ID_AA64SMFR0_EL1 | Jon's Arm Reference

Provides information about the implemented features of the AArch64 Scalable Matrix Extension.

The fields in this register do not follow the standard ID scheme. See 'Alternative ID scheme used for ID_AA64SMFR0_EL1'.

Configuration

Attributes

Field descriptions

FA64, bit [63]

Indicates support for execution of the full A64 instruction set when the PE is in Streaming SVE mode.

The value of this field is an IMPLEMENTATION DEFINED choice of:

FA64	Meaning
0b0	Only those A64 instructions defined as being legal can be executed in Streaming SVE mode.
0b1	All implemented A64 instructions are legal for execution in Streaming SVE mode, when enabled by SMCR_EL1.FA64, SMCR_EL2.FA64, and SMCR_EL3.FA64.

FEAT_SME_FA64 implements the functionality identified by the value 0b1.

Access to this field is RO.

Bits [62:61]

LUTv2, bit [60]

Indicates support for the following additional variants of SME2 lookup table instructions LUTI4 and MOVT:

A LUTI4 instruction having 8-bit result elements, two consecutively numbered source vectors, and four consecutively numbered destination vectors.
If FEAT_SME2p1 is implemented, a LUTI4 instruction having 8-bit result elements, two consecutively numbered source vectors, and four destination vectors with strided register numbering.
A MOVT instruction that copies a single Z source vector to ZT0.

The value of this field is an IMPLEMENTATION DEFINED choice of:

LUTv2	Meaning
0b0	The specified instructions are not implemented.
0b1	The specified instructions are implemented.

All other values are reserved.

FEAT_SME_LUTv2 implements the functionality identified by the value 0b1.

Access to this field is RO.

SMEver, bits [59:56]

When ID_AA64PFR1_EL1.SME != 0b0000:

Indicates support for SME instructions when FEAT_SME is implemented.

The value of this field is an IMPLEMENTATION DEFINED choice of:

SMEver	Meaning
0b0000	The mandatory SME instructions are implemented.
0b0001	As 0b0000, and adds the mandatory SME2 instructions.
0b0010	As 0b0001, and adds the mandatory SME2.1 instructions.

All other values are reserved.

FEAT_SME implements the functionality identified by the value 0b0000.

FEAT_SME2 implements the functionality identified by the value 0b0001.

FEAT_SME2p1 implements the functionality identified by the value 0b0010.

From Armv9.4, the value 0b0001 is not permitted.

Access to this field is RO.

Otherwise:

I16I64, bits [55:52]

Indicates SME support for instructions that accumulate into 64-bit integer elements in the ZA array.

The value of this field is an IMPLEMENTATION DEFINED choice of:

I16I64	Meaning
0b0000	Instructions that accumulate into 64-bit integer elements in the ZA array are not implemented.
0b1111	The variants of the ADDHA, ADDVA, SMOPA, SMOPS, SUMOPA, SUMOPS, UMOPA, UMOPS, USMOPA, and USMOPS instructions that accumulate into 64-bit integer tiles are implemented. When FEAT_SME2 is implemented, the variants of the ADD, ADDA, SDOT, SMLALL, SMLSLL, SUB, SUBA, SVDOT, UDOT, UMLALL, UMLSLL, and UVDOT instructions that accumulate into 64-bit integer elements in ZA array vectors are implemented.

I16I64

Meaning

0b0000

Instructions that accumulate into 64-bit integer elements in the ZA array are not implemented.

0b1111

The variants of the ADDHA, ADDVA, SMOPA, SMOPS, SUMOPA, SUMOPS, UMOPA, UMOPS, USMOPA, and USMOPS instructions that accumulate into 64-bit integer tiles are implemented.

When FEAT_SME2 is implemented, the variants of the ADD, ADDA, SDOT, SMLALL, SMLSLL, SUB, SUBA, SVDOT, UDOT, UMLALL, UMLSLL, and UVDOT instructions that accumulate into 64-bit integer elements in ZA array vectors are implemented.

All other values are reserved.

FEAT_SME_I16I64 implements the functionality identified by the value 0b1111.

Access to this field is RO.

Bits [51:49]

F64F64, bit [48]

Indicates SME support for instructions that accumulate into FP64 double-precision floating-point elements in the ZA array.

The value of this field is an IMPLEMENTATION DEFINED choice of:

F64F64	Meaning
0b0	Instructions that accumulate into double-precision floating-point elements in the ZA array are not implemented.
0b1	The variants of the FMOPA and FMOPS instructions that accumulate into double-precision tiles are implemented. When FEAT_SME2 is implemented, the variants of the FADD, FMLA, FMLS, and FSUB instructions that accumulate into double-precision elements in ZA array vectors are implemented.

F64F64

Meaning

0b0

Instructions that accumulate into double-precision floating-point elements in the ZA array are not implemented.

0b1

The variants of the FMOPA and FMOPS instructions that accumulate into double-precision tiles are implemented.

When FEAT_SME2 is implemented, the variants of the FADD, FMLA, FMLS, and FSUB instructions that accumulate into double-precision elements in ZA array vectors are implemented.

FEAT_SME_F64F64 implements the functionality identified by the value 0b1.

Access to this field is RO.

I16I32, bits [47:44]

Indicates SME2 support for instructions that accumulate 16-bit outer products into 32-bit integer tiles.

The value of this field is an IMPLEMENTATION DEFINED choice of:

I16I32	Meaning
0b0000	Instructions that accumulate 16-bit outer products into 32-bit integer tiles are not implemented.
0b0101	The SMOPA (2-way), SMOPS (2-way), UMOPA (2-way), and UMOPS (2-way) instructions that accumulate 16-bit outer products into 32-bit integer tiles are implemented.

All other values are reserved.

If FEAT_SME2 is implemented, the only permitted value is 0b0101. Otherwise, the only permitted value is 0b0000.

Access to this field is RO.

B16B16, bit [43]

Indicates support for SME2 non-widening BFloat16 instructions.

The value of this field is an IMPLEMENTATION DEFINED choice of:

B16B16	Meaning
0b0	SME2 non-widening BFloat16 instructions are not implemented.
0b1	SME2 BFADD, BFCLAMP, BFMAX, BFMAXNM, BFMIN, BFMINNM, BFMLA, BFMLS, BFMOPA, BFMOPS, and BFSUB instructions with BFloat16 operands and results are implemented.

FEAT_SVE_B16B16 implements the functionality identified by the value 0b1.

This field must indicate the same level of support as ID_AA64ZFR0_EL1.B16B16.

Access to this field is RO.

F16F16, bit [42]

Indicates support for SME2 half-precision floating-point instructions.

The value of this field is an IMPLEMENTATION DEFINED choice of:

F16F16	Meaning
0b0	SME2 half-precision floating-point instructions are not implemented.
0b1	The following SME2 half-precision floating-point instructions are implemented: FMOPA and FMOPS instructions that accumulate half-precision outer-products into half-precision tiles. Multi-vector FADD, FMLA, FMLS, and FSUB instructions with half-precision operands and results. Multi-vector FCVT and FCVTL instructions that convert half-precision inputs to single-precision results.

F16F16

Meaning

0b0

SME2 half-precision floating-point instructions are not implemented.

0b1

The following SME2 half-precision floating-point instructions are implemented:

FMOPA and FMOPS instructions that accumulate half-precision outer-products into half-precision tiles.
Multi-vector FADD, FMLA, FMLS, and FSUB instructions with half-precision operands and results.
Multi-vector FCVT and FCVTL instructions that convert half-precision inputs to single-precision results.

FEAT_SME_F16F16 implements the functionality identified by the value 0b1.

Access to this field is RO.

F8F16, bit [41]

Indicates support for the following SME2 instructions:

The following ZA-targeting FP8 instructions that accumulate into half-precision floating-point elements:
- FDOT (2-way).
- FMLAL.
- FMOPA (2-way).
- FVDOT.
ZA-targeting non-widening half-precision FADD and FSUB instructions.

The value of this field is an IMPLEMENTATION DEFINED choice of:

F8F16	Meaning
0b0	The specified instructions are not implemented.
0b1	The specified instructions are implemented.

If FEAT_SSVE_FP8DOT2 is implemented, the only permitted value is 1.

FEAT_SME_F8F16 implements the functionality identified by the value 1.

Access to this field is RO.

F8F32, bit [40]

Indicates support for the following SME2 ZA-targeting FP8 instructions that accumulate into single-precision floating-point elements:

FDOT (4-way).
FMLALL.
FMOPA (4-way).
FVDOTB.
FVDOTT.

The value of this field is an IMPLEMENTATION DEFINED choice of:

F8F32	Meaning
0b0	The specified instructions are not implemented.
0b1	The specified instructions are implemented.

If FEAT_SSVE_FP8FMA is implemented, the only permitted value is 1.

FEAT_SME_F8F32 implements the functionality identified by the value 1.

Access to this field is RO.

I8I32, bits [39:36]

Indicates SME support for instructions that accumulate 8-bit integer outer products into 32-bit integer tiles.

The value of this field is an IMPLEMENTATION DEFINED choice of:

I8I32	Meaning
0b0000	Instructions that accumulate 8-bit outer products into 32-bit tiles are not implemented.
0b1111	The SMOPA, SMOPS, SUMOPA, SUMOPS, UMOPA, UMOPS, USMOPA, and USMOPS instructions that accumulate 8-bit outer products into 32-bit tiles are implemented.

All other values are reserved.

If FEAT_SME is implemented, the only permitted value is 0b1111.

Access to this field is RO.

F16F32, bit [35]

Indicates SME support for instructions that accumulate FP16 half-precision floating-point outer products into FP32 single-precision floating-point tiles.

The value of this field is an IMPLEMENTATION DEFINED choice of:

F16F32	Meaning
0b0	Instructions that accumulate half-precision outer products into single-precision tiles are not implemented.
0b1	The FMOPA and FMOPS instructions that accumulate half-precision outer products into single-precision tiles are implemented.

If FEAT_SME is implemented, the only permitted value is 0b1.

Access to this field is RO.

B16F32, bit [34]

Indicates SME support for instructions that accumulate BFloat16 outer products into FP32 single-precision floating-point tiles.

The value of this field is an IMPLEMENTATION DEFINED choice of:

B16F32	Meaning
0b0	Instructions that accumulate BFloat16 outer products into single-precision tiles are not implemented.
0b1	The BFMOPA and BFMOPS instructions that accumulate BFloat16 outer products into single-precision tiles are implemented.

If FEAT_SME is implemented, the only permitted value is 0b1.

Access to this field is RO.

BI32I32, bit [33]

Indicates SME support for instructions that accumulate thirty-two 1-bit binary outer products into 32-bit integer tiles.

The value of this field is an IMPLEMENTATION DEFINED choice of:

BI32I32	Meaning
0b0	Instructions that accumulate 1-bit binary outer products into 32-bit integer tiles are not implemented.
0b1	The BMOPA and BMOPS instructions that accumulate 1-bit binary outer products into 32-bit integer tiles are implemented.

If FEAT_SME2 is implemented, the only permitted value is 0b1. Otherwise, the only permitted value is 0b0.

Access to this field is RO.

F32F32, bit [32]

Indicates SME support for instructions that accumulate FP32 single-precision floating-point outer products into single-precision floating-point tiles.

The value of this field is an IMPLEMENTATION DEFINED choice of:

F32F32	Meaning
0b0	Instructions that accumulate single-precision outer products into single-precision tiles are not implemented.
0b1	The FMOPA and FMOPS instructions that accumulate single-precision outer products into single-precision tiles are implemented.

If FEAT_SME is implemented, the only permitted value is 0b1.

Access to this field is RO.

Bit [31]

SF8FMA, bit [30]

Indicates support for the following SVE2 FP8 to single-precision and half-precision multiply-accumulate instructions when the PE is in Streaming SVE mode:

FMLALB.
FMLALT.
FMLALLBB.
FMLALLBT.
FMLALLTB.
FMLALLTT.

The value of this field is an IMPLEMENTATION DEFINED choice of:

SF8FMA	Meaning
0b0	The specified instructions are not implemented by this feature.
0b1	When the PE is in Streaming SVE mode, the specified instructions are implemented.

Note

Other features may implement some of the specified instructions.

If FEAT_SME2 and FEAT_FP8FMA are implemented, the only permitted value is 1.

FEAT_SSVE_FP8FMA implements the functionality identified by the value 1.

Access to this field is RO.

SF8DP4, bit [29]

Indicates support for the SVE2 FP8 to single-precision 4-way dot product FDOT (4-way) instructions when the PE is in Streaming SVE mode.

The value of this field is an IMPLEMENTATION DEFINED choice of:

SF8DP4	Meaning
0b0	The specified instructions are not implemented by this feature.
0b1	When the PE is in Streaming SVE mode, the specified instructions are implemented.

Note

Other features may implement some of the specified instructions.

When FEAT_SME2 and FEAT_FP8DOT4 are implemented, the only permitted value is 1.

FEAT_SSVE_FP8DOT4 implements the functionality identified by the value 1.

Access to this field is RO.

SF8DP2, bit [28]

Indicates support for the SVE2 FP8 to half-precision 2-way dot product FDOT (2-way) instructions when the PE is in Streaming SVE mode.

The value of this field is an IMPLEMENTATION DEFINED choice of:

SF8DP2	Meaning
0b0	The specified instructions are not implemented by this feature.
0b1	When the PE is in Streaming SVE mode, the specified instructions are implemented.

Note

Other features may implement some of the specified instructions.

When FEAT_SME2 and FEAT_FP8DOT2 are implemented, the only value permitted is 1.

FEAT_SSVE_FP8DOT2 implements the functionality identified by the value 1.

Access to this field is RO.

Bits [27:0]

Accessing ID_AA64SMFR0_EL1

This register is read-only and can be accessed from EL1 and higher.

This register is only accessible from the AArch64 state.

Accesses to this register use the following encodings in the System register encoding space:

MRS <Xt>, ID_AA64SMFR0_EL1

op0	op1	CRn	CRm	op2
0b11	0b000	0b0000	0b0100	0b101

if PSTATE.EL == EL0 then if IsFeatureImplemented(FEAT_IDST) then if EL2Enabled() && HCR_EL2.TGE == '1' then AArch64.SystemAccessTrap(EL2, 0x18); else AArch64.SystemAccessTrap(EL1, 0x18); else UNDEFINED; elsif PSTATE.EL == EL1 then if EL2Enabled() && (IsFeatureImplemented(FEAT_FGT) || !IsZero(ID_AA64SMFR0_EL1) || boolean IMPLEMENTATION_DEFINED "ID_AA64SMFR0_EL1 trapped by HCR_EL2.TID3") && HCR_EL2.TID3 == '1' then AArch64.SystemAccessTrap(EL2, 0x18); else X[t, 64] = ID_AA64SMFR0_EL1; elsif PSTATE.EL == EL2 then X[t, 64] = ID_AA64SMFR0_EL1; elsif PSTATE.EL == EL3 then X[t, 64] = ID_AA64SMFR0_EL1;

ID_AA64SMFR0_EL1: SME Feature ID Register 0

Purpose

Configuration

Attributes

Field descriptions

FA64, bit [63]

Bits [62:61]

LUTv2, bit [60]

SMEver, bits [59:56]

When ID_AA64PFR1_EL1.SME != 0b0000:

Otherwise:

I16I64, bits [55:52]

Bits [51:49]

F64F64, bit [48]

I16I32, bits [47:44]

B16B16, bit [43]

F16F16, bit [42]

F8F16, bit [41]

F8F32, bit [40]

I8I32, bits [39:36]

F16F32, bit [35]

B16F32, bit [34]

BI32I32, bit [33]

F32F32, bit [32]

Bit [31]

SF8FMA, bit [30]

SF8DP4, bit [29]

SF8DP2, bit [28]

Bits [27:0]

Accessing ID_AA64SMFR0_EL1

63	62	61	60	59	58	57	56	55	54	53	52	51	50	49	48	47	46	45	44	43	42	41	40	39	38	37	36	35	34	33	32
31	30	29	28	27	26	25	24	23	22	21	20	19	18	17	16	15	14	13	12	11	10	9	8	7	6	5	4	3	2	1	0
FA64	RES0		LUTv2	SMEver				I16I64				RES0			F64F64	I16I32				B16B16	F16F16	F8F16	F8F32	I8I32				F16F32	B16F32	BI32I32	F32F32
RES0	SF8FMA	SF8DP4	SF8DP2	RES0

ID_AA64SMFR0_EL1: SME Feature ID Register 0

Purpose

Configuration

Attributes

Field descriptions

FA64, bit [63]

Bits [62:61]

LUTv2, bit [60]

SMEver, bits [59:56]When ID_AA64PFR1_EL1.SME != 0b0000:

Otherwise:

I16I64, bits [55:52]

Bits [51:49]

F64F64, bit [48]

I16I32, bits [47:44]

B16B16, bit [43]

F16F16, bit [42]

F8F16, bit [41]

F8F32, bit [40]

I8I32, bits [39:36]

F16F32, bit [35]

B16F32, bit [34]

BI32I32, bit [33]

F32F32, bit [32]

Bit [31]

SF8FMA, bit [30]

SF8DP4, bit [29]

SF8DP2, bit [28]

Bits [27:0]

Accessing ID_AA64SMFR0_EL1

SMEver, bits [59:56]

When ID_AA64PFR1_EL1.SME != 0b0000: