@samitouri / QOSamiQemu / commits / 908c7a23d5

target/arm: Enable FEAT_FP8DOT4, FEAT_SSVE_FP8DOT4 for -cpu max

Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20260609192110.752384-37-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Richard Henderson committed Jun 9, 2026 at 12:21 UTC 908c7a23d5984e94934cdc7a29ccfdf404ca8f42
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_FP8DOT4 (FP8 4-way dot product to single-precision instructions)
80 - FEAT_FP8FMA (FP8 multiply-accumulate to half-precision and single-precision instructions)
81 - FEAT_FPAC (Faulting on AUT* instructions)
82 - FEAT_FPACCOMBINE (Faulting on combined pointer authentication instructions)
@@ -167,6 +168,7 @@ the following architecture extensions:
168 - FEAT_SME_F64F64 (Double-precision floating-point outer product instructions)
169 - FEAT_SME_I16I64 (16-bit to 64-bit integer widening outer product instructions)
170 - FEAT_SME_LUTv2 (Lookup table instructions with 4-bit indices and 8-bit elements)
171 +- FEAT_SSVE_FP8DOT4 (SVE2 FP8 4-way dot product to single-precision instructions in Streaming SVE mode)
172 - FEAT_SSVE_FP8FMA (SVE2 FP8 multiply-accumulate to half-precision and single-precision instructions in Streaming SVE mode)
173 - FEAT_SVE (Scalable Vector Extension)
174 - FEAT_SVE_AES (Scalable Vector AES instructions)
linux-user/aarch64/elfload.c
+2
@@ -227,7 +227,9 @@ abi_ulong get_elf_hwcap2(CPUState *cs)
227 ARM_HWCAP2_A64_F8E4M3 |
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_ssve_f8fma, ARM_HWCAP2_A64_SME_SF8FMA);
232 + GET_FEATURE_ID(aa64_ssve_f8dp4, ARM_HWCAP2_A64_SME_SF8DP4);
233
234 return hwcaps;
235 }
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, SF8DP4, 1); /* FEAT_SSVE_FP8DOT4 */
1387 t = FIELD_DP64(t, ID_AA64SMFR0, SF8FMA, 1); /* FEAT_SSVE_FP8FMA */
1388 t = FIELD_DP64(t, ID_AA64SMFR0, F32F32, 1); /* FEAT_SME */
1389 t = FIELD_DP64(t, ID_AA64SMFR0, BI32I32, 1); /* FEAT_SME2 */
@@ -1402,6 +1403,7 @@ void aarch64_max_tcg_initfn(Object *obj)
1403 t = GET_IDREG(isar, ID_AA64FPFR0);
1404 t = FIELD_DP64(t, ID_AA64FPFR0, F8E5M2, 1); /* FEAT_FP8 */
1405 t = FIELD_DP64(t, ID_AA64FPFR0, F8E4M3, 1); /* FEAT_FP8 */
1406 + t = FIELD_DP64(t, ID_AA64FPFR0, F8DP4, 1); /* FEAT_FP8DOT4 */
1407 t = FIELD_DP64(t, ID_AA64FPFR0, F8FMA, 1); /* FEAT_FP8FMA */
1408 t = FIELD_DP64(t, ID_AA64FPFR0, F8CVT, 1); /* FEAT_FP8 */
1409 SET_IDREG(isar, ID_AA64FPFR0, t);