, gen_helper_neon_##FUNC##64); \ - } \ - static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \ - { \ - static NeonGenTwoOpEnvFn * const fns[] = { \ - gen_helper_neon_##FUNC##8, \ - gen_helper_neon_##FUNC##16, \ - gen_helper_neon_##FUNC##32, \ - }; \ - assert(a->size < ARRAY_SIZE(fns)); \ - return do_2shift_env_32(s, a, fns[a->size]); \ - } - -DO_2SHIFT_ENV(VQSHLU, qshlu_s) -DO_2SHIFT_ENV(VQSHL_U, qshl_u) -DO_2SHIFT_ENV(VQSHL_S, qshl_s) +DO_2SH(VQSHLU, gen_neon_sqshlui) +DO_2SH(VQSHL_U, gen_neon_uqshli) +DO_2SH(VQSHL_S, gen_neon_sqshli) static bool do_2shift_narrow_64(DisasContext *s, arg_2reg_shift *a, NeonGenTwo64OpFn *shiftfn, diff --git a/target/arm/tcg/neon-dp.decode b/target/arm/tcg/neon-dp.decode index 788578c8fa..e883c6ab58 100644 --- a/target/arm/tcg/neon-dp.decode +++ b/target/arm/tcg/neon-dp.decode @@ -291,17 +291,17 @@ VSLI_2sh 1111 001 1 1 . ...... .... 0101 . . . 1 .... @2reg_shl_s VSLI_2sh 1111 001 1 1 . ...... .... 0101 . . . 1 .... @2reg_shl_h VSLI_2sh 1111 001 1 1 . ...... .... 0101 . . . 1 .... @2reg_shl_b -VQSHLU_64_2sh 1111 001 1 1 . ...... .... 0110 . . . 1 .... @2reg_shl_d +VQSHLU_2sh 1111 001 1 1 . ...... .... 0110 . . . 1 .... @2reg_shl_d VQSHLU_2sh 1111 001 1 1 . ...... .... 0110 . . . 1 .... @2reg_shl_s VQSHLU_2sh 1111 001 1 1 . ...... .... 0110 . . . 1 .... @2reg_shl_h VQSHLU_2sh 1111 001 1 1 . ...... .... 0110 . . . 1 .... @2reg_shl_b -VQSHL_S_64_2sh 1111 001 0 1 . ...... .... 0111 . . . 1 .... @2reg_shl_d +VQSHL_S_2sh 1111 001 0 1 . ...... .... 0111 . . . 1 .... @2reg_shl_d VQSHL_S_2sh 1111 001 0 1 . ...... .... 0111 . . . 1 .... @2reg_shl_s VQSHL_S_2sh 1111 001 0 1 . ...... .... 0111 . . . 1 .... @2reg_shl_h VQSHL_S_2sh 1111 001 0 1 . ...... .... 0111 . . . 1 .... @2reg_shl_b -VQSHL_U_64_2sh 1111 001 1 1 . ...... .... 0111 . . . 1 .... @2reg_shl_d +VQSHL_U_2sh 1111 001 1 1 . ...... .... 0111 . . . 1 .... @2reg_shl_d VQSHL_U_2sh 1111 001 1 1 . ...... .... 0111 . . . 1 .... @2reg_shl_s VQSHL_U_2sh 1111 001 1 1 . ...... .... 0111 . . . 1 .... @2reg_shl_h VQSHL_U_2sh 1111 001 1 1 . ...... .... 0111 . . . 1 .... @2reg_shl_b

[v2,25/29] target/arm: Convert VQSHL, VQSHLU to gvec

Commit Message

Patch