https://gcc.gnu.org/g:22d25b1fee6656df3203dc8ffee6d1d55fc1a7fd
commit r15-5086-g22d25b1fee6656df3203dc8ffee6d1d55fc1a7fd Author: Richard Sandiford <richard.sandif...@arm.com> Date: Mon Nov 11 12:32:14 2024 +0000 aarch64: Test TARGET_STREAMING instead of TARGET_STREAMING_SME g:ede97598e2c recorded separate ISA requirements for streaming and non-streaming mode. The premise there was that AARCH64_FL_SME should not be included in the streaming mode requirements, since: (a) an __arm_streaming_compatible function wouldn't be in streaming mode if SME wasn't available. (b) __arm_streaming_compatible functions only allow things that are possible in non-streaming mode, so the non-streaming architecture is enough to assemble the code, even if +sme isn't enabled. (c) we reject __arm_streaming if +sme isn't enabled, so don't need to test it for individual intrinsics as well. Later patches lean into this further. This patch applies the same reasoning to the .md constructs for base streaming-only SME instructions, guarding them with TARGET_STREAMING rather than TARGET_STREAMING_SME. gcc/ * config/aarch64/aarch64.h (TARGET_SME): Expand comment. (TARGET_STREAMING_SME): Delete. * config/aarch64/aarch64-sme.md: Use TARGET_STREAMING instead of TARGET_STREAMING_SME. * config/aarch64/aarch64-sve2.md: Likewise. Diff: --- gcc/config/aarch64/aarch64-sme.md | 28 ++++++++++++++-------------- gcc/config/aarch64/aarch64-sve2.md | 8 ++++---- gcc/config/aarch64/aarch64.h | 6 ++---- 3 files changed, 20 insertions(+), 22 deletions(-) diff --git a/gcc/config/aarch64/aarch64-sme.md b/gcc/config/aarch64/aarch64-sme.md index 9215f51b01f8..8fca138314c2 100644 --- a/gcc/config/aarch64/aarch64-sme.md +++ b/gcc/config/aarch64/aarch64-sme.md @@ -481,7 +481,7 @@ (match_operand:<VPRED> 2 "register_operand" "Upl") (match_operand:SME_ZA_I 3 "aarch64_sve_ldff1_operand" "Utf")] SME_LD1))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "ld1<Vesize>\t{ za%0<hv>.<Vetype>[%w1, 0] }, %2/z, %3" ) @@ -496,7 +496,7 @@ (match_operand:<VPRED> 3 "register_operand" "Upl") (match_operand:SME_ZA_I 4 "aarch64_sve_ldff1_operand" "Utf")] SME_LD1))] - "TARGET_STREAMING_SME + "TARGET_STREAMING && UINTVAL (operands[2]) < 128 / <elem_bits>" "ld1<Vesize>\t{ za%0<hv>.<Vetype>[%w1, %2] }, %3/z, %4" ) @@ -583,7 +583,7 @@ (match_operand:SI 2 "register_operand" "Ucj") (match_operand:<VPRED> 3 "register_operand" "Upl")] SME_ST1))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "st1<Vesize>\t{ za%1<hv>.<Vetype>[%w2, 0] }, %3, %0" ) @@ -598,7 +598,7 @@ (match_operand:SI 3 "const_int_operand")) (match_operand:<VPRED> 4 "register_operand" "Upl")] SME_ST1))] - "TARGET_STREAMING_SME + "TARGET_STREAMING && UINTVAL (operands[3]) < 128 / <elem_bits>" "st1<Vesize>\t{ za%1<hv>.<Vetype>[%w2, %3] }, %4, %0" ) @@ -663,7 +663,7 @@ (match_operand:DI 3 "const_int_operand") (match_operand:SI 4 "register_operand" "Ucj")] SME_READ))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "mova\t%0.<Vetype>, %2/m, za%3<hv>.<Vetype>[%w4, 0]" ) @@ -678,7 +678,7 @@ (plus:SI (match_operand:SI 4 "register_operand" "Ucj") (match_operand:SI 5 "const_int_operand"))] SME_READ))] - "TARGET_STREAMING_SME + "TARGET_STREAMING && UINTVAL (operands[5]) < 128 / <elem_bits>" "mova\t%0.<Vetype>, %2/m, za%3<hv>.<Vetype>[%w4, %5]" ) @@ -693,7 +693,7 @@ (match_operand:DI 3 "const_int_operand") (match_operand:SI 4 "register_operand" "Ucj")] SME_READ))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "mova\t%0.q, %2/m, za%3<hv>.q[%w4, 0]" ) @@ -707,7 +707,7 @@ (match_operand:<VPRED> 2 "register_operand" "Upl") (match_operand:SVE_FULL 3 "register_operand" "w")] SME_WRITE))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "mova\tza%0<hv>.<Vetype>[%w1, 0], %2/m, %3.<Vetype>" ) @@ -722,7 +722,7 @@ (match_operand:<VPRED> 3 "register_operand" "Upl") (match_operand:SVE_FULL 4 "register_operand" "w")] SME_WRITE))] - "TARGET_STREAMING_SME + "TARGET_STREAMING && UINTVAL (operands[2]) < 128 / <elem_bits>" "mova\tza%0<hv>.<Vetype>[%w1, %2], %3/m, %4.<Vetype>" ) @@ -737,7 +737,7 @@ (match_operand:VNx2BI 2 "register_operand" "Upl") (match_operand:SVE_FULL 3 "register_operand" "w")] SME_WRITE))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "mova\tza%0<hv>.q[%w1, 0], %2/m, %3.q" ) @@ -917,7 +917,7 @@ (match_operand:<VPRED> 2 "register_operand" "Upl") (match_operand:SME_ZA_SDI 3 "register_operand" "w")] SME_BINARY_SDI))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "<optab>\tza%0.<Vetype>, %1/m, %2/m, %3.<Vetype>" ) @@ -1479,7 +1479,7 @@ (match_operand:VNx16QI_ONLY 3 "register_operand" "w") (match_operand:VNx16QI_ONLY 4 "register_operand" "w")] SME_INT_MOP))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" "<optab>\tza%0.s, %1/m, %2/m, %3.b, %4.b" ) @@ -1494,7 +1494,7 @@ (match_operand:VNx8HI_ONLY 3 "register_operand" "w") (match_operand:VNx8HI_ONLY 4 "register_operand" "w")] SME_INT_MOP))] - "TARGET_STREAMING_SME && TARGET_SME_I16I64" + "TARGET_STREAMING && TARGET_SME_I16I64" "<optab>\tza%0.d, %1/m, %2/m, %3.h, %4.h" ) @@ -1887,7 +1887,7 @@ (match_operand:SME_MOP_HSDF 3 "register_operand" "w") (match_operand:SME_MOP_HSDF 4 "register_operand" "w")] SME_FP_MOP))] - "TARGET_STREAMING_SME + "TARGET_STREAMING && (<SME_ZA_SDF_I:elem_bits> == 32) == (<SME_MOP_HSDF:elem_bits> <= 32)" "<b><optab>\tza%0.<SME_ZA_SDF_I:Vetype>, %1/m, %2/m, %3.<SME_MOP_HSDF:Vetype>, %4.<SME_MOP_HSDF:Vetype>" ) diff --git a/gcc/config/aarch64/aarch64-sve2.md b/gcc/config/aarch64/aarch64-sve2.md index 38ecdd1ccc1f..a7b29daeba40 100644 --- a/gcc/config/aarch64/aarch64-sve2.md +++ b/gcc/config/aarch64/aarch64-sve2.md @@ -560,7 +560,7 @@ (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_SME" + "TARGET_STREAMING" {@ [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 @@ UNSPEC_PRED_X) (match_operand:SVE_FULL_I 3 "register_operand"))] UNSPEC_PRED_X))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" {@ [cons: =0, 1, 2, 3; attrs: movprfx] [ w, %0, w, w; * ] # [ ?&w, w, w, w; yes ] # @@ -3182,7 +3182,7 @@ [(match_operand:SVE_FULL 2 "register_operand")] UNSPEC_REVD_ONLY)] UNSPEC_PRED_X))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" {@ [ 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 @@ UNSPEC_REVD_ONLY) (match_operand:SVE_FULL 3 "register_operand")] UNSPEC_SEL))] - "TARGET_STREAMING_SME" + "TARGET_STREAMING" {@ [ 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 593319fd4723..d17f40ce22e8 100644 --- a/gcc/config/aarch64/aarch64.h +++ b/gcc/config/aarch64/aarch64.h @@ -339,12 +339,10 @@ constexpr auto AARCH64_FL_DEFAULT_ISA_MODE ATTRIBUTE_UNUSED #define TARGET_SVE2_SM4 (AARCH64_HAVE_ISA (SVE2_SM4) && TARGET_NON_STREAMING) /* SME instructions, enabled through +sme. Note that this does not - imply anything about the state of PSTATE.SM. */ + imply anything about the state of PSTATE.SM; instructions that require + SME and streaming mode should use TARGET_STREAMING instead. */ #define TARGET_SME AARCH64_HAVE_ISA (SME) -/* Same with streaming mode enabled. */ -#define TARGET_STREAMING_SME (TARGET_STREAMING && TARGET_SME) - /* The FEAT_SME_I16I64 extension to SME, enabled through +sme-i16i64. */ #define TARGET_SME_I16I64 AARCH64_HAVE_ISA (SME_I16I64)