target/arm: Enable FEAT_FP8DOT2, FEAT_SSVE_FP8DOT2 for -cpu max
Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20260609192110.752384-40-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Richard Henderson committed
Jun 9, 2026 at 12:21 UTC
c67d5463dabbf6418784fa0d4ac3a920c7b83aa2
3 files changed
+6
docs/system/arm/emulation.rst
+2
@@ -76,6 +76,7 @@ the following architecture extensions:
76
- FEAT_FP (Floating Point extensions)
77
- FEAT_FP16 (Half-precision floating-point data processing)
78
- FEAT_FP8 (FP8 convert instructions)
79
+- FEAT_FP8DOT2 (FP8 2-way dot product to half-precision instructions)
80
- FEAT_FP8DOT4 (FP8 4-way dot product to single-precision instructions)
81
- FEAT_FP8FMA (FP8 multiply-accumulate to half-precision and single-precision instructions)
82
- FEAT_FPAC (Faulting on AUT* instructions)
@@ -168,6 +169,7 @@ the following architecture extensions:
169
- FEAT_SME_F64F64 (Double-precision floating-point outer product instructions)
170
- FEAT_SME_I16I64 (16-bit to 64-bit integer widening outer product instructions)
171
- FEAT_SME_LUTv2 (Lookup table instructions with 4-bit indices and 8-bit elements)
172
+- FEAT_SSVE_FP8DOT2 (SVE2 FP8 2-way dot product to half-precision instructions in Streaming SVE mode)
173
- FEAT_SSVE_FP8DOT4 (SVE2 FP8 4-way dot product to single-precision instructions in Streaming SVE mode)
174
- FEAT_SSVE_FP8FMA (SVE2 FP8 multiply-accumulate to half-precision and single-precision instructions in Streaming SVE mode)
175
- FEAT_SVE (Scalable Vector Extension)
linux-user/aarch64/elfload.c
+2
@@ -228,8 +228,10 @@ abi_ulong get_elf_hwcap2(CPUState *cs)
228
ARM_HWCAP2_A64_F8E5M2);
229
GET_FEATURE_ID(aa64_f8fma, ARM_HWCAP2_A64_F8FMA);
230
GET_FEATURE_ID(aa64_f8dp4, ARM_HWCAP2_A64_F8DP4);
231
+ GET_FEATURE_ID(aa64_f8dp2, ARM_HWCAP2_A64_F8DP2);
232
GET_FEATURE_ID(aa64_ssve_f8fma, ARM_HWCAP2_A64_SME_SF8FMA);
233
GET_FEATURE_ID(aa64_ssve_f8dp4, ARM_HWCAP2_A64_SME_SF8DP4);
234
+ GET_FEATURE_ID(aa64_ssve_f8dp2, ARM_HWCAP2_A64_SME_SF8DP2);
235
236
return hwcaps;
237
}
target/arm/tcg/cpu64.c
+2
@@ -1383,6 +1383,7 @@ void aarch64_max_tcg_initfn(Object *obj)
1383
SET_IDREG(isar, ID_AA64DFR0, t);
1384
1385
t = GET_IDREG(isar, ID_AA64SMFR0);
1386
+ t = FIELD_DP64(t, ID_AA64SMFR0, SF8DP2, 1); /* FEAT_SSVE_FP8DOT2 */
1387
t = FIELD_DP64(t, ID_AA64SMFR0, SF8DP4, 1); /* FEAT_SSVE_FP8DOT4 */
1388
t = FIELD_DP64(t, ID_AA64SMFR0, SF8FMA, 1); /* FEAT_SSVE_FP8FMA */
1389
t = FIELD_DP64(t, ID_AA64SMFR0, F32F32, 1); /* FEAT_SME */
@@ -1403,6 +1404,7 @@ void aarch64_max_tcg_initfn(Object *obj)
1404
t = GET_IDREG(isar, ID_AA64FPFR0);
1405
t = FIELD_DP64(t, ID_AA64FPFR0, F8E5M2, 1); /* FEAT_FP8 */
1406
t = FIELD_DP64(t, ID_AA64FPFR0, F8E4M3, 1); /* FEAT_FP8 */
1407
+ t = FIELD_DP64(t, ID_AA64FPFR0, F8DP2, 1); /* FEAT_FP8DOT2 */
1408
t = FIELD_DP64(t, ID_AA64FPFR0, F8DP4, 1); /* FEAT_FP8DOT4 */
1409
t = FIELD_DP64(t, ID_AA64FPFR0, F8FMA, 1); /* FEAT_FP8FMA */
1410
t = FIELD_DP64(t, ID_AA64FPFR0, F8CVT, 1); /* FEAT_FP8 */