[07/23] target/arm: Implement SVE Integer Binary Arithmetic - Predicated Group

Message ID	20171218174552.18871-8-richard.henderson@linaro.org
State	New
Headers	show Delivered-To: patch@linaro.org Received-SPF: pass (google.com: domain of qemu-devel-bounces+patch=linaro.org@nongnu.org designates 2001:4830:134:3::11 as permitted sender) client-ip=2001:4830:134:3::11; From: Richard Henderson <richard.henderson@linaro.org> To: qemu-devel@nongnu.org Date: Mon, 18 Dec 2017 09:45:36 -0800 Message-Id: <20171218174552.18871-8-richard.henderson@linaro.org> In-Reply-To: <20171218174552.18871-1-richard.henderson@linaro.org> References: <20171218174552.18871-1-richard.henderson@linaro.org> Subject: [Qemu-devel] [PATCH 07/23] target/arm: Implement SVE Integer Binary Arithmetic - Predicated Group Precedence: list Cc: peter.maydell@linaro.org, qemu-arm@nongnu.org Errors-To: qemu-devel-bounces+patch=linaro.org@nongnu.org Sender: "Qemu-devel" <qemu-devel-bounces+patch=linaro.org@nongnu.org>
Series	target/arm: decode generator and initial sve patches \| expand [RFC,00/23] target/arm: decode generator and initial sve patches [01/23] scripts: Add decodetree.py [02/23] target/arm: Add SVE decode skeleton [03/23] target/arm: Implement SVE Bitwise Logical - Unpredicated Group [04/23] target/arm: Implement PTRUE, PFALSE, SETFFR [05/23] target/arm: Implement SVE predicate logical operations [06/23] target/arm: Implement SVE load vector/predicate [07/23] target/arm: Implement SVE Integer Binary Arithmetic - Predicated Group [08/23] target/arm: Handle SVE registers in write_fp_dreg [09/23] target/arm: Handle SVE registers when using clear_vec_high [10/23] target/arm: Implement SVE Integer Reduction Group [11/23] target/arm: Implement SVE bitwise shift by immediate (predicated) [12/23] target/arm: Implement SVE bitwise shift by vector (predicated) [13/23] target/arm: Implement SVE bitwise shift by wide elements (predicated) [14/23] target/arm: Implement SVE Integer Arithmetic - Unary Predicated Group [15/23] target/arm: Implement SVE Integer Multiply-Add Group [16/23] target/arm: Implement SVE Integer Arithmetic - Unpredicated Group [17/23] target/arm: Implement SVE Index Generation Group [18/23] target/arm: Implement SVE Stack Allocation Group [19/23] target/arm: Implement SVE Bitwise Shift - Unpredicated Group [20/23] target/arm: Implement SVE Compute Vector Address Group [21/23] target/arm: Implement SVE floating-point exponential accelerator [22/23] target/arm: Implement SVE floating-point trig select coefficient [23/23] target/arm: Implement SVE Element Count Group, register destinations

diff --git a/target/arm/helper-sve.h b/target/arm/helper-sve.h index 8b382a962d..964b15b104 100644 --- a/target/arm/helper-sve.h +++ b/target/arm/helper-sve.h @@ -17,6 +17,151 @@ * License along with this library; if not, see <http://www.gnu.org/licenses/>. */ +DEF_HELPER_FLAGS_5(sve_and_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_and_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_and_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_and_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_eor_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_eor_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_eor_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_eor_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_orr_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_orr_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_orr_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_orr_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_bic_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_bic_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_bic_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_bic_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_add_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_add_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_add_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_add_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_sub_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_sub_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_sub_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_sub_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_smax_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smax_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smax_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smax_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_umax_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umax_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umax_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umax_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_smin_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smin_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smin_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smin_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_umin_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umin_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umin_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umin_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_sabd_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_sabd_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_sabd_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_sabd_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_uabd_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_uabd_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_uabd_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_uabd_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_mul_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_mul_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_mul_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_mul_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_smulh_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smulh_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smulh_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_smulh_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_umulh_zpzz_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umulh_zpzz_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umulh_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_umulh_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_sdiv_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_sdiv_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve_udiv_zpzz_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sve_udiv_zpzz_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_5(sve_and_pred, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sve_bic_pred, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sve_eor_pred, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/sve_helper.c b/target/arm/sve_helper.c index a605e623f7..b617ea2c04 100644 --- a/target/arm/sve_helper.c +++ b/target/arm/sve_helper.c @@ -28,13 +28,17 @@ /* Note that vector data is stored in host-endian 64-bit chunks, so addressing units smaller than that needs a host-endian fixup. */ #ifdef HOST_WORDS_BIGENDIAN -#define H1(x) ((x) ^ 7) -#define H2(x) ((x) ^ 3) -#define H4(x) ((x) ^ 1) +#define H1(x) ((x) ^ 7) +#define H1_2(x) ((x) ^ 6) +#define H1_4(x) ((x) ^ 4) +#define H2(x) ((x) ^ 3) +#define H4(x) ((x) ^ 1) #else -#define H1(x) (x) -#define H2(x) (x) -#define H4(x) (x) +#define H1(x) (x) +#define H1_2(x) (x) +#define H1_4(x) (x) +#define H2(x) (x) +#define H4(x) (x) #endif @@ -117,7 +121,7 @@ LOGICAL_PRED_FLAGS(sve_nands_pred, DO_NAND) #undef LOGICAL_PRED #undef LOGICAL_PRED_FLAGS -#undef DO_ADD +#undef DO_AND #undef DO_BIC #undef DO_EOR #undef DO_ORR @@ -126,6 +130,191 @@ LOGICAL_PRED_FLAGS(sve_nands_pred, DO_NAND) #undef DO_NAND #undef DO_SEL +/* Fully general three-operand expander, controlled by a predicate. + * This is complicated by the host-endian storage of the register file. + */ +/* ??? I don't expect the compiler could ever vectorize this itself. + * With some tables we can convert bit masks to byte masks, and with + * extra care wrt byte/word ordering we could use gcc generic vectors + * and do 16 bytes at a time. + */ +#define DO_ZPZZ(NAME, TYPE, H, OP) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, void *vg, uint32_t desc) \ +{ \ + intptr_t iv = 0, ib = 0, opr_sz = simd_oprsz(desc); \ + for (iv = ib = 0; iv < opr_sz; iv += 16, ib += 2) { \ + uint16_t pg = *(uint16_t *)(vg + H2(ib)); \ + intptr_t i = 0; \ + do { \ + if (pg & 1) { \ + TYPE nn = *(TYPE *)(vn + iv + H(i)); \ + TYPE mm = *(TYPE *)(vm + iv + H(i)); \ + *(TYPE *)(vd + iv + H(i)) = OP(nn, mm); \ + } \ + i += sizeof(TYPE), pg >>= sizeof(TYPE); \ + } while (pg); \ + } \ +} + +/* Similarly, specialized for 64-bit operands. */ +#define DO_ZPZZ_D(NAME, TYPE, OP) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, void *vg, uint32_t desc) \ +{ \ + intptr_t i, opr_sz = simd_oprsz(desc) / 8; \ + TYPE *d = vd, *n = vn, *m = vm; \ + uint8_t *pg = vg; \ + for (i = 0; i < opr_sz; i += 1) { \ + if (pg[H1(i)] & 1) { \ + TYPE nn = n[i], mm = m[i]; \ + d[i] = OP(nn, mm); \ + } \ + } \ +} + +#define DO_AND(N, M) (N & M) +#define DO_EOR(N, M) (N ^ M) +#define DO_ORR(N, M) (N | M) +#define DO_BIC(N, M) (N &~ M) + +DO_ZPZZ(sve_and_zpzz_b, uint8_t, H1, DO_AND) +DO_ZPZZ(sve_orr_zpzz_b, uint8_t, H1, DO_ORR) +DO_ZPZZ(sve_eor_zpzz_b, uint8_t, H1, DO_EOR) +DO_ZPZZ(sve_bic_zpzz_b, uint8_t, H1, DO_BIC) + +DO_ZPZZ(sve_and_zpzz_h, uint16_t, H1_2, DO_AND) +DO_ZPZZ(sve_orr_zpzz_h, uint16_t, H1_2, DO_ORR) +DO_ZPZZ(sve_eor_zpzz_h, uint16_t, H1_2, DO_EOR) +DO_ZPZZ(sve_bic_zpzz_h, uint16_t, H1_2, DO_BIC) + +DO_ZPZZ(sve_and_zpzz_s, uint32_t, H1_4, DO_AND) +DO_ZPZZ(sve_orr_zpzz_s, uint32_t, H1_4, DO_ORR) +DO_ZPZZ(sve_eor_zpzz_s, uint32_t, H1_4, DO_EOR) +DO_ZPZZ(sve_bic_zpzz_s, uint32_t, H1_4, DO_BIC) + +DO_ZPZZ_D(sve_and_zpzz_d, uint64_t, DO_AND) +DO_ZPZZ_D(sve_orr_zpzz_d, uint64_t, DO_ORR) +DO_ZPZZ_D(sve_eor_zpzz_d, uint64_t, DO_EOR) +DO_ZPZZ_D(sve_bic_zpzz_d, uint64_t, DO_BIC) + +#undef DO_AND +#undef DO_ORR +#undef DO_EOR +#undef DO_BIC + +#define DO_ADD(N, M) (N + M) +#define DO_SUB(N, M) (N - M) + +DO_ZPZZ(sve_add_zpzz_b, uint8_t, H1, DO_ADD) +DO_ZPZZ(sve_sub_zpzz_b, uint8_t, H1, DO_SUB) + +DO_ZPZZ(sve_add_zpzz_h, uint16_t, H1_2, DO_ADD) +DO_ZPZZ(sve_sub_zpzz_h, uint16_t, H1_2, DO_SUB) + +DO_ZPZZ(sve_add_zpzz_s, uint32_t, H1_4, DO_ADD) +DO_ZPZZ(sve_sub_zpzz_s, uint32_t, H1_4, DO_SUB) + +DO_ZPZZ_D(sve_add_zpzz_d, uint64_t, DO_ADD) +DO_ZPZZ_D(sve_sub_zpzz_d, uint64_t, DO_SUB) + +#undef DO_ADD +#undef DO_SUB + +#define DO_MAX(N, M) ((N) >= (M) ? (N) : (M)) +#define DO_MIN(N, M) ((N) >= (M) ? (M) : (N)) +#define DO_ABD(N, M) ((N) >= (M) ? (N) - (M) : (M) - (N)) + +DO_ZPZZ(sve_smax_zpzz_b, int8_t, H1, DO_MAX) +DO_ZPZZ(sve_umax_zpzz_b, uint8_t, H1, DO_MAX) +DO_ZPZZ(sve_smin_zpzz_b, int8_t, H1, DO_MIN) +DO_ZPZZ(sve_umin_zpzz_b, uint8_t, H1, DO_MIN) +DO_ZPZZ(sve_sabd_zpzz_b, int8_t, H1, DO_ABD) +DO_ZPZZ(sve_uabd_zpzz_b, uint8_t, H1, DO_ABD) + +DO_ZPZZ(sve_smax_zpzz_h, int16_t, H1_2, DO_MAX) +DO_ZPZZ(sve_umax_zpzz_h, uint16_t, H1_2, DO_MAX) +DO_ZPZZ(sve_smin_zpzz_h, int16_t, H1_2, DO_MIN) +DO_ZPZZ(sve_umin_zpzz_h, uint16_t, H1_2, DO_MIN) +DO_ZPZZ(sve_sabd_zpzz_h, int16_t, H1_2, DO_ABD) +DO_ZPZZ(sve_uabd_zpzz_h, uint16_t, H1_2, DO_ABD) + +DO_ZPZZ(sve_smax_zpzz_s, int32_t, H1_4, DO_MAX) +DO_ZPZZ(sve_umax_zpzz_s, uint32_t, H1_4, DO_MAX) +DO_ZPZZ(sve_smin_zpzz_s, int32_t, H1_4, DO_MIN) +DO_ZPZZ(sve_umin_zpzz_s, uint32_t, H1_4, DO_MIN) +DO_ZPZZ(sve_sabd_zpzz_s, int32_t, H1_4, DO_ABD) +DO_ZPZZ(sve_uabd_zpzz_s, uint32_t, H1_4, DO_ABD) + +DO_ZPZZ_D(sve_smax_zpzz_d, int64_t, DO_MAX) +DO_ZPZZ_D(sve_umax_zpzz_d, uint64_t, DO_MAX) +DO_ZPZZ_D(sve_smin_zpzz_d, int64_t, DO_MIN) +DO_ZPZZ_D(sve_umin_zpzz_d, uint64_t, DO_MIN) +DO_ZPZZ_D(sve_sabd_zpzz_d, int64_t, DO_ABD) +DO_ZPZZ_D(sve_uabd_zpzz_d, uint64_t, DO_ABD) + +#undef DO_MAX +#undef DO_MIN +#undef DO_ABD + +#define DO_MUL(N, M) (N * M) +#define DO_DIV(N, M) (M ? N / M : 0) + +/* Because the computation type is at least twice as large as required, + these work for both signed and unsigned source types. */ +static inline uint8_t do_mulh_b(int32_t n, int32_t m) +{ + return (n * m) >> 8; +} + +static inline uint16_t do_mulh_h(int32_t n, int32_t m) +{ + return (n * m) >> 16; +} + +static inline uint32_t do_mulh_s(int64_t n, int64_t m) +{ + return (n * m) >> 32; +} + +static inline uint64_t do_smulh_d(uint64_t n, uint64_t m) +{ + uint64_t lo, hi; + muls64(&lo, &hi, n, m); + return hi; +} + +static inline uint64_t do_umulh_d(uint64_t n, uint64_t m) +{ + uint64_t lo, hi; + mulu64(&lo, &hi, n, m); + return hi; +} + +DO_ZPZZ(sve_mul_zpzz_b, uint8_t, H1, DO_MUL) +DO_ZPZZ(sve_smulh_zpzz_b, int8_t, H1, do_mulh_b) +DO_ZPZZ(sve_umulh_zpzz_b, uint8_t, H1, do_mulh_b) + +DO_ZPZZ(sve_mul_zpzz_h, uint16_t, H1_2, DO_MUL) +DO_ZPZZ(sve_smulh_zpzz_h, int16_t, H1_2, do_mulh_h) +DO_ZPZZ(sve_umulh_zpzz_h, uint16_t, H1_2, do_mulh_h) + +DO_ZPZZ(sve_mul_zpzz_s, uint32_t, H1_4, DO_MUL) +DO_ZPZZ(sve_smulh_zpzz_s, int32_t, H1_4, do_mulh_s) +DO_ZPZZ(sve_umulh_zpzz_s, uint32_t, H1_4, do_mulh_s) +DO_ZPZZ(sve_sdiv_zpzz_s, int32_t, H1_4, DO_DIV) +DO_ZPZZ(sve_udiv_zpzz_s, uint32_t, H1_4, DO_DIV) + +DO_ZPZZ_D(sve_mul_zpzz_d, uint64_t, DO_MUL) +DO_ZPZZ_D(sve_smulh_zpzz_d, uint64_t, do_smulh_d) +DO_ZPZZ_D(sve_umulh_zpzz_d, uint64_t, do_umulh_d) +DO_ZPZZ_D(sve_sdiv_zpzz_d, int64_t, DO_DIV) +DO_ZPZZ_D(sve_udiv_zpzz_d, uint64_t, DO_DIV) + +#undef DO_MUL +#undef DO_DIV + +#undef DO_ZPZZ +#undef DO_ZPZZ_D + void HELPER(sve_ldr)(CPUARMState *env, void *d, target_ulong addr, uint32_t len) { intptr_t i, len_align = QEMU_ALIGN_DOWN(len, 8); diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c index 0e988c03aa..d8b34020bb 100644 --- a/target/arm/translate-sve.c +++ b/target/arm/translate-sve.c @@ -126,6 +126,81 @@ void trans_BIC_zzz(DisasContext *s, arg_BIC_zzz *a, uint32_t insn) do_zzz_genfn(s, a, tcg_gen_gvec_andc); } +static void do_zpzz_ool(DisasContext *s, arg_rprr_esz *a, gen_helper_gvec_4 *fn) +{ + unsigned vsz = size_for_gvec(vec_full_reg_size(s)); + tcg_gen_gvec_4_ool(vec_full_reg_offset(s, a->rd), + vec_full_reg_offset(s, a->rn), + vec_full_reg_offset(s, a->rm), + pred_full_reg_offset(s, a->pg), + vsz, vsz, 0, fn); +} + +#define DO_ZPZZ(NAME, name) \ +void trans_##NAME##_zpzz(DisasContext *s, arg_rprr_esz *a, uint32_t insn) \ +{ \ + static gen_helper_gvec_4 * const fns[4] = { \ + gen_helper_sve_##name##_zpzz_b, gen_helper_sve_##name##_zpzz_h, \ + gen_helper_sve_##name##_zpzz_s, gen_helper_sve_##name##_zpzz_d, \ + }; \ + do_zpzz_ool(s, a, fns[a->esz]); \ +} + +DO_ZPZZ(AND, and) +DO_ZPZZ(EOR, eor) +DO_ZPZZ(ORR, orr) +DO_ZPZZ(BIC, bic) + +DO_ZPZZ(ADD, add) +DO_ZPZZ(SUB, sub) + +DO_ZPZZ(SMAX, smax) +DO_ZPZZ(UMAX, umax) +DO_ZPZZ(SMIN, smin) +DO_ZPZZ(UMIN, umin) +DO_ZPZZ(SABD, sabd) +DO_ZPZZ(UABD, uabd) + +DO_ZPZZ(MUL, mul) +DO_ZPZZ(SMULH, smulh) +DO_ZPZZ(UMULH, umulh) + +void trans_SDIV_zpzz(DisasContext *s, arg_rprr_esz *a, uint32_t insn) +{ + gen_helper_gvec_4 *fn; + switch (a->esz) { + default: + unallocated_encoding(s); + return; + case 2: + fn = gen_helper_sve_sdiv_zpzz_s; + break; + case 3: + fn = gen_helper_sve_sdiv_zpzz_d; + break; + }; + do_zpzz_ool(s, a, fn); +} + +void trans_UDIV_zpzz(DisasContext *s, arg_rprr_esz *a, uint32_t insn) +{ + gen_helper_gvec_4 *fn; + switch (a->esz) { + default: + unallocated_encoding(s); + return; + case 2: + fn = gen_helper_sve_udiv_zpzz_s; + break; + case 3: + fn = gen_helper_sve_udiv_zpzz_d; + break; + }; + do_zpzz_ool(s, a, fn); +} + +#undef DO_ZPZZ + static uint64_t pred_esz_mask[4] = { 0xffffffffffffffffull, 0x5555555555555555ull, 0x1111111111111111ull, 0x0101010101010101ull diff --git a/target/arm/sve.def b/target/arm/sve.def index d1172296e0..3bb4faaf89 100644 --- a/target/arm/sve.def +++ b/target/arm/sve.def @@ -24,6 +24,10 @@ %imm9_16_10 16:s6 10:3 +# Either a copy of rd (at bit 0), or a different source +# as propagated via the MOVPRFX instruction. +%reg_movprfx 0:5 + ########################################################################### # Named attribute sets. These are used to make nice(er) names # when creating helpers common to those for the individual @@ -44,6 +48,10 @@ # Three prediate operand, with governing predicate, unused vector element size @pd_pg_pn_pm ........ .... rm:4 .. pg:4 . rn:4 . rd:4 &rprr_esz esz=0 +# Two register operand, with governing predicate, vector element size +@rdn_pg_rm_esz ........ esz:2 ... ... ... pg:3 rm:5 rd:5 &rprr_esz rn=%reg_movprfx +@rdm_pg_rn_esz ........ esz:2 ... ... ... pg:3 rn:5 rd:5 &rprr_esz rm=%reg_movprfx + # Basic Load/Store with 9-bit immediate offset @pd_rn_i9 ........ ........ ...... rn:5 . rd:4 &rri imm=%imm9_16_10 @rd_rn_i9 ........ ........ ...... rn:5 rd:5 &rri imm=%imm9_16_10 @@ -51,6 +59,37 @@ ########################################################################### # Instruction patterns. Grouped according to the SVE encodingindex.xhtml. +### SVE Integer Arithmetic - Binary Predicated Group + +# SVE bitwise logical vector operations (predicated) +ORR_zpzz 00000100 .. 011 000 000 ... ..... ..... @rdn_pg_rm_esz +EOR_zpzz 00000100 .. 011 001 000 ... ..... ..... @rdn_pg_rm_esz +AND_zpzz 00000100 .. 011 010 000 ... ..... ..... @rdn_pg_rm_esz +BIC_zpzz 00000100 .. 011 011 000 ... ..... ..... @rdn_pg_rm_esz + +# SVE integer add/subtract vectors (predicated) +ADD_zpzz 00000100 .. 000 000 000 ... ..... ..... @rdn_pg_rm_esz +SUB_zpzz 00000100 .. 000 001 000 ... ..... ..... @rdn_pg_rm_esz +SUB_zpzz 00000100 .. 000 011 000 ... ..... ..... @rdm_pg_rn_esz # SUBR + +# SVE integer min/max/difference (predicated) +SMAX_zpzz 00000100 .. 001 000 000 ... ..... ..... @rdn_pg_rm_esz +UMAX_zpzz 00000100 .. 001 001 000 ... ..... ..... @rdn_pg_rm_esz +SMIN_zpzz 00000100 .. 001 010 000 ... ..... ..... @rdn_pg_rm_esz +UMIN_zpzz 00000100 .. 001 011 000 ... ..... ..... @rdn_pg_rm_esz +SABD_zpzz 00000100 .. 001 100 000 ... ..... ..... @rdn_pg_rm_esz +UABD_zpzz 00000100 .. 001 101 000 ... ..... ..... @rdn_pg_rm_esz + +# SVE integer multiply/divide (predicated) +MUL_zpzz 00000100 .. 010 000 000 ... ..... ..... @rdn_pg_rm_esz +SMULH_zpzz 00000100 .. 010 010 000 ... ..... ..... @rdn_pg_rm_esz +UMULH_zpzz 00000100 .. 010 011 000 ... ..... ..... @rdn_pg_rm_esz +# Note that divide requires size >= 2; below 2 is unallocated. +SDIV_zpzz 00000100 .. 010 100 000 ... ..... ..... @rdn_pg_rm_esz +UDIV_zpzz 00000100 .. 010 101 000 ... ..... ..... @rdn_pg_rm_esz +SDIV_zpzz 00000100 .. 010 110 000 ... ..... ..... @rdm_pg_rn_esz # SDIVR +UDIV_zpzz 00000100 .. 010 111 000 ... ..... ..... @rdm_pg_rn_esz # UDIVR + ### SVE Logical - Unpredicated Group # SVE bitwise logical operations (unpredicated)

[07/23] target/arm: Implement SVE Integer Binary Arithmetic - Predicated Group

Commit Message

Patch