Richard Sandiford <richard.sandif...@arm.com> writes:
> Some instructions that were previously restricted to streaming mode
> can also be used in non-streaming mode with SVE2.1. This patch adds
> support for those, as well as the usual new-extension boilerplate.
> A later patch will add the feature macro.
>
> gcc/
> * config/aarch64/aarch64-option-extensions.def (sve2p1): New extension.
Gah, just realised that I forgot to document this :( Will fix.
Richard
> * config/aarch64/aarch64-sve-builtins-sve2.def: Mark instructions
> that are common to both SVE2p1 and SME.
> * config/aarch64/aarch64.h (TARGET_SVE2p1): New macro.
> (TARGET_SVE2p1_OR_SME): Likewise.
> * config/aarch64/aarch64-sve2.md
> (@aarch64_sve_psel<BHSD_BITS>): Require TARGET_SVE2p1_OR_SME
> instead of TARGET_STREAMING.
> (*aarch64_sve_psel<BHSD_BITS>_plus): Likewise.
> (@aarch64_sve_<su>clamp<mode>): Likewise.
> (*aarch64_sve_<su>clamp<mode>_x): Likewise.
> (@aarch64_pred_<optab><mode>): Likewise.
> (@cond_<optab><mode>): Likewise.
>
> gcc/testsuite/
> * lib/target-supports.exp
> (check_effective_target_aarch64_asm_sve2p1_ok): New procedure.
> * gcc.target/aarch64/sve/clamp_1.c: New test.
> * gcc.target/aarch64/sve/clamp_2.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_s16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_s32.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_s64.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_s8.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_u16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_u32.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_u64.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/clamp_u8.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_b16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_b32.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_b64.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_b8.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_c16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_c32.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_c64.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/psel_lane_c8.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_bf16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_f16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_f32.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_f64.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_s16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_s32.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_s64.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_s8.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_u16.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_u32.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_u64.c: Likewise.
> * gcc.target/aarch64/sve2/acle/asm/revd_u8.c: Likewise.
> ---
> .../aarch64/aarch64-option-extensions.def | 2 +
> .../aarch64/aarch64-sve-builtins-sve2.def | 2 +-
> gcc/config/aarch64/aarch64-sve2.md | 12 +--
> gcc/config/aarch64/aarch64.h | 9 ++
> .../gcc.target/aarch64/sve/clamp_1.c | 40 ++++++++
> .../gcc.target/aarch64/sve/clamp_2.c | 34 +++++++
> .../aarch64/sve2/acle/asm/clamp_s16.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/clamp_s32.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/clamp_s64.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/clamp_s8.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/clamp_u16.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/clamp_u32.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/clamp_u64.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/clamp_u8.c | 46 +++++++++
> .../aarch64/sve2/acle/asm/psel_lane_b16.c | 93 +++++++++++++++++++
> .../aarch64/sve2/acle/asm/psel_lane_b32.c | 93 +++++++++++++++++++
> .../aarch64/sve2/acle/asm/psel_lane_b64.c | 84 +++++++++++++++++
> .../aarch64/sve2/acle/asm/psel_lane_b8.c | 93 +++++++++++++++++++
> .../aarch64/sve2/acle/asm/psel_lane_c16.c | 93 +++++++++++++++++++
> .../aarch64/sve2/acle/asm/psel_lane_c32.c | 93 +++++++++++++++++++
> .../aarch64/sve2/acle/asm/psel_lane_c64.c | 84 +++++++++++++++++
> .../aarch64/sve2/acle/asm/psel_lane_c8.c | 93 +++++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_bf16.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_f16.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_f32.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_f64.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_s16.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_s32.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_s64.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_s8.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_u16.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_u32.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_u64.c | 80 ++++++++++++++++
> .../aarch64/sve2/acle/asm/revd_u8.c | 80 ++++++++++++++++
> gcc/testsuite/lib/target-supports.exp | 10 ++
> 35 files changed, 2156 insertions(+), 7 deletions(-)
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/clamp_1.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/clamp_2.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_s16.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_s32.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_s64.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_s8.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_u16.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_u32.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_u64.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/clamp_u8.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_b16.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_b32.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_b64.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_b8.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_c16.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_c32.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_c64.c
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/psel_lane_c8.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_bf16.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_f16.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_f32.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_f64.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_s16.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_s32.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_s64.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_s8.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_u16.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_u32.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_u64.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/revd_u8.c
>
> diff --git a/gcc/config/aarch64/aarch64-option-extensions.def
> b/gcc/config/aarch64/aarch64-option-extensions.def
> index 8279f5a76ea..c9d419afc8f 100644
> --- a/gcc/config/aarch64/aarch64-option-extensions.def
> +++ b/gcc/config/aarch64/aarch64-option-extensions.def
> @@ -192,6 +192,8 @@ AARCH64_OPT_EXTENSION("sve2-sm4", SVE2_SM4, (SVE2, SM4),
> (), (), "svesm4")
>
> AARCH64_FMV_FEATURE("sve2-sm4", SVE_SM4, (SVE2_SM4))
>
> +AARCH64_OPT_EXTENSION("sve2p1", SVE2p1, (SVE2), (), (), "")
> +
> AARCH64_OPT_FMV_EXTENSION("sme", SME, (BF16, SVE2), (), (), "sme")
>
> AARCH64_OPT_EXTENSION("memtag", MEMTAG, (), (), (), "")
> diff --git a/gcc/config/aarch64/aarch64-sve-builtins-sve2.def
> b/gcc/config/aarch64/aarch64-sve-builtins-sve2.def
> index 12548fe39cb..5cc32aa8871 100644
> --- a/gcc/config/aarch64/aarch64-sve-builtins-sve2.def
> +++ b/gcc/config/aarch64/aarch64-sve-builtins-sve2.def
> @@ -220,7 +220,7 @@ DEF_SVE_FUNCTION (svsm4e, binary, s_unsigned, none)
> DEF_SVE_FUNCTION (svsm4ekey, binary, s_unsigned, none)
> #undef REQUIRED_EXTENSIONS
>
> -#define REQUIRED_EXTENSIONS streaming_only (0)
> +#define REQUIRED_EXTENSIONS sve_and_sme (AARCH64_FL_SVE2p1, 0)
> DEF_SVE_FUNCTION (svclamp, clamp, all_integer, none)
> DEF_SVE_FUNCTION (svpsel_lane, select_pred, all_pred_count, none)
> DEF_SVE_FUNCTION (svrevd, unary, all_data, mxz)
> diff --git a/gcc/config/aarch64/aarch64-sve2.md
> b/gcc/config/aarch64/aarch64-sve2.md
> index a7b29daeba4..fd4bd42b6d9 100644
> --- a/gcc/config/aarch64/aarch64-sve2.md
> +++ b/gcc/config/aarch64/aarch64-sve2.md
> @@ -418,7 +418,7 @@ (define_insn "@aarch64_sve_psel<BHSD_BITS>"
> (match_operand:SI 3 "register_operand" "Ucj")
> (const_int BHSD_BITS)]
> UNSPEC_PSEL))]
> - "TARGET_STREAMING"
> + "TARGET_SVE2p1_OR_SME"
> "psel\t%0, %1, %2.<bits_etype>[%w3, 0]"
> )
>
> @@ -432,7 +432,7 @@ (define_insn "*aarch64_sve_psel<BHSD_BITS>_plus"
> (match_operand:SI 4 "const_int_operand"))
> (const_int BHSD_BITS)]
> UNSPEC_PSEL))]
> - "TARGET_STREAMING
> + "TARGET_SVE2p1_OR_SME
> && UINTVAL (operands[4]) < 128 / <BHSD_BITS>"
> "psel\t%0, %1, %2.<bits_etype>[%w3, %4]"
> )
> @@ -560,7 +560,7 @@ (define_insn "@aarch64_sve_<su>clamp<mode>"
> (match_operand:SVE_FULL_I 1 "register_operand")
> (match_operand:SVE_FULL_I 2 "register_operand"))
> (match_operand:SVE_FULL_I 3 "register_operand")))]
> - "TARGET_STREAMING"
> + "TARGET_SVE2p1_OR_SME"
> {@ [cons: =0, 1, 2, 3; attrs: movprfx]
> [ w, %0, w, w; * ] <su>clamp\t%0.<Vetype>,
> %2.<Vetype>, %3.<Vetype>
> [ ?&w, w, w, w; yes ] movprfx\t%0,
> %1\;<su>clamp\t%0.<Vetype>, %2.<Vetype>, %3.<Vetype>
> @@ -580,7 +580,7 @@ (define_insn_and_split "*aarch64_sve_<su>clamp<mode>_x"
> UNSPEC_PRED_X)
> (match_operand:SVE_FULL_I 3 "register_operand"))]
> UNSPEC_PRED_X))]
> - "TARGET_STREAMING"
> + "TARGET_SVE2p1_OR_SME"
> {@ [cons: =0, 1, 2, 3; attrs: movprfx]
> [ w, %0, w, w; * ] #
> [ ?&w, w, w, w; yes ] #
> @@ -3182,7 +3182,7 @@ (define_insn "@aarch64_pred_<optab><mode>"
> [(match_operand:SVE_FULL 2 "register_operand")]
> UNSPEC_REVD_ONLY)]
> UNSPEC_PRED_X))]
> - "TARGET_STREAMING"
> + "TARGET_SVE2p1_OR_SME"
> {@ [ cons: =0 , 1 , 2 ; attrs: movprfx ]
> [ w , Upl , 0 ; * ] revd\t%0.q, %1/m, %2.q
> [ ?&w , Upl , w ; yes ] movprfx\t%0, %2\;revd\t%0.q,
> %1/m, %2.q
> @@ -3198,7 +3198,7 @@ (define_insn "@cond_<optab><mode>"
> UNSPEC_REVD_ONLY)
> (match_operand:SVE_FULL 3 "register_operand")]
> UNSPEC_SEL))]
> - "TARGET_STREAMING"
> + "TARGET_SVE2p1_OR_SME"
> {@ [ cons: =0 , 1 , 2 , 3 ; attrs: movprfx ]
> [ w , Upl , w , 0 ; * ] revd\t%0.q, %1/m, %2.q
> [ ?&w , Upl , w , w ; yes ] movprfx\t%0,
> %3\;revd\t%0.q, %1/m, %2.q
> diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h
> index d17f40ce22e..404efa16c28 100644
> --- a/gcc/config/aarch64/aarch64.h
> +++ b/gcc/config/aarch64/aarch64.h
> @@ -338,6 +338,9 @@ constexpr auto AARCH64_FL_DEFAULT_ISA_MODE
> ATTRIBUTE_UNUSED
> /* SVE2 SM4 instructions, enabled through +sve2-sm4. */
> #define TARGET_SVE2_SM4 (AARCH64_HAVE_ISA (SVE2_SM4) && TARGET_NON_STREAMING)
>
> +/* SVE2p1 instructions, enabled through +sve2p1. */
> +#define TARGET_SVE2p1 AARCH64_HAVE_ISA (SVE2p1)
> +
> /* SME instructions, enabled through +sme. Note that this does not
> imply anything about the state of PSTATE.SM; instructions that require
> SME and streaming mode should use TARGET_STREAMING instead. */
> @@ -481,6 +484,12 @@ constexpr auto AARCH64_FL_DEFAULT_ISA_MODE
> ATTRIBUTE_UNUSED
> /* fp8 instructions are enabled through +fp8. */
> #define TARGET_FP8 AARCH64_HAVE_ISA (FP8)
>
> +/* Combinatorial tests. */
> +
> +/* There's no need to check TARGET_SME for streaming or streaming-compatible
> + functions, since streaming mode itself implies SME. */
> +#define TARGET_SVE2p1_OR_SME (TARGET_SVE2p1 || TARGET_STREAMING)
> +
> /* Standard register usage. */
>
> /* 31 64-bit general purpose registers R0-R30:
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/clamp_1.c
> b/gcc/testsuite/gcc.target/aarch64/sve/clamp_1.c
> new file mode 100644
> index 00000000000..92fef098865
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/clamp_1.c
> @@ -0,0 +1,40 @@
> +// { dg-options "-O" }
> +
> +#include <arm_sve.h>
> +
> +#pragma GCC target "+sve2p1"
> +
> +#define TEST(TYPE) \
> + TYPE
> \
> + tied1_##TYPE(TYPE a, TYPE b, TYPE c)
> \
> + { \
> + return svmin_x(svptrue_b8(), svmax_x(svptrue_b8(), a, b), c); \
> + } \
> + \
> + TYPE
> \
> + tied2_##TYPE(TYPE a, TYPE b, TYPE c)
> \
> + { \
> + return svmin_x(svptrue_b8(), svmax_x(svptrue_b8(), b, a), c); \
> + }
> +
> +TEST(svint8_t)
> +TEST(svint16_t)
> +TEST(svint32_t)
> +TEST(svint64_t)
> +
> +TEST(svuint8_t)
> +TEST(svuint16_t)
> +TEST(svuint32_t)
> +TEST(svuint64_t)
> +
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.b, z1\.b, z2\.b\n} 2 } }
> */
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.h, z1\.h, z2\.h\n} 2 } }
> */
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.s, z1\.s, z2\.s\n} 2 } }
> */
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.d, z1\.d, z2\.d\n} 2 } }
> */
> +
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.b, z1\.b, z2\.b\n} 2 } }
> */
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.h, z1\.h, z2\.h\n} 2 } }
> */
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.s, z1\.s, z2\.s\n} 2 } }
> */
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.d, z1\.d, z2\.d\n} 2 } }
> */
> +
> +/* { dg-final { scan-assembler-not {\tmovprfx\t} } } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/clamp_2.c
> b/gcc/testsuite/gcc.target/aarch64/sve/clamp_2.c
> new file mode 100644
> index 00000000000..f96c0046465
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/clamp_2.c
> @@ -0,0 +1,34 @@
> +// { dg-options "-O" }
> +
> +#include <arm_sve.h>
> +
> +#pragma GCC target "+sve2p1"
> +
> +#define TEST(TYPE) \
> + TYPE
> \
> + untied_##TYPE(TYPE a, TYPE b, TYPE c, TYPE d)
> \
> + { \
> + return svmin_x(svptrue_b8(), svmax_x(svptrue_b8(), b, c), d); \
> + }
> +
> +TEST(svint8_t)
> +TEST(svint16_t)
> +TEST(svint32_t)
> +TEST(svint64_t)
> +
> +TEST(svuint8_t)
> +TEST(svuint16_t)
> +TEST(svuint32_t)
> +TEST(svuint64_t)
> +
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.b, z2\.b, z3\.b\n} 1 } }
> */
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.h, z2\.h, z3\.h\n} 1 } }
> */
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.s, z2\.s, z3\.s\n} 1 } }
> */
> +/* { dg-final { scan-assembler-times {\tsclamp\tz0\.d, z2\.d, z3\.d\n} 1 } }
> */
> +
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.b, z2\.b, z3\.b\n} 1 } }
> */
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.h, z2\.h, z3\.h\n} 1 } }
> */
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.s, z2\.s, z3\.s\n} 1 } }
> */
> +/* { dg-final { scan-assembler-times {\tuclamp\tz0\.d, z2\.d, z3\.d\n} 1 } }
> */
> +
> +/* { dg-final { scan-assembler-times {\tmovprfx\tz0, z1\n} 8 } } */
> diff --git a/gcc/testsuite/lib/target-supports.exp
> b/gcc/testsuite/lib/target-supports.exp
> index 75703ddca60..a8833d585c6 100644
> --- a/gcc/testsuite/lib/target-supports.exp
> +++ b/gcc/testsuite/lib/target-supports.exp
> @@ -12100,6 +12100,16 @@ foreach { aarch64_ext } { "fp" "simd" "crypto" "crc"
> "lse" "dotprod" "sve"
> }]
> }
>
> +proc check_effective_target_aarch64_asm_sve2p1_ok { } {
> + if { [istarget aarch64*-*-*] } {
> + return [check_no_compiler_messages aarch64_sve2p1_assembler object {
> + __asm__ (".arch_extension sve2p1; ld1w {z0.q},p7/z,[x0]");
> + } "-march=armv8-a+sve2p1"]
> + } else {
> + return 0
> + }
> +}
> +
> proc check_effective_target_aarch64_small { } {
> if { [istarget aarch64*-*-*] } {
> return [check_no_compiler_messages aarch64_small object {
