Skip to contentSkip to conversation

i386: Add new CPU model SapphireRapids

PATCH In review Wang, Lei 6 participants 11 messages Aug 12, 2022 – Feb 17, 2023

lore
Share & tools
lore

Wang, Lei

author
#0 lore raw

PATCH 1/1 i386: Add new CPU model SapphireRapids

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

---
 target/i386/cpu.c | 128 ++++++++++++++++++++++++++++++++++++++++++++++
 target/i386/cpu.h |   4 ++
 2 files changed, 132 insertions(+)
target/i386/cpu.c +128 -0
index 1db1278a59..abb43853d4 100644
@@ -3467,6 +3467,134 @@ static const X86CPUDefinition builtin_x86_defs[] = {
             { /* end of list */ }
         }
     },
+    {
+        .name = "SapphireRapids",
+        .level = 0x20,
+        .vendor = CPUID_VENDOR_INTEL,
+        .family = 6,
+        .model = 143,
+        .stepping = 4,
+        /*
+         * please keep the ascending order so that we can have a clear view of
+         * bit position of each feature.
+         */
+        .features[FEAT_1_EDX] =
+            CPUID_FP87 | CPUID_VME | CPUID_DE | CPUID_PSE | CPUID_TSC |
+            CPUID_MSR | CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC |
+            CPUID_SEP | CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV |
+            CPUID_PAT | CPUID_PSE36 | CPUID_CLFLUSH | CPUID_MMX | CPUID_FXSR |
+            CPUID_SSE | CPUID_SSE2,
+        .features[FEAT_1_ECX] =
+            CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | CPUID_EXT_SSSE3 |
+            CPUID_EXT_FMA | CPUID_EXT_CX16 | CPUID_EXT_PCID | CPUID_EXT_SSE41 |
+            CPUID_EXT_SSE42 | CPUID_EXT_X2APIC | CPUID_EXT_MOVBE |
+            CPUID_EXT_POPCNT | CPUID_EXT_TSC_DEADLINE_TIMER | CPUID_EXT_AES |
+            CPUID_EXT_XSAVE | CPUID_EXT_AVX | CPUID_EXT_F16C | CPUID_EXT_RDRAND,
+        .features[FEAT_8000_0001_EDX] =
+            CPUID_EXT2_SYSCALL | CPUID_EXT2_NX | CPUID_EXT2_PDPE1GB |
+            CPUID_EXT2_RDTSCP | CPUID_EXT2_LM,
+        .features[FEAT_8000_0001_ECX] =
+            CPUID_EXT3_LAHF_LM | CPUID_EXT3_ABM | CPUID_EXT3_3DNOWPREFETCH,
+        .features[FEAT_8000_0008_EBX] =
+            CPUID_8000_0008_EBX_WBNOINVD,
+        .features[FEAT_7_0_EBX] =
+            CPUID_7_0_EBX_FSGSBASE | CPUID_7_0_EBX_BMI1 | CPUID_7_0_EBX_HLE |
+            CPUID_7_0_EBX_AVX2 | CPUID_7_0_EBX_SMEP | CPUID_7_0_EBX_BMI2 |
+            CPUID_7_0_EBX_ERMS | CPUID_7_0_EBX_INVPCID | CPUID_7_0_EBX_RTM |
+            CPUID_7_0_EBX_AVX512F | CPUID_7_0_EBX_AVX512DQ |
+            CPUID_7_0_EBX_RDSEED | CPUID_7_0_EBX_ADX | CPUID_7_0_EBX_SMAP |
+            CPUID_7_0_EBX_AVX512IFMA | CPUID_7_0_EBX_CLFLUSHOPT |
+            CPUID_7_0_EBX_CLWB | CPUID_7_0_EBX_AVX512CD | CPUID_7_0_EBX_SHA_NI |
+            CPUID_7_0_EBX_AVX512BW | CPUID_7_0_EBX_AVX512VL,
+        .features[FEAT_7_0_ECX] =
+            CPUID_7_0_ECX_AVX512_VBMI | CPUID_7_0_ECX_UMIP | CPUID_7_0_ECX_PKU |
+            CPUID_7_0_ECX_AVX512_VBMI2 | CPUID_7_0_ECX_GFNI |
+            CPUID_7_0_ECX_VAES | CPUID_7_0_ECX_VPCLMULQDQ |
+            CPUID_7_0_ECX_AVX512VNNI | CPUID_7_0_ECX_AVX512BITALG |
+            CPUID_7_0_ECX_AVX512_VPOPCNTDQ | CPUID_7_0_ECX_LA57 |
+            CPUID_7_0_ECX_RDPID | CPUID_7_0_ECX_BUS_LOCK_DETECT,
+        .features[FEAT_7_0_EDX] =
+            CPUID_7_0_EDX_FSRM | CPUID_7_0_EDX_SERIALIZE |
+            CPUID_7_0_EDX_TSX_LDTRK | CPUID_7_0_EDX_AMX_BF16 |
+            CPUID_7_0_EDX_AVX512_FP16 | CPUID_7_0_EDX_AMX_TILE |
+            CPUID_7_0_EDX_AMX_INT8 | CPUID_7_0_EDX_SPEC_CTRL |
+            CPUID_7_0_EDX_ARCH_CAPABILITIES | CPUID_7_0_EDX_SPEC_CTRL_SSBD,
+        .features[FEAT_ARCH_CAPABILITIES] =
+            MSR_ARCH_CAP_RDCL_NO | MSR_ARCH_CAP_IBRS_ALL |
+            MSR_ARCH_CAP_SKIP_L1DFL_VMENTRY | MSR_ARCH_CAP_MDS_NO |
+            MSR_ARCH_CAP_PSCHANGE_MC_NO | MSR_ARCH_CAP_TAA_NO,
+        .features[FEAT_XSAVE] =
+            CPUID_XSAVE_XSAVEOPT | CPUID_XSAVE_XSAVEC |
+            CPUID_XSAVE_XGETBV1 | CPUID_XSAVE_XSAVES | CPUID_D_1_EAX_XFD,
+        .features[FEAT_6_EAX] =
+            CPUID_6_EAX_ARAT,
+        .features[FEAT_7_1_EAX] =
+            CPUID_7_1_EAX_AVX_VNNI | CPUID_7_1_EAX_AVX512_BF16,
+        .features[FEAT_VMX_BASIC] =
+            MSR_VMX_BASIC_INS_OUTS | MSR_VMX_BASIC_TRUE_CTLS,
+        .features[FEAT_VMX_ENTRY_CTLS] =
+            VMX_VM_ENTRY_LOAD_DEBUG_CONTROLS | VMX_VM_ENTRY_IA32E_MODE |
+            VMX_VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_ENTRY_LOAD_IA32_PAT | VMX_VM_ENTRY_LOAD_IA32_EFER,
+        .features[FEAT_VMX_EPT_VPID_CAPS] =
+            MSR_VMX_EPT_EXECONLY |
+            MSR_VMX_EPT_PAGE_WALK_LENGTH_4 | MSR_VMX_EPT_PAGE_WALK_LENGTH_5 |
+            MSR_VMX_EPT_WB | MSR_VMX_EPT_2MB | MSR_VMX_EPT_1GB |
+            MSR_VMX_EPT_INVEPT | MSR_VMX_EPT_AD_BITS |
+            MSR_VMX_EPT_INVEPT_SINGLE_CONTEXT | MSR_VMX_EPT_INVEPT_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID | MSR_VMX_EPT_INVVPID_SINGLE_ADDR |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT |
+            MSR_VMX_EPT_INVVPID_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT_NOGLOBALS,
+        .features[FEAT_VMX_EXIT_CTLS] =
+            VMX_VM_EXIT_SAVE_DEBUG_CONTROLS |
+            VMX_VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_EXIT_ACK_INTR_ON_EXIT | VMX_VM_EXIT_SAVE_IA32_PAT |
+            VMX_VM_EXIT_LOAD_IA32_PAT | VMX_VM_EXIT_SAVE_IA32_EFER |
+            VMX_VM_EXIT_LOAD_IA32_EFER | VMX_VM_EXIT_SAVE_VMX_PREEMPTION_TIMER,
+        .features[FEAT_VMX_MISC] =
+            MSR_VMX_MISC_STORE_LMA | MSR_VMX_MISC_ACTIVITY_HLT |
+            MSR_VMX_MISC_VMWRITE_VMEXIT,
+        .features[FEAT_VMX_PINBASED_CTLS] =
+            VMX_PIN_BASED_EXT_INTR_MASK | VMX_PIN_BASED_NMI_EXITING |
+            VMX_PIN_BASED_VIRTUAL_NMIS | VMX_PIN_BASED_VMX_PREEMPTION_TIMER |
+            VMX_PIN_BASED_POSTED_INTR,
+        .features[FEAT_VMX_PROCBASED_CTLS] =
+            VMX_CPU_BASED_VIRTUAL_INTR_PENDING |
+            VMX_CPU_BASED_USE_TSC_OFFSETING | VMX_CPU_BASED_HLT_EXITING |
+            VMX_CPU_BASED_INVLPG_EXITING | VMX_CPU_BASED_MWAIT_EXITING |
+            VMX_CPU_BASED_RDPMC_EXITING | VMX_CPU_BASED_RDTSC_EXITING |
+            VMX_CPU_BASED_CR3_LOAD_EXITING | VMX_CPU_BASED_CR3_STORE_EXITING |
+            VMX_CPU_BASED_CR8_LOAD_EXITING | VMX_CPU_BASED_CR8_STORE_EXITING |
+            VMX_CPU_BASED_TPR_SHADOW | VMX_CPU_BASED_VIRTUAL_NMI_PENDING |
+            VMX_CPU_BASED_MOV_DR_EXITING | VMX_CPU_BASED_UNCOND_IO_EXITING |
+            VMX_CPU_BASED_USE_IO_BITMAPS | VMX_CPU_BASED_MONITOR_TRAP_FLAG |
+            VMX_CPU_BASED_USE_MSR_BITMAPS | VMX_CPU_BASED_MONITOR_EXITING |
+            VMX_CPU_BASED_PAUSE_EXITING |
+            VMX_CPU_BASED_ACTIVATE_SECONDARY_CONTROLS,
+        .features[FEAT_VMX_SECONDARY_CTLS] =
+            VMX_SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
+            VMX_SECONDARY_EXEC_ENABLE_EPT | VMX_SECONDARY_EXEC_DESC |
+            VMX_SECONDARY_EXEC_RDTSCP |
+            VMX_SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
+            VMX_SECONDARY_EXEC_ENABLE_VPID | VMX_SECONDARY_EXEC_WBINVD_EXITING |
+            VMX_SECONDARY_EXEC_UNRESTRICTED_GUEST |
+            VMX_SECONDARY_EXEC_APIC_REGISTER_VIRT |
+            VMX_SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
+            VMX_SECONDARY_EXEC_RDRAND_EXITING |
+            VMX_SECONDARY_EXEC_ENABLE_INVPCID |
+            VMX_SECONDARY_EXEC_ENABLE_VMFUNC | VMX_SECONDARY_EXEC_SHADOW_VMCS |
+            VMX_SECONDARY_EXEC_RDSEED_EXITING | VMX_SECONDARY_EXEC_ENABLE_PML |
+            VMX_SECONDARY_EXEC_XSAVES,
+        .features[FEAT_VMX_VMFUNC] =
+            MSR_VMX_VMFUNC_EPT_SWITCHING,
+        .xlevel = 0x80000008,
+        .model_id = "Intel Xeon Processor (SapphireRapids)",
+        .versions = (X86CPUVersionDefinition[]) {
+            { .version = 1 },
+            { /* end of list */ },
+        },
+    },
     {
         .name = "Denverton",
         .level = 21,
target/i386/cpu.h +4 -0
index 82004b65b9..ef3e8a5ed5 100644
@@ -879,10 +879,14 @@ uint64_t x86_cpu_get_supported_feature_word(FeatureWord w,
 #define CPUID_7_0_EDX_TSX_LDTRK         (1U << 16)
 /* Architectural LBRs */
 #define CPUID_7_0_EDX_ARCH_LBR          (1U << 19)
+/* AMX_BF16 instruction */
+#define CPUID_7_0_EDX_AMX_BF16          (1U << 22)
 /* AVX512_FP16 instruction */
 #define CPUID_7_0_EDX_AVX512_FP16       (1U << 23)
 /* AMX tile (two-dimensional register) */
 #define CPUID_7_0_EDX_AMX_TILE          (1U << 24)
+/* AMX_INT8 instruction */
+#define CPUID_7_0_EDX_AMX_INT8          (1U << 25)
 /* Speculation Control */
 #define CPUID_7_0_EDX_SPEC_CTRL         (1U << 26)
 /* Single Thread Indirect Branch Predictors */
Signature · 1 line
2.34.1
Acked-by on v1 1/1
#1 lore raw
Quoted 27 lines · Lei

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines

looks fine to me,

Quoted 163 lines
---
 target/i386/cpu.c | 128 ++++++++++++++++++++++++++++++++++++++++++++++
 target/i386/cpu.h |   4 ++
 2 files changed, 132 insertions(+)
target/i386/cpu.c +128 -0
index 1db1278a59..abb43853d4 100644
@@ -3467,6 +3467,134 @@ static const X86CPUDefinition builtin_x86_defs[] = {
             { /* end of list */ }
         }
     },
+    {
+        .name = "SapphireRapids",
+        .level = 0x20,
+        .vendor = CPUID_VENDOR_INTEL,
+        .family = 6,
+        .model = 143,
+        .stepping = 4,
+        /*
+         * please keep the ascending order so that we can have a clear view of
+         * bit position of each feature.
+         */
+        .features[FEAT_1_EDX] =
+            CPUID_FP87 | CPUID_VME | CPUID_DE | CPUID_PSE | CPUID_TSC |
+            CPUID_MSR | CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC |
+            CPUID_SEP | CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV |
+            CPUID_PAT | CPUID_PSE36 | CPUID_CLFLUSH | CPUID_MMX | CPUID_FXSR |
+            CPUID_SSE | CPUID_SSE2,
+        .features[FEAT_1_ECX] =
+            CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | CPUID_EXT_SSSE3 |
+            CPUID_EXT_FMA | CPUID_EXT_CX16 | CPUID_EXT_PCID | CPUID_EXT_SSE41 |
+            CPUID_EXT_SSE42 | CPUID_EXT_X2APIC | CPUID_EXT_MOVBE |
+            CPUID_EXT_POPCNT | CPUID_EXT_TSC_DEADLINE_TIMER | CPUID_EXT_AES |
+            CPUID_EXT_XSAVE | CPUID_EXT_AVX | CPUID_EXT_F16C | CPUID_EXT_RDRAND,
+        .features[FEAT_8000_0001_EDX] =
+            CPUID_EXT2_SYSCALL | CPUID_EXT2_NX | CPUID_EXT2_PDPE1GB |
+            CPUID_EXT2_RDTSCP | CPUID_EXT2_LM,
+        .features[FEAT_8000_0001_ECX] =
+            CPUID_EXT3_LAHF_LM | CPUID_EXT3_ABM | CPUID_EXT3_3DNOWPREFETCH,
+        .features[FEAT_8000_0008_EBX] =
+            CPUID_8000_0008_EBX_WBNOINVD,
+        .features[FEAT_7_0_EBX] =
+            CPUID_7_0_EBX_FSGSBASE | CPUID_7_0_EBX_BMI1 | CPUID_7_0_EBX_HLE |
+            CPUID_7_0_EBX_AVX2 | CPUID_7_0_EBX_SMEP | CPUID_7_0_EBX_BMI2 |
+            CPUID_7_0_EBX_ERMS | CPUID_7_0_EBX_INVPCID | CPUID_7_0_EBX_RTM |
+            CPUID_7_0_EBX_AVX512F | CPUID_7_0_EBX_AVX512DQ |
+            CPUID_7_0_EBX_RDSEED | CPUID_7_0_EBX_ADX | CPUID_7_0_EBX_SMAP |
+            CPUID_7_0_EBX_AVX512IFMA | CPUID_7_0_EBX_CLFLUSHOPT |
+            CPUID_7_0_EBX_CLWB | CPUID_7_0_EBX_AVX512CD | CPUID_7_0_EBX_SHA_NI |
+            CPUID_7_0_EBX_AVX512BW | CPUID_7_0_EBX_AVX512VL,
+        .features[FEAT_7_0_ECX] =
+            CPUID_7_0_ECX_AVX512_VBMI | CPUID_7_0_ECX_UMIP | CPUID_7_0_ECX_PKU |
+            CPUID_7_0_ECX_AVX512_VBMI2 | CPUID_7_0_ECX_GFNI |
+            CPUID_7_0_ECX_VAES | CPUID_7_0_ECX_VPCLMULQDQ |
+            CPUID_7_0_ECX_AVX512VNNI | CPUID_7_0_ECX_AVX512BITALG |
+            CPUID_7_0_ECX_AVX512_VPOPCNTDQ | CPUID_7_0_ECX_LA57 |
+            CPUID_7_0_ECX_RDPID | CPUID_7_0_ECX_BUS_LOCK_DETECT,
+        .features[FEAT_7_0_EDX] =
+            CPUID_7_0_EDX_FSRM | CPUID_7_0_EDX_SERIALIZE |
+            CPUID_7_0_EDX_TSX_LDTRK | CPUID_7_0_EDX_AMX_BF16 |
+            CPUID_7_0_EDX_AVX512_FP16 | CPUID_7_0_EDX_AMX_TILE |
+            CPUID_7_0_EDX_AMX_INT8 | CPUID_7_0_EDX_SPEC_CTRL |
+            CPUID_7_0_EDX_ARCH_CAPABILITIES | CPUID_7_0_EDX_SPEC_CTRL_SSBD,
+        .features[FEAT_ARCH_CAPABILITIES] =
+            MSR_ARCH_CAP_RDCL_NO | MSR_ARCH_CAP_IBRS_ALL |
+            MSR_ARCH_CAP_SKIP_L1DFL_VMENTRY | MSR_ARCH_CAP_MDS_NO |
+            MSR_ARCH_CAP_PSCHANGE_MC_NO | MSR_ARCH_CAP_TAA_NO,
+        .features[FEAT_XSAVE] =
+            CPUID_XSAVE_XSAVEOPT | CPUID_XSAVE_XSAVEC |
+            CPUID_XSAVE_XGETBV1 | CPUID_XSAVE_XSAVES | CPUID_D_1_EAX_XFD,
+        .features[FEAT_6_EAX] =
+            CPUID_6_EAX_ARAT,
+        .features[FEAT_7_1_EAX] =
+            CPUID_7_1_EAX_AVX_VNNI | CPUID_7_1_EAX_AVX512_BF16,
+        .features[FEAT_VMX_BASIC] =
+            MSR_VMX_BASIC_INS_OUTS | MSR_VMX_BASIC_TRUE_CTLS,
+        .features[FEAT_VMX_ENTRY_CTLS] =
+            VMX_VM_ENTRY_LOAD_DEBUG_CONTROLS | VMX_VM_ENTRY_IA32E_MODE |
+            VMX_VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_ENTRY_LOAD_IA32_PAT | VMX_VM_ENTRY_LOAD_IA32_EFER,
+        .features[FEAT_VMX_EPT_VPID_CAPS] =
+            MSR_VMX_EPT_EXECONLY |
+            MSR_VMX_EPT_PAGE_WALK_LENGTH_4 | MSR_VMX_EPT_PAGE_WALK_LENGTH_5 |
+            MSR_VMX_EPT_WB | MSR_VMX_EPT_2MB | MSR_VMX_EPT_1GB |
+            MSR_VMX_EPT_INVEPT | MSR_VMX_EPT_AD_BITS |
+            MSR_VMX_EPT_INVEPT_SINGLE_CONTEXT | MSR_VMX_EPT_INVEPT_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID | MSR_VMX_EPT_INVVPID_SINGLE_ADDR |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT |
+            MSR_VMX_EPT_INVVPID_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT_NOGLOBALS,
+        .features[FEAT_VMX_EXIT_CTLS] =
+            VMX_VM_EXIT_SAVE_DEBUG_CONTROLS |
+            VMX_VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_EXIT_ACK_INTR_ON_EXIT | VMX_VM_EXIT_SAVE_IA32_PAT |
+            VMX_VM_EXIT_LOAD_IA32_PAT | VMX_VM_EXIT_SAVE_IA32_EFER |
+            VMX_VM_EXIT_LOAD_IA32_EFER | VMX_VM_EXIT_SAVE_VMX_PREEMPTION_TIMER,
+        .features[FEAT_VMX_MISC] =
+            MSR_VMX_MISC_STORE_LMA | MSR_VMX_MISC_ACTIVITY_HLT |
+            MSR_VMX_MISC_VMWRITE_VMEXIT,
+        .features[FEAT_VMX_PINBASED_CTLS] =
+            VMX_PIN_BASED_EXT_INTR_MASK | VMX_PIN_BASED_NMI_EXITING |
+            VMX_PIN_BASED_VIRTUAL_NMIS | VMX_PIN_BASED_VMX_PREEMPTION_TIMER |
+            VMX_PIN_BASED_POSTED_INTR,
+        .features[FEAT_VMX_PROCBASED_CTLS] =
+            VMX_CPU_BASED_VIRTUAL_INTR_PENDING |
+            VMX_CPU_BASED_USE_TSC_OFFSETING | VMX_CPU_BASED_HLT_EXITING |
+            VMX_CPU_BASED_INVLPG_EXITING | VMX_CPU_BASED_MWAIT_EXITING |
+            VMX_CPU_BASED_RDPMC_EXITING | VMX_CPU_BASED_RDTSC_EXITING |
+            VMX_CPU_BASED_CR3_LOAD_EXITING | VMX_CPU_BASED_CR3_STORE_EXITING |
+            VMX_CPU_BASED_CR8_LOAD_EXITING | VMX_CPU_BASED_CR8_STORE_EXITING |
+            VMX_CPU_BASED_TPR_SHADOW | VMX_CPU_BASED_VIRTUAL_NMI_PENDING |
+            VMX_CPU_BASED_MOV_DR_EXITING | VMX_CPU_BASED_UNCOND_IO_EXITING |
+            VMX_CPU_BASED_USE_IO_BITMAPS | VMX_CPU_BASED_MONITOR_TRAP_FLAG |
+            VMX_CPU_BASED_USE_MSR_BITMAPS | VMX_CPU_BASED_MONITOR_EXITING |
+            VMX_CPU_BASED_PAUSE_EXITING |
+            VMX_CPU_BASED_ACTIVATE_SECONDARY_CONTROLS,
+        .features[FEAT_VMX_SECONDARY_CTLS] =
+            VMX_SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
+            VMX_SECONDARY_EXEC_ENABLE_EPT | VMX_SECONDARY_EXEC_DESC |
+            VMX_SECONDARY_EXEC_RDTSCP |
+            VMX_SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
+            VMX_SECONDARY_EXEC_ENABLE_VPID | VMX_SECONDARY_EXEC_WBINVD_EXITING |
+            VMX_SECONDARY_EXEC_UNRESTRICTED_GUEST |
+            VMX_SECONDARY_EXEC_APIC_REGISTER_VIRT |
+            VMX_SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
+            VMX_SECONDARY_EXEC_RDRAND_EXITING |
+            VMX_SECONDARY_EXEC_ENABLE_INVPCID |
+            VMX_SECONDARY_EXEC_ENABLE_VMFUNC | VMX_SECONDARY_EXEC_SHADOW_VMCS |
+            VMX_SECONDARY_EXEC_RDSEED_EXITING | VMX_SECONDARY_EXEC_ENABLE_PML |
+            VMX_SECONDARY_EXEC_XSAVES,
+        .features[FEAT_VMX_VMFUNC] =
+            MSR_VMX_VMFUNC_EPT_SWITCHING,
+        .xlevel = 0x80000008,
+        .model_id = "Intel Xeon Processor (SapphireRapids)",
+        .versions = (X86CPUVersionDefinition[]) {
+            { .version = 1 },
+            { /* end of list */ },
+        },
+    },
     {
         .name = "Denverton",
         .level = 21,
target/i386/cpu.h +4 -0
index 82004b65b9..ef3e8a5ed5 100644
@@ -879,10 +879,14 @@ uint64_t x86_cpu_get_supported_feature_word(FeatureWord w,
 #define CPUID_7_0_EDX_TSX_LDTRK         (1U << 16)
 /* Architectural LBRs */
 #define CPUID_7_0_EDX_ARCH_LBR          (1U << 19)
+/* AMX_BF16 instruction */
+#define CPUID_7_0_EDX_AMX_BF16          (1U << 22)
 /* AVX512_FP16 instruction */
 #define CPUID_7_0_EDX_AVX512_FP16       (1U << 23)
 /* AMX tile (two-dimensional register) */
 #define CPUID_7_0_EDX_AMX_TILE          (1U << 24)
+/* AMX_INT8 instruction */
+#define CPUID_7_0_EDX_AMX_INT8          (1U << 25)
 /* Speculation Control */
 #define CPUID_7_0_EDX_SPEC_CTRL         (1U << 26)
 /* Single Thread Indirect Branch Predictors */
#2 lore raw
In reply to Wang, Lei
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?
In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

Dave

Quoted 167 lines
---
 target/i386/cpu.c | 128 ++++++++++++++++++++++++++++++++++++++++++++++
 target/i386/cpu.h |   4 ++
 2 files changed, 132 insertions(+)
target/i386/cpu.c +128 -0
index 1db1278a59..abb43853d4 100644
@@ -3467,6 +3467,134 @@ static const X86CPUDefinition builtin_x86_defs[] = {
             { /* end of list */ }
         }
     },
+    {
+        .name = "SapphireRapids",
+        .level = 0x20,
+        .vendor = CPUID_VENDOR_INTEL,
+        .family = 6,
+        .model = 143,
+        .stepping = 4,
+        /*
+         * please keep the ascending order so that we can have a clear view of
+         * bit position of each feature.
+         */
+        .features[FEAT_1_EDX] =
+            CPUID_FP87 | CPUID_VME | CPUID_DE | CPUID_PSE | CPUID_TSC |
+            CPUID_MSR | CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC |
+            CPUID_SEP | CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV |
+            CPUID_PAT | CPUID_PSE36 | CPUID_CLFLUSH | CPUID_MMX | CPUID_FXSR |
+            CPUID_SSE | CPUID_SSE2,
+        .features[FEAT_1_ECX] =
+            CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | CPUID_EXT_SSSE3 |
+            CPUID_EXT_FMA | CPUID_EXT_CX16 | CPUID_EXT_PCID | CPUID_EXT_SSE41 |
+            CPUID_EXT_SSE42 | CPUID_EXT_X2APIC | CPUID_EXT_MOVBE |
+            CPUID_EXT_POPCNT | CPUID_EXT_TSC_DEADLINE_TIMER | CPUID_EXT_AES |
+            CPUID_EXT_XSAVE | CPUID_EXT_AVX | CPUID_EXT_F16C | CPUID_EXT_RDRAND,
+        .features[FEAT_8000_0001_EDX] =
+            CPUID_EXT2_SYSCALL | CPUID_EXT2_NX | CPUID_EXT2_PDPE1GB |
+            CPUID_EXT2_RDTSCP | CPUID_EXT2_LM,
+        .features[FEAT_8000_0001_ECX] =
+            CPUID_EXT3_LAHF_LM | CPUID_EXT3_ABM | CPUID_EXT3_3DNOWPREFETCH,
+        .features[FEAT_8000_0008_EBX] =
+            CPUID_8000_0008_EBX_WBNOINVD,
+        .features[FEAT_7_0_EBX] =
+            CPUID_7_0_EBX_FSGSBASE | CPUID_7_0_EBX_BMI1 | CPUID_7_0_EBX_HLE |
+            CPUID_7_0_EBX_AVX2 | CPUID_7_0_EBX_SMEP | CPUID_7_0_EBX_BMI2 |
+            CPUID_7_0_EBX_ERMS | CPUID_7_0_EBX_INVPCID | CPUID_7_0_EBX_RTM |
+            CPUID_7_0_EBX_AVX512F | CPUID_7_0_EBX_AVX512DQ |
+            CPUID_7_0_EBX_RDSEED | CPUID_7_0_EBX_ADX | CPUID_7_0_EBX_SMAP |
+            CPUID_7_0_EBX_AVX512IFMA | CPUID_7_0_EBX_CLFLUSHOPT |
+            CPUID_7_0_EBX_CLWB | CPUID_7_0_EBX_AVX512CD | CPUID_7_0_EBX_SHA_NI |
+            CPUID_7_0_EBX_AVX512BW | CPUID_7_0_EBX_AVX512VL,
+        .features[FEAT_7_0_ECX] =
+            CPUID_7_0_ECX_AVX512_VBMI | CPUID_7_0_ECX_UMIP | CPUID_7_0_ECX_PKU |
+            CPUID_7_0_ECX_AVX512_VBMI2 | CPUID_7_0_ECX_GFNI |
+            CPUID_7_0_ECX_VAES | CPUID_7_0_ECX_VPCLMULQDQ |
+            CPUID_7_0_ECX_AVX512VNNI | CPUID_7_0_ECX_AVX512BITALG |
+            CPUID_7_0_ECX_AVX512_VPOPCNTDQ | CPUID_7_0_ECX_LA57 |
+            CPUID_7_0_ECX_RDPID | CPUID_7_0_ECX_BUS_LOCK_DETECT,
+        .features[FEAT_7_0_EDX] =
+            CPUID_7_0_EDX_FSRM | CPUID_7_0_EDX_SERIALIZE |
+            CPUID_7_0_EDX_TSX_LDTRK | CPUID_7_0_EDX_AMX_BF16 |
+            CPUID_7_0_EDX_AVX512_FP16 | CPUID_7_0_EDX_AMX_TILE |
+            CPUID_7_0_EDX_AMX_INT8 | CPUID_7_0_EDX_SPEC_CTRL |
+            CPUID_7_0_EDX_ARCH_CAPABILITIES | CPUID_7_0_EDX_SPEC_CTRL_SSBD,
+        .features[FEAT_ARCH_CAPABILITIES] =
+            MSR_ARCH_CAP_RDCL_NO | MSR_ARCH_CAP_IBRS_ALL |
+            MSR_ARCH_CAP_SKIP_L1DFL_VMENTRY | MSR_ARCH_CAP_MDS_NO |
+            MSR_ARCH_CAP_PSCHANGE_MC_NO | MSR_ARCH_CAP_TAA_NO,
+        .features[FEAT_XSAVE] =
+            CPUID_XSAVE_XSAVEOPT | CPUID_XSAVE_XSAVEC |
+            CPUID_XSAVE_XGETBV1 | CPUID_XSAVE_XSAVES | CPUID_D_1_EAX_XFD,
+        .features[FEAT_6_EAX] =
+            CPUID_6_EAX_ARAT,
+        .features[FEAT_7_1_EAX] =
+            CPUID_7_1_EAX_AVX_VNNI | CPUID_7_1_EAX_AVX512_BF16,
+        .features[FEAT_VMX_BASIC] =
+            MSR_VMX_BASIC_INS_OUTS | MSR_VMX_BASIC_TRUE_CTLS,
+        .features[FEAT_VMX_ENTRY_CTLS] =
+            VMX_VM_ENTRY_LOAD_DEBUG_CONTROLS | VMX_VM_ENTRY_IA32E_MODE |
+            VMX_VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_ENTRY_LOAD_IA32_PAT | VMX_VM_ENTRY_LOAD_IA32_EFER,
+        .features[FEAT_VMX_EPT_VPID_CAPS] =
+            MSR_VMX_EPT_EXECONLY |
+            MSR_VMX_EPT_PAGE_WALK_LENGTH_4 | MSR_VMX_EPT_PAGE_WALK_LENGTH_5 |
+            MSR_VMX_EPT_WB | MSR_VMX_EPT_2MB | MSR_VMX_EPT_1GB |
+            MSR_VMX_EPT_INVEPT | MSR_VMX_EPT_AD_BITS |
+            MSR_VMX_EPT_INVEPT_SINGLE_CONTEXT | MSR_VMX_EPT_INVEPT_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID | MSR_VMX_EPT_INVVPID_SINGLE_ADDR |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT |
+            MSR_VMX_EPT_INVVPID_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT_NOGLOBALS,
+        .features[FEAT_VMX_EXIT_CTLS] =
+            VMX_VM_EXIT_SAVE_DEBUG_CONTROLS |
+            VMX_VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_EXIT_ACK_INTR_ON_EXIT | VMX_VM_EXIT_SAVE_IA32_PAT |
+            VMX_VM_EXIT_LOAD_IA32_PAT | VMX_VM_EXIT_SAVE_IA32_EFER |
+            VMX_VM_EXIT_LOAD_IA32_EFER | VMX_VM_EXIT_SAVE_VMX_PREEMPTION_TIMER,
+        .features[FEAT_VMX_MISC] =
+            MSR_VMX_MISC_STORE_LMA | MSR_VMX_MISC_ACTIVITY_HLT |
+            MSR_VMX_MISC_VMWRITE_VMEXIT,
+        .features[FEAT_VMX_PINBASED_CTLS] =
+            VMX_PIN_BASED_EXT_INTR_MASK | VMX_PIN_BASED_NMI_EXITING |
+            VMX_PIN_BASED_VIRTUAL_NMIS | VMX_PIN_BASED_VMX_PREEMPTION_TIMER |
+            VMX_PIN_BASED_POSTED_INTR,
+        .features[FEAT_VMX_PROCBASED_CTLS] =
+            VMX_CPU_BASED_VIRTUAL_INTR_PENDING |
+            VMX_CPU_BASED_USE_TSC_OFFSETING | VMX_CPU_BASED_HLT_EXITING |
+            VMX_CPU_BASED_INVLPG_EXITING | VMX_CPU_BASED_MWAIT_EXITING |
+            VMX_CPU_BASED_RDPMC_EXITING | VMX_CPU_BASED_RDTSC_EXITING |
+            VMX_CPU_BASED_CR3_LOAD_EXITING | VMX_CPU_BASED_CR3_STORE_EXITING |
+            VMX_CPU_BASED_CR8_LOAD_EXITING | VMX_CPU_BASED_CR8_STORE_EXITING |
+            VMX_CPU_BASED_TPR_SHADOW | VMX_CPU_BASED_VIRTUAL_NMI_PENDING |
+            VMX_CPU_BASED_MOV_DR_EXITING | VMX_CPU_BASED_UNCOND_IO_EXITING |
+            VMX_CPU_BASED_USE_IO_BITMAPS | VMX_CPU_BASED_MONITOR_TRAP_FLAG |
+            VMX_CPU_BASED_USE_MSR_BITMAPS | VMX_CPU_BASED_MONITOR_EXITING |
+            VMX_CPU_BASED_PAUSE_EXITING |
+            VMX_CPU_BASED_ACTIVATE_SECONDARY_CONTROLS,
+        .features[FEAT_VMX_SECONDARY_CTLS] =
+            VMX_SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
+            VMX_SECONDARY_EXEC_ENABLE_EPT | VMX_SECONDARY_EXEC_DESC |
+            VMX_SECONDARY_EXEC_RDTSCP |
+            VMX_SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
+            VMX_SECONDARY_EXEC_ENABLE_VPID | VMX_SECONDARY_EXEC_WBINVD_EXITING |
+            VMX_SECONDARY_EXEC_UNRESTRICTED_GUEST |
+            VMX_SECONDARY_EXEC_APIC_REGISTER_VIRT |
+            VMX_SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
+            VMX_SECONDARY_EXEC_RDRAND_EXITING |
+            VMX_SECONDARY_EXEC_ENABLE_INVPCID |
+            VMX_SECONDARY_EXEC_ENABLE_VMFUNC | VMX_SECONDARY_EXEC_SHADOW_VMCS |
+            VMX_SECONDARY_EXEC_RDSEED_EXITING | VMX_SECONDARY_EXEC_ENABLE_PML |
+            VMX_SECONDARY_EXEC_XSAVES,
+        .features[FEAT_VMX_VMFUNC] =
+            MSR_VMX_VMFUNC_EPT_SWITCHING,
+        .xlevel = 0x80000008,
+        .model_id = "Intel Xeon Processor (SapphireRapids)",
+        .versions = (X86CPUVersionDefinition[]) {
+            { .version = 1 },
+            { /* end of list */ },
+        },
+    },
     {
         .name = "Denverton",
         .level = 21,
target/i386/cpu.h +4 -0
index 82004b65b9..ef3e8a5ed5 100644
@@ -879,10 +879,14 @@ uint64_t x86_cpu_get_supported_feature_word(FeatureWord w,
 #define CPUID_7_0_EDX_TSX_LDTRK         (1U << 16)
 /* Architectural LBRs */
 #define CPUID_7_0_EDX_ARCH_LBR          (1U << 19)
+/* AMX_BF16 instruction */
+#define CPUID_7_0_EDX_AMX_BF16          (1U << 22)
 /* AVX512_FP16 instruction */
 #define CPUID_7_0_EDX_AVX512_FP16       (1U << 23)
 /* AMX tile (two-dimensional register) */
 #define CPUID_7_0_EDX_AMX_TILE          (1U << 24)
+/* AMX_INT8 instruction */
+#define CPUID_7_0_EDX_AMX_INT8          (1U << 25)
 /* Speculation Control */
 #define CPUID_7_0_EDX_SPEC_CTRL         (1U << 26)
 /* Single Thread Indirect Branch Predictors */
Signature · 1 line
2.34.1
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
#3 lore raw
Quoted 31 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines
Quoted 9 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?
In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

We shouldn't assume this will only be used on 'SappireRapids' host silicon. Thi named CPU model is likely to be used by a guest running on any host silicon generations that follow SappireRapids too.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#4 lore raw
Quoted 42 lines · * Daniel P. Berrangé ([email protected]) wrote:
Quoted 31 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines
Quoted 9 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?
In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

Quoted 3 lines

We shouldn't assume this will only be used on 'SappireRapids' host silicon. Thi named CPU model is likely to be used by a guest running on any host silicon generations that follow SappireRapids too.

Indeed, but I wanted to check the opposite question first; whether all SappireRapids had the same sizes; I think you're asking the opposite question.

Dave

Quoted 7 lines

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
#5 lore raw
In reply to Dr. David Alan Gilbert
Quoted 35 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?

Quoted 7 lines

In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

There is only one palette in current design. This palette include 8 tiles. Those two CPUID leafs defined bytes_per_tile, total_tile_bytes, max_rows and etc, the AMX tool will configure those values into TILECFG with ldtilecfg instrcutions. Once tiles are configured, we can use tileload instruction to load data into those tiles.

We did migration between two SappireRapids with amx self test tool (tools/testing/selftests/x86/amx.c)started in two sides, the migration work well.

As for SappireRapids and more newer cpu types, those two CPUID leafs definitions are all same on AMX. So, on AMX perspective, the migration should be workable on subsequent cpu types. thanks!

Yang

Quoted 173 lines

Dave

Quoted 167 lines
---
 target/i386/cpu.c | 128 ++++++++++++++++++++++++++++++++++++++++++++++
 target/i386/cpu.h |   4 ++
 2 files changed, 132 insertions(+)
target/i386/cpu.c +128 -0
index 1db1278a59..abb43853d4 100644
@@ -3467,6 +3467,134 @@ static const X86CPUDefinition builtin_x86_defs[] = {
             { /* end of list */ }
         }
     },
+    {
+        .name = "SapphireRapids",
+        .level = 0x20,
+        .vendor = CPUID_VENDOR_INTEL,
+        .family = 6,
+        .model = 143,
+        .stepping = 4,
+        /*
+         * please keep the ascending order so that we can have a clear view of
+         * bit position of each feature.
+         */
+        .features[FEAT_1_EDX] =
+            CPUID_FP87 | CPUID_VME | CPUID_DE | CPUID_PSE | CPUID_TSC |
+            CPUID_MSR | CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC |
+            CPUID_SEP | CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV |
+            CPUID_PAT | CPUID_PSE36 | CPUID_CLFLUSH | CPUID_MMX | CPUID_FXSR |
+            CPUID_SSE | CPUID_SSE2,
+        .features[FEAT_1_ECX] =
+            CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | CPUID_EXT_SSSE3 |
+            CPUID_EXT_FMA | CPUID_EXT_CX16 | CPUID_EXT_PCID | CPUID_EXT_SSE41 |
+            CPUID_EXT_SSE42 | CPUID_EXT_X2APIC | CPUID_EXT_MOVBE |
+            CPUID_EXT_POPCNT | CPUID_EXT_TSC_DEADLINE_TIMER | CPUID_EXT_AES |
+            CPUID_EXT_XSAVE | CPUID_EXT_AVX | CPUID_EXT_F16C | CPUID_EXT_RDRAND,
+        .features[FEAT_8000_0001_EDX] =
+            CPUID_EXT2_SYSCALL | CPUID_EXT2_NX | CPUID_EXT2_PDPE1GB |
+            CPUID_EXT2_RDTSCP | CPUID_EXT2_LM,
+        .features[FEAT_8000_0001_ECX] =
+            CPUID_EXT3_LAHF_LM | CPUID_EXT3_ABM | CPUID_EXT3_3DNOWPREFETCH,
+        .features[FEAT_8000_0008_EBX] =
+            CPUID_8000_0008_EBX_WBNOINVD,
+        .features[FEAT_7_0_EBX] =
+            CPUID_7_0_EBX_FSGSBASE | CPUID_7_0_EBX_BMI1 | CPUID_7_0_EBX_HLE |
+            CPUID_7_0_EBX_AVX2 | CPUID_7_0_EBX_SMEP | CPUID_7_0_EBX_BMI2 |
+            CPUID_7_0_EBX_ERMS | CPUID_7_0_EBX_INVPCID | CPUID_7_0_EBX_RTM |
+            CPUID_7_0_EBX_AVX512F | CPUID_7_0_EBX_AVX512DQ |
+            CPUID_7_0_EBX_RDSEED | CPUID_7_0_EBX_ADX | CPUID_7_0_EBX_SMAP |
+            CPUID_7_0_EBX_AVX512IFMA | CPUID_7_0_EBX_CLFLUSHOPT |
+            CPUID_7_0_EBX_CLWB | CPUID_7_0_EBX_AVX512CD | CPUID_7_0_EBX_SHA_NI |
+            CPUID_7_0_EBX_AVX512BW | CPUID_7_0_EBX_AVX512VL,
+        .features[FEAT_7_0_ECX] =
+            CPUID_7_0_ECX_AVX512_VBMI | CPUID_7_0_ECX_UMIP | CPUID_7_0_ECX_PKU |
+            CPUID_7_0_ECX_AVX512_VBMI2 | CPUID_7_0_ECX_GFNI |
+            CPUID_7_0_ECX_VAES | CPUID_7_0_ECX_VPCLMULQDQ |
+            CPUID_7_0_ECX_AVX512VNNI | CPUID_7_0_ECX_AVX512BITALG |
+            CPUID_7_0_ECX_AVX512_VPOPCNTDQ | CPUID_7_0_ECX_LA57 |
+            CPUID_7_0_ECX_RDPID | CPUID_7_0_ECX_BUS_LOCK_DETECT,
+        .features[FEAT_7_0_EDX] =
+            CPUID_7_0_EDX_FSRM | CPUID_7_0_EDX_SERIALIZE |
+            CPUID_7_0_EDX_TSX_LDTRK | CPUID_7_0_EDX_AMX_BF16 |
+            CPUID_7_0_EDX_AVX512_FP16 | CPUID_7_0_EDX_AMX_TILE |
+            CPUID_7_0_EDX_AMX_INT8 | CPUID_7_0_EDX_SPEC_CTRL |
+            CPUID_7_0_EDX_ARCH_CAPABILITIES | CPUID_7_0_EDX_SPEC_CTRL_SSBD,
+        .features[FEAT_ARCH_CAPABILITIES] =
+            MSR_ARCH_CAP_RDCL_NO | MSR_ARCH_CAP_IBRS_ALL |
+            MSR_ARCH_CAP_SKIP_L1DFL_VMENTRY | MSR_ARCH_CAP_MDS_NO |
+            MSR_ARCH_CAP_PSCHANGE_MC_NO | MSR_ARCH_CAP_TAA_NO,
+        .features[FEAT_XSAVE] =
+            CPUID_XSAVE_XSAVEOPT | CPUID_XSAVE_XSAVEC |
+            CPUID_XSAVE_XGETBV1 | CPUID_XSAVE_XSAVES | CPUID_D_1_EAX_XFD,
+        .features[FEAT_6_EAX] =
+            CPUID_6_EAX_ARAT,
+        .features[FEAT_7_1_EAX] =
+            CPUID_7_1_EAX_AVX_VNNI | CPUID_7_1_EAX_AVX512_BF16,
+        .features[FEAT_VMX_BASIC] =
+            MSR_VMX_BASIC_INS_OUTS | MSR_VMX_BASIC_TRUE_CTLS,
+        .features[FEAT_VMX_ENTRY_CTLS] =
+            VMX_VM_ENTRY_LOAD_DEBUG_CONTROLS | VMX_VM_ENTRY_IA32E_MODE |
+            VMX_VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_ENTRY_LOAD_IA32_PAT | VMX_VM_ENTRY_LOAD_IA32_EFER,
+        .features[FEAT_VMX_EPT_VPID_CAPS] =
+            MSR_VMX_EPT_EXECONLY |
+            MSR_VMX_EPT_PAGE_WALK_LENGTH_4 | MSR_VMX_EPT_PAGE_WALK_LENGTH_5 |
+            MSR_VMX_EPT_WB | MSR_VMX_EPT_2MB | MSR_VMX_EPT_1GB |
+            MSR_VMX_EPT_INVEPT | MSR_VMX_EPT_AD_BITS |
+            MSR_VMX_EPT_INVEPT_SINGLE_CONTEXT | MSR_VMX_EPT_INVEPT_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID | MSR_VMX_EPT_INVVPID_SINGLE_ADDR |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT |
+            MSR_VMX_EPT_INVVPID_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT_NOGLOBALS,
+        .features[FEAT_VMX_EXIT_CTLS] =
+            VMX_VM_EXIT_SAVE_DEBUG_CONTROLS |
+            VMX_VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_EXIT_ACK_INTR_ON_EXIT | VMX_VM_EXIT_SAVE_IA32_PAT |
+            VMX_VM_EXIT_LOAD_IA32_PAT | VMX_VM_EXIT_SAVE_IA32_EFER |
+            VMX_VM_EXIT_LOAD_IA32_EFER | VMX_VM_EXIT_SAVE_VMX_PREEMPTION_TIMER,
+        .features[FEAT_VMX_MISC] =
+            MSR_VMX_MISC_STORE_LMA | MSR_VMX_MISC_ACTIVITY_HLT |
+            MSR_VMX_MISC_VMWRITE_VMEXIT,
+        .features[FEAT_VMX_PINBASED_CTLS] =
+            VMX_PIN_BASED_EXT_INTR_MASK | VMX_PIN_BASED_NMI_EXITING |
+            VMX_PIN_BASED_VIRTUAL_NMIS | VMX_PIN_BASED_VMX_PREEMPTION_TIMER |
+            VMX_PIN_BASED_POSTED_INTR,
+        .features[FEAT_VMX_PROCBASED_CTLS] =
+            VMX_CPU_BASED_VIRTUAL_INTR_PENDING |
+            VMX_CPU_BASED_USE_TSC_OFFSETING | VMX_CPU_BASED_HLT_EXITING |
+            VMX_CPU_BASED_INVLPG_EXITING | VMX_CPU_BASED_MWAIT_EXITING |
+            VMX_CPU_BASED_RDPMC_EXITING | VMX_CPU_BASED_RDTSC_EXITING |
+            VMX_CPU_BASED_CR3_LOAD_EXITING | VMX_CPU_BASED_CR3_STORE_EXITING |
+            VMX_CPU_BASED_CR8_LOAD_EXITING | VMX_CPU_BASED_CR8_STORE_EXITING |
+            VMX_CPU_BASED_TPR_SHADOW | VMX_CPU_BASED_VIRTUAL_NMI_PENDING |
+            VMX_CPU_BASED_MOV_DR_EXITING | VMX_CPU_BASED_UNCOND_IO_EXITING |
+            VMX_CPU_BASED_USE_IO_BITMAPS | VMX_CPU_BASED_MONITOR_TRAP_FLAG |
+            VMX_CPU_BASED_USE_MSR_BITMAPS | VMX_CPU_BASED_MONITOR_EXITING |
+            VMX_CPU_BASED_PAUSE_EXITING |
+            VMX_CPU_BASED_ACTIVATE_SECONDARY_CONTROLS,
+        .features[FEAT_VMX_SECONDARY_CTLS] =
+            VMX_SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
+            VMX_SECONDARY_EXEC_ENABLE_EPT | VMX_SECONDARY_EXEC_DESC |
+            VMX_SECONDARY_EXEC_RDTSCP |
+            VMX_SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
+            VMX_SECONDARY_EXEC_ENABLE_VPID | VMX_SECONDARY_EXEC_WBINVD_EXITING |
+            VMX_SECONDARY_EXEC_UNRESTRICTED_GUEST |
+            VMX_SECONDARY_EXEC_APIC_REGISTER_VIRT |
+            VMX_SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
+            VMX_SECONDARY_EXEC_RDRAND_EXITING |
+            VMX_SECONDARY_EXEC_ENABLE_INVPCID |
+            VMX_SECONDARY_EXEC_ENABLE_VMFUNC | VMX_SECONDARY_EXEC_SHADOW_VMCS |
+            VMX_SECONDARY_EXEC_RDSEED_EXITING | VMX_SECONDARY_EXEC_ENABLE_PML |
+            VMX_SECONDARY_EXEC_XSAVES,
+        .features[FEAT_VMX_VMFUNC] =
+            MSR_VMX_VMFUNC_EPT_SWITCHING,
+        .xlevel = 0x80000008,
+        .model_id = "Intel Xeon Processor (SapphireRapids)",
+        .versions = (X86CPUVersionDefinition[]) {
+            { .version = 1 },
+            { /* end of list */ },
+        },
+    },
     {
         .name = "Denverton",
         .level = 21,
target/i386/cpu.h +4 -0
index 82004b65b9..ef3e8a5ed5 100644
@@ -879,10 +879,14 @@ uint64_t x86_cpu_get_supported_feature_word(FeatureWord w,
 #define CPUID_7_0_EDX_TSX_LDTRK         (1U << 16)
 /* Architectural LBRs */
 #define CPUID_7_0_EDX_ARCH_LBR          (1U << 19)
+/* AMX_BF16 instruction */
+#define CPUID_7_0_EDX_AMX_BF16          (1U << 22)
 /* AVX512_FP16 instruction */
 #define CPUID_7_0_EDX_AVX512_FP16       (1U << 23)
 /* AMX tile (two-dimensional register) */
 #define CPUID_7_0_EDX_AMX_TILE          (1U << 24)
+/* AMX_INT8 instruction */
+#define CPUID_7_0_EDX_AMX_INT8          (1U << 25)
 /* Speculation Control */
 #define CPUID_7_0_EDX_SPEC_CTRL         (1U << 26)
 /* Single Thread Indirect Branch Predictors */
Signature · 1 line
2.34.1
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
#6 lore raw
Quoted 50 lines · Yang Zhong
Quoted 35 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?

Quoted 7 lines

In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

There is only one palette in current design. This palette include 8 tiles. Those two CPUID leafs defined bytes_per_tile, total_tile_bytes, max_rows and etc, the AMX tool will configure those values into TILECFG with ldtilecfg instrcutions. Once tiles are configured, we can use tileload instruction to load data into those tiles.

Quoted 6 lines

We did migration between two SappireRapids with amx self test tool (tools/testing/selftests/x86/amx.c)started in two sides, the migration work well.

As for SappireRapids and more newer cpu types, those two CPUID leafs definitions are all same on AMX.

I'm not sure what definitions mean here. Are you saying the CPUID values of leaf 0x1D and 0x1E won't change for any future Intel Silicion?

Personally, I doubt it. And we shouldn't take such assumption unless Intel states it SDM.

Quoted 2 lines

So, on AMX perspective, the migration
should be workable on subsequent cpu types. thanks!

I think what Dave worried is that when migrating one VM created with "SapphireRapids" model on SPR machine to some newer platform in the future, where the newer platform reports different value on CPUID leaves 0x1D and 0x1E than SPR platform.

I think we need to contain CPUID leaves 0x1D and 0x1E into CPU model as well. Otherwise we will hit the same as Intel PT that SPR reports less capabilities that ICX.

#7 lore raw
Quoted 54 lines · Xiaoyao Li
Quoted 53 lines · Yang Zhong
Quoted 31 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines
Quoted 10 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?
In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

There is only one palette in current design. This palette include 8 tiles. Those two CPUID leafs defined bytes_per_tile, total_tile_bytes, max_rows and etc, the AMX tool will configure those values into TILECFG with ldtilecfg instrcutions. Once tiles are configured, we can use tileload instruction to load data into those tiles.

We did migration between two SappireRapids with amx self test tool (tools/testing/selftests/x86/amx.c)started in two sides, the migration work well.

Quoted 8 lines
Quoted 2 lines

As for SappireRapids and more newer cpu types, those two CPUID leafs definitions are all same on AMX.

I'm not sure what definitions mean here. Are you saying the CPUID values of leaf 0x1D and 0x1E won't change for any future Intel Silicion?

Personally, I doubt it. And we shouldn't take such assumption unless Intel states it SDM.

The current 0x1D and 0x1E definitions as below:

  /* CPUID Leaf 0x1D constants: */
  #define INTEL_AMX_TILE_MAX_SUBLEAF     0x1
  #define INTEL_AMX_TOTAL_TILE_BYTES     0x2000
  #define INTEL_AMX_BYTES_PER_TILE       0x400
  #define INTEL_AMX_BYTES_PER_ROW        0x40
  #define INTEL_AMX_TILE_MAX_NAMES       0x8
  #define INTEL_AMX_TILE_MAX_ROWS        0x10

  /* CPUID Leaf 0x1E constants: */
  #define INTEL_AMX_TMUL_MAX_K           0x10
  #define INTEL_AMX_TMUL_MAX_N           0x40

These values are defined from SDM, and from the new developping CPU, these values are still same with SappireRapids. thanks!

Yang

Quoted 13 lines
Quoted 3 lines

So, on AMX perspective, the migration
should be workable on subsequent cpu types. thanks!

I think what Dave worried is that when migrating one VM created with "SapphireRapids" model on SPR machine to some newer platform in the future, where the newer platform reports different value on CPUID leaves 0x1D and 0x1E than SPR platform.

I think we need to contain CPUID leaves 0x1D and 0x1E into CPU model as well. Otherwise we will hit the same as Intel PT that SPR reports less capabilities that ICX.

#8 lore raw
Quoted 78 lines · * Yang Zhong ([email protected]) wrote:
Quoted 54 lines · Xiaoyao Li
Quoted 53 lines · Yang Zhong
Quoted 31 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines
Quoted 10 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?
In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

There is only one palette in current design. This palette include 8 tiles. Those two CPUID leafs defined bytes_per_tile, total_tile_bytes, max_rows and etc, the AMX tool will configure those values into TILECFG with ldtilecfg instrcutions. Once tiles are configured, we can use tileload instruction to load data into those tiles.

We did migration between two SappireRapids with amx self test tool (tools/testing/selftests/x86/amx.c)started in two sides, the migration work well.

Quoted 8 lines
Quoted 2 lines

As for SappireRapids and more newer cpu types, those two CPUID leafs definitions are all same on AMX.

I'm not sure what definitions mean here. Are you saying the CPUID values of leaf 0x1D and 0x1E won't change for any future Intel Silicion?

Personally, I doubt it. And we shouldn't take such assumption unless Intel states it SDM.

The current 0x1D and 0x1E definitions as below:

  /* CPUID Leaf 0x1D constants: */
  #define INTEL_AMX_TILE_MAX_SUBLEAF     0x1
  #define INTEL_AMX_TOTAL_TILE_BYTES     0x2000
  #define INTEL_AMX_BYTES_PER_TILE       0x400
  #define INTEL_AMX_BYTES_PER_ROW        0x40
  #define INTEL_AMX_TILE_MAX_NAMES       0x8
  #define INTEL_AMX_TILE_MAX_ROWS        0x10

  /* CPUID Leaf 0x1E constants: */
  #define INTEL_AMX_TMUL_MAX_K           0x10
  #define INTEL_AMX_TMUL_MAX_N           0x40
Quoted 2 lines

These values are defined from SDM, and from the new developping CPU, these values are still same with SappireRapids. thanks!

But there's nothing stopping them increasing in future versions ?

Dave

Quoted 15 lines

Yang

Quoted 13 lines
Quoted 3 lines

So, on AMX perspective, the migration
should be workable on subsequent cpu types. thanks!

I think what Dave worried is that when migrating one VM created with "SapphireRapids" model on SPR machine to some newer platform in the future, where the newer platform reports different value on CPUID leaves 0x1D and 0x1E than SPR platform.

I think we need to contain CPUID leaves 0x1D and 0x1E into CPU model as well. Otherwise we will hit the same as Intel PT that SPR reports less capabilities that ICX.

Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
#9 lore raw
Quoted 75 lines · Dr. David Alan Gilbert
Quoted 74 lines · * Yang Zhong ([email protected]) wrote:
Quoted 54 lines · Xiaoyao Li
Quoted 53 lines · Yang Zhong
Quoted 31 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 2 lines
Quoted 10 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?
In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?
Would you expect all devices called a 'SappireRapids' to have the same
sizes?

There is only one palette in current design. This palette include 8 tiles. Those two CPUID leafs defined bytes_per_tile, total_tile_bytes, max_rows and etc, the AMX tool will configure those values into TILECFG with ldtilecfg instrcutions. Once tiles are configured, we can use tileload instruction to load data into those tiles.

We did migration between two SappireRapids with amx self test tool (tools/testing/selftests/x86/amx.c)started in two sides, the migration work well.

Quoted 8 lines
Quoted 2 lines

As for SappireRapids and more newer cpu types, those two CPUID leafs definitions are all same on AMX.

I'm not sure what definitions mean here. Are you saying the CPUID values of leaf 0x1D and 0x1E won't change for any future Intel Silicion?

Personally, I doubt it. And we shouldn't take such assumption unless Intel states it SDM.

The current 0x1D and 0x1E definitions as below:

  /* CPUID Leaf 0x1D constants: */
  #define INTEL_AMX_TILE_MAX_SUBLEAF     0x1
  #define INTEL_AMX_TOTAL_TILE_BYTES     0x2000
  #define INTEL_AMX_BYTES_PER_TILE       0x400
  #define INTEL_AMX_BYTES_PER_ROW        0x40
  #define INTEL_AMX_TILE_MAX_NAMES       0x8
  #define INTEL_AMX_TILE_MAX_ROWS        0x10
Quoted 9 lines
Quoted 6 lines
  /* CPUID Leaf 0x1E constants: */
  #define INTEL_AMX_TMUL_MAX_K           0x10
  #define INTEL_AMX_TMUL_MAX_N           0x40

These values are defined from SDM, and from the new developping CPU, these values are still same with SappireRapids. thanks!

But there's nothing stopping them increasing in future versions ?

Okay, thanks! We will add these CPUID leafs in this cpu model.

Yang

Quoted 21 lines

Dave

Quoted 15 lines

Yang

Quoted 13 lines
Quoted 3 lines

So, on AMX perspective, the migration
should be workable on subsequent cpu types. thanks!

I think what Dave worried is that when migrating one VM created with "SapphireRapids" model on SPR machine to some newer platform in the future, where the newer platform reports different value on CPUID leaves 0x1D and 0x1E than SPR platform.

I think we need to contain CPUID leaves 0x1D and 0x1E into CPU model as well. Otherwise we will hit the same as Intel PT that SPR reports less capabilities that ICX.

Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
#10 lore raw
In reply to Dr. David Alan Gilbert
Quoted 28 lines · Dr. David Alan Gilbert
Quoted 27 lines · * Wang, Lei ([email protected]) wrote:

The new CPU model mostly inherits features from Icelake-Server, while adding new features:
- AMX (Advance Matrix eXtensions)
- Bus Lock Debug Exception
and new instructions:
- AVX VNNI (Vector Neural Network Instruction):
- VPDPBUS: Multiply and Add Unsigned and Signed Bytes
- VPDPBUSDS: Multiply and Add Unsigned and Signed Bytes with Saturation
- VPDPWSSD: Multiply and Add Signed Word Integers
- VPDPWSSDS: Multiply and Add Signed Integers with Saturation
- FP16: Replicates existing AVX512 computational SP (FP32) instructions using FP16 instead of FP32 for ~2X performance gain
- SERIALIZE: Provide software with a simple way to force the processor to complete all modifications, faster, allowed in all privilege levels and not causing an unconditional VM exit
- TSX Suspend Load Address Tracking: Allows programmers to choose which memory accesses do not need to be tracked in the TSX read set
- AVX512_BF16: Vector Neural Network Instructions supporting BFLOAT16 inputs and conversion instructions from IEEE single precision

Features may be added in future versions:
- CET (virtualization support hasn't been merged)
Instructions may be added in future versions:
- fast zero-length MOVSB (KVM doesn't support yet)
- fast short STOSB (KVM doesn't support yet)
- fast short CMPSB, SCASB (KVM doesn't support yet)

Quoted 6 lines
Quoted 2 lines

Hi,
What fills in the AMX tile and tmul information leafs
(0x1D, 0x1E)?

Current QEMU hard-codes the value of AMX tile and tmul information leafs (0x1D, 0x1E) if AMX is exposed to guest. In cpu_x86_cpuid(), target/i386/cpu.c

case 0x1D: {
        /* AMX TILE */
        *eax = 0;
        *ebx = 0;
        *ecx = 0;
        *edx = 0;
        if (!(env->features[FEAT_7_0_EDX] & CPUID_7_0_EDX_AMX_TILE)) {
            break;
        }

        if (count == 0) {
            /* Highest numbered palette subleaf */
            *eax = INTEL_AMX_TILE_MAX_SUBLEAF;
        } else if (count == 1) {
            *eax = INTEL_AMX_TOTAL_TILE_BYTES |
                   (INTEL_AMX_BYTES_PER_TILE << 16);
            *ebx = INTEL_AMX_BYTES_PER_ROW | (INTEL_AMX_TILE_MAX_NAMES 
<< 16);
            *ecx = INTEL_AMX_TILE_MAX_ROWS;
        }
        break;
    }
    case 0x1E: {
        /* AMX TMUL */
        *eax = 0;
        *ebx = 0;
        *ecx = 0;
        *edx = 0;
        if (!(env->features[FEAT_7_0_EDX] & CPUID_7_0_EDX_AMX_TILE)) {
            break;
        }

        if (count == 0) {
            /* Highest numbered palette subleaf */
            *ebx = INTEL_AMX_TMUL_MAX_K | (INTEL_AMX_TMUL_MAX_N << 8);
        }
        break;
    }
Quoted 4 lines

In particular, how would we make sure when we migrate between two
generations of AMX/Tile/Tmul capable devices with different register/palette/tmul limits that the migration is tied to the CPU type correctly?

Since they'are hard-coded. The value of guest never change no matter what HW is.

Quoted 2 lines

Would you expect all devices called a 'SappireRapids' to have the same
sizes?

I suppose here the devices you mean HW platform? If so, Intel commits that palette 1 value (CPUID leaf 0x1d, subleaf 0x1) will never change. And TMUL capability (CPUID leaf 0x1e) are constant for a few generations after SPR.

But this has no impact on migration safety as long as CPUID value of leaf 0x1d and 0x1e are tied to a named CPU model and doesn't vary on different hosts. And current QEMU code satisfy it since the values are hard-coded.

So, IMHO, it seems OK to define AMX in SPR cpu model with current QEMU as this patch does. Although hard-coded value of 0x1E seems to have potential issue if far future product reports smaller value (Intel only can commit it doesn't change for a few generations so far), it's another thing we can handle separately for AMX.

Dave and Paolo, what do you think?

Quoted 169 lines

Dave

Quoted 167 lines
---
  target/i386/cpu.c | 128 ++++++++++++++++++++++++++++++++++++++++++++++
  target/i386/cpu.h |   4 ++
  2 files changed, 132 insertions(+)
target/i386/cpu.c +128 -0
index 1db1278a59..abb43853d4 100644
@@ -3467,6 +3467,134 @@ static const X86CPUDefinition builtin_x86_defs[] = {
              { /* end of list */ }
          }
      },
+    {
+        .name = "SapphireRapids",
+        .level = 0x20,
+        .vendor = CPUID_VENDOR_INTEL,
+        .family = 6,
+        .model = 143,
+        .stepping = 4,
+        /*
+         * please keep the ascending order so that we can have a clear view of
+         * bit position of each feature.
+         */
+        .features[FEAT_1_EDX] =
+            CPUID_FP87 | CPUID_VME | CPUID_DE | CPUID_PSE | CPUID_TSC |
+            CPUID_MSR | CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC |
+            CPUID_SEP | CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV |
+            CPUID_PAT | CPUID_PSE36 | CPUID_CLFLUSH | CPUID_MMX | CPUID_FXSR |
+            CPUID_SSE | CPUID_SSE2,
+        .features[FEAT_1_ECX] =
+            CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | CPUID_EXT_SSSE3 |
+            CPUID_EXT_FMA | CPUID_EXT_CX16 | CPUID_EXT_PCID | CPUID_EXT_SSE41 |
+            CPUID_EXT_SSE42 | CPUID_EXT_X2APIC | CPUID_EXT_MOVBE |
+            CPUID_EXT_POPCNT | CPUID_EXT_TSC_DEADLINE_TIMER | CPUID_EXT_AES |
+            CPUID_EXT_XSAVE | CPUID_EXT_AVX | CPUID_EXT_F16C | CPUID_EXT_RDRAND,
+        .features[FEAT_8000_0001_EDX] =
+            CPUID_EXT2_SYSCALL | CPUID_EXT2_NX | CPUID_EXT2_PDPE1GB |
+            CPUID_EXT2_RDTSCP | CPUID_EXT2_LM,
+        .features[FEAT_8000_0001_ECX] =
+            CPUID_EXT3_LAHF_LM | CPUID_EXT3_ABM | CPUID_EXT3_3DNOWPREFETCH,
+        .features[FEAT_8000_0008_EBX] =
+            CPUID_8000_0008_EBX_WBNOINVD,
+        .features[FEAT_7_0_EBX] =
+            CPUID_7_0_EBX_FSGSBASE | CPUID_7_0_EBX_BMI1 | CPUID_7_0_EBX_HLE |
+            CPUID_7_0_EBX_AVX2 | CPUID_7_0_EBX_SMEP | CPUID_7_0_EBX_BMI2 |
+            CPUID_7_0_EBX_ERMS | CPUID_7_0_EBX_INVPCID | CPUID_7_0_EBX_RTM |
+            CPUID_7_0_EBX_AVX512F | CPUID_7_0_EBX_AVX512DQ |
+            CPUID_7_0_EBX_RDSEED | CPUID_7_0_EBX_ADX | CPUID_7_0_EBX_SMAP |
+            CPUID_7_0_EBX_AVX512IFMA | CPUID_7_0_EBX_CLFLUSHOPT |
+            CPUID_7_0_EBX_CLWB | CPUID_7_0_EBX_AVX512CD | CPUID_7_0_EBX_SHA_NI |
+            CPUID_7_0_EBX_AVX512BW | CPUID_7_0_EBX_AVX512VL,
+        .features[FEAT_7_0_ECX] =
+            CPUID_7_0_ECX_AVX512_VBMI | CPUID_7_0_ECX_UMIP | CPUID_7_0_ECX_PKU |
+            CPUID_7_0_ECX_AVX512_VBMI2 | CPUID_7_0_ECX_GFNI |
+            CPUID_7_0_ECX_VAES | CPUID_7_0_ECX_VPCLMULQDQ |
+            CPUID_7_0_ECX_AVX512VNNI | CPUID_7_0_ECX_AVX512BITALG |
+            CPUID_7_0_ECX_AVX512_VPOPCNTDQ | CPUID_7_0_ECX_LA57 |
+            CPUID_7_0_ECX_RDPID | CPUID_7_0_ECX_BUS_LOCK_DETECT,
+        .features[FEAT_7_0_EDX] =
+            CPUID_7_0_EDX_FSRM | CPUID_7_0_EDX_SERIALIZE |
+            CPUID_7_0_EDX_TSX_LDTRK | CPUID_7_0_EDX_AMX_BF16 |
+            CPUID_7_0_EDX_AVX512_FP16 | CPUID_7_0_EDX_AMX_TILE |
+            CPUID_7_0_EDX_AMX_INT8 | CPUID_7_0_EDX_SPEC_CTRL |
+            CPUID_7_0_EDX_ARCH_CAPABILITIES | CPUID_7_0_EDX_SPEC_CTRL_SSBD,
+        .features[FEAT_ARCH_CAPABILITIES] =
+            MSR_ARCH_CAP_RDCL_NO | MSR_ARCH_CAP_IBRS_ALL |
+            MSR_ARCH_CAP_SKIP_L1DFL_VMENTRY | MSR_ARCH_CAP_MDS_NO |
+            MSR_ARCH_CAP_PSCHANGE_MC_NO | MSR_ARCH_CAP_TAA_NO,
+        .features[FEAT_XSAVE] =
+            CPUID_XSAVE_XSAVEOPT | CPUID_XSAVE_XSAVEC |
+            CPUID_XSAVE_XGETBV1 | CPUID_XSAVE_XSAVES | CPUID_D_1_EAX_XFD,
+        .features[FEAT_6_EAX] =
+            CPUID_6_EAX_ARAT,
+        .features[FEAT_7_1_EAX] =
+            CPUID_7_1_EAX_AVX_VNNI | CPUID_7_1_EAX_AVX512_BF16,
+        .features[FEAT_VMX_BASIC] =
+            MSR_VMX_BASIC_INS_OUTS | MSR_VMX_BASIC_TRUE_CTLS,
+        .features[FEAT_VMX_ENTRY_CTLS] =
+            VMX_VM_ENTRY_LOAD_DEBUG_CONTROLS | VMX_VM_ENTRY_IA32E_MODE |
+            VMX_VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_ENTRY_LOAD_IA32_PAT | VMX_VM_ENTRY_LOAD_IA32_EFER,
+        .features[FEAT_VMX_EPT_VPID_CAPS] =
+            MSR_VMX_EPT_EXECONLY |
+            MSR_VMX_EPT_PAGE_WALK_LENGTH_4 | MSR_VMX_EPT_PAGE_WALK_LENGTH_5 |
+            MSR_VMX_EPT_WB | MSR_VMX_EPT_2MB | MSR_VMX_EPT_1GB |
+            MSR_VMX_EPT_INVEPT | MSR_VMX_EPT_AD_BITS |
+            MSR_VMX_EPT_INVEPT_SINGLE_CONTEXT | MSR_VMX_EPT_INVEPT_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID | MSR_VMX_EPT_INVVPID_SINGLE_ADDR |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT |
+            MSR_VMX_EPT_INVVPID_ALL_CONTEXT |
+            MSR_VMX_EPT_INVVPID_SINGLE_CONTEXT_NOGLOBALS,
+        .features[FEAT_VMX_EXIT_CTLS] =
+            VMX_VM_EXIT_SAVE_DEBUG_CONTROLS |
+            VMX_VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
+            VMX_VM_EXIT_ACK_INTR_ON_EXIT | VMX_VM_EXIT_SAVE_IA32_PAT |
+            VMX_VM_EXIT_LOAD_IA32_PAT | VMX_VM_EXIT_SAVE_IA32_EFER |
+            VMX_VM_EXIT_LOAD_IA32_EFER | VMX_VM_EXIT_SAVE_VMX_PREEMPTION_TIMER,
+        .features[FEAT_VMX_MISC] =
+            MSR_VMX_MISC_STORE_LMA | MSR_VMX_MISC_ACTIVITY_HLT |
+            MSR_VMX_MISC_VMWRITE_VMEXIT,
+        .features[FEAT_VMX_PINBASED_CTLS] =
+            VMX_PIN_BASED_EXT_INTR_MASK | VMX_PIN_BASED_NMI_EXITING |
+            VMX_PIN_BASED_VIRTUAL_NMIS | VMX_PIN_BASED_VMX_PREEMPTION_TIMER |
+            VMX_PIN_BASED_POSTED_INTR,
+        .features[FEAT_VMX_PROCBASED_CTLS] =
+            VMX_CPU_BASED_VIRTUAL_INTR_PENDING |
+            VMX_CPU_BASED_USE_TSC_OFFSETING | VMX_CPU_BASED_HLT_EXITING |
+            VMX_CPU_BASED_INVLPG_EXITING | VMX_CPU_BASED_MWAIT_EXITING |
+            VMX_CPU_BASED_RDPMC_EXITING | VMX_CPU_BASED_RDTSC_EXITING |
+            VMX_CPU_BASED_CR3_LOAD_EXITING | VMX_CPU_BASED_CR3_STORE_EXITING |
+            VMX_CPU_BASED_CR8_LOAD_EXITING | VMX_CPU_BASED_CR8_STORE_EXITING |
+            VMX_CPU_BASED_TPR_SHADOW | VMX_CPU_BASED_VIRTUAL_NMI_PENDING |
+            VMX_CPU_BASED_MOV_DR_EXITING | VMX_CPU_BASED_UNCOND_IO_EXITING |
+            VMX_CPU_BASED_USE_IO_BITMAPS | VMX_CPU_BASED_MONITOR_TRAP_FLAG |
+            VMX_CPU_BASED_USE_MSR_BITMAPS | VMX_CPU_BASED_MONITOR_EXITING |
+            VMX_CPU_BASED_PAUSE_EXITING |
+            VMX_CPU_BASED_ACTIVATE_SECONDARY_CONTROLS,
+        .features[FEAT_VMX_SECONDARY_CTLS] =
+            VMX_SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
+            VMX_SECONDARY_EXEC_ENABLE_EPT | VMX_SECONDARY_EXEC_DESC |
+            VMX_SECONDARY_EXEC_RDTSCP |
+            VMX_SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
+            VMX_SECONDARY_EXEC_ENABLE_VPID | VMX_SECONDARY_EXEC_WBINVD_EXITING |
+            VMX_SECONDARY_EXEC_UNRESTRICTED_GUEST |
+            VMX_SECONDARY_EXEC_APIC_REGISTER_VIRT |
+            VMX_SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
+            VMX_SECONDARY_EXEC_RDRAND_EXITING |
+            VMX_SECONDARY_EXEC_ENABLE_INVPCID |
+            VMX_SECONDARY_EXEC_ENABLE_VMFUNC | VMX_SECONDARY_EXEC_SHADOW_VMCS |
+            VMX_SECONDARY_EXEC_RDSEED_EXITING | VMX_SECONDARY_EXEC_ENABLE_PML |
+            VMX_SECONDARY_EXEC_XSAVES,
+        .features[FEAT_VMX_VMFUNC] =
+            MSR_VMX_VMFUNC_EPT_SWITCHING,
+        .xlevel = 0x80000008,
+        .model_id = "Intel Xeon Processor (SapphireRapids)",
+        .versions = (X86CPUVersionDefinition[]) {
+            { .version = 1 },
+            { /* end of list */ },
+        },
+    },
      {
          .name = "Denverton",
          .level = 21,
target/i386/cpu.h +4 -0
index 82004b65b9..ef3e8a5ed5 100644
@@ -879,10 +879,14 @@ uint64_t x86_cpu_get_supported_feature_word(FeatureWord w,
  #define CPUID_7_0_EDX_TSX_LDTRK         (1U << 16)
  /* Architectural LBRs */
  #define CPUID_7_0_EDX_ARCH_LBR          (1U << 19)
+/* AMX_BF16 instruction */
+#define CPUID_7_0_EDX_AMX_BF16          (1U << 22)
  /* AVX512_FP16 instruction */
  #define CPUID_7_0_EDX_AVX512_FP16       (1U << 23)
  /* AMX tile (two-dimensional register) */
  #define CPUID_7_0_EDX_AMX_TILE          (1U << 24)
+/* AMX_INT8 instruction */
+#define CPUID_7_0_EDX_AMX_INT8          (1U << 25)
  /* Speculation Control */
  #define CPUID_7_0_EDX_SPEC_CTRL         (1U << 26)
  /* Single Thread Indirect Branch Predictors */
Signature · 1 line
2.34.1