[AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA operations #163161

jthackray · 2025-10-13T09:38:53Z

Add instructions for SVE2p3 DOT and MLA operations:

BFMMLA (non-widening)
FMMLA (non-widening)
SDOT (2-way, vectors)
SDOT (2-way, indexed)
UDOT (2-way, vectors)
UDOT (2-way, indexed)

as documented here:

jthackray · 2025-10-13T09:39:11Z

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-10-13T09:41:11Z

@llvm/pr-subscribers-clang
@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-clang-driver

Author: Jonathan Thackray (jthackray)

Changes

Add instructions for SVE2p3 DOT and MLA operations:

BFMMLA (non-widening)
FMMLA (non-widening)
SDOT (2-way, vectors)
SDOT (2-way, indexed)
UDOT (2-way, vectors)
UDOT (2-way, indexed)

as documented here:

Patch is 41.36 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/163161.diff

18 Files Affected:

(modified) clang/test/Driver/aarch64-v97a.c (+8)
(modified) clang/test/Driver/print-supported-extensions-aarch64.c (+2)
(modified) llvm/lib/Target/AArch64/AArch64.td (+1-1)
(modified) llvm/lib/Target/AArch64/AArch64Features.td (+6)
(modified) llvm/lib/Target/AArch64/AArch64InstrInfo.td (+6)
(modified) llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td (+26-4)
(modified) llvm/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp (+2)
(modified) llvm/lib/Target/AArch64/SVEInstrFormats.td (+25-11)
(modified) llvm/test/MC/AArch64/SVE/bfmmla-diagnostics.s (-5)
(modified) llvm/test/MC/AArch64/SVE2p1/sdot-diagnostics.s (+1-1)
(modified) llvm/test/MC/AArch64/SVE2p1/udot-diagnostics.s (+2-2)
(added) llvm/test/MC/AArch64/SVE2p2/fmmla-diagnostics.s (+24)
(added) llvm/test/MC/AArch64/SVE2p2/fmmla.s (+45)
(added) llvm/test/MC/AArch64/SVE2p3/bfmmla-diagnostics.s (+34)
(added) llvm/test/MC/AArch64/SVE2p3/bfmmla.s (+45)
(added) llvm/test/MC/AArch64/SVE2p3/dot-diagnostics.s (+188)
(added) llvm/test/MC/AArch64/SVE2p3/dot.s (+173)
(modified) llvm/unittests/TargetParser/TargetParserTest.cpp (+12-3)

diff --git a/clang/test/Driver/aarch64-v97a.c b/clang/test/Driver/aarch64-v97a.c
index cfa9cad692914..171b2b44e3daa 100644
--- a/clang/test/Driver/aarch64-v97a.c
+++ b/clang/test/Driver/aarch64-v97a.c
@@ -26,6 +26,14 @@
 // RUN: %clang -target aarch64 -march=armv9.7-a+sve2p3 -### -c %s 2>&1 | FileCheck -check-prefix=V97A-SVE2p3 %s
 // V97A-SVE2p3: "-cc1"{{.*}} "-triple" "aarch64{{.*}}" "-target-cpu" "generic" "-target-feature" "+v9.7a"{{.*}} "-target-feature" "+sve2p3"
 
+// RUN: %clang -target aarch64 -march=armv9.7a+sve-b16mm -### -c %s 2>&1 | FileCheck -check-prefix=V97A-SVE-B16MM %s
+// RUN: %clang -target aarch64 -march=armv9.7-a+sve-b16mm -### -c %s 2>&1 | FileCheck -check-prefix=V97A-SVE-B16MM %s
+// V97A-SVE-B16MM: "-cc1"{{.*}} "-triple" "aarch64{{.*}}" "-target-cpu" "generic" "-target-feature" "+v9.7a"{{.*}} "-target-feature" "+sve-b16mm"
+
+// RUN: %clang -target aarch64 -march=armv9.7a+f16mm -### -c %s 2>&1 | FileCheck -check-prefix=V97A-F16MM %s
+// RUN: %clang -target aarch64 -march=armv9.7-a+f16mm -### -c %s 2>&1 | FileCheck -check-prefix=V97A-F16MM %s
+// V97A-F16MM: "-cc1"{{.*}} "-triple" "aarch64{{.*}}" "-target-cpu" "generic" "-target-feature" "+v9.7a"{{.*}} "-target-feature" "+f16mm"
+
 // RUN: %clang -target aarch64 -march=armv9.7a+cmh -### -c %s 2>&1 | FileCheck -check-prefix=V97A-CMH %s
 // RUN: %clang -target aarch64 -march=armv9.7-a+cmh -### -c %s 2>&1 | FileCheck -check-prefix=V97A-CMH %s
 // V97A-CMH: "-cc1"{{.*}} "-triple" "aarch64{{.*}}" "-target-cpu" "generic" "-target-feature" "+v9.7a"{{.*}} "-target-feature" "+cmh"
diff --git a/clang/test/Driver/print-supported-extensions-aarch64.c b/clang/test/Driver/print-supported-extensions-aarch64.c
index ef53f9fba4c43..50c3610123646 100644
--- a/clang/test/Driver/print-supported-extensions-aarch64.c
+++ b/clang/test/Driver/print-supported-extensions-aarch64.c
@@ -18,6 +18,7 @@
 // CHECK-NEXT:     d128                FEAT_D128, FEAT_LVA3, FEAT_SYSREG128, FEAT_SYSINSTR128 Enable Armv9.4-A 128-bit Page Table Descriptors, System Registers and instructions
 // CHECK-NEXT:     dit                 FEAT_DIT                                               Enable Armv8.4-A Data Independent Timing instructions
 // CHECK-NEXT:     dotprod             FEAT_DotProd                                           Enable dot product support
+// CHECK-NEXT:     f16mm               FEAT_F16MM                                             Enable Armv9.7-A non-widening half-precision matrix multiply-accumulate
 // CHECK-NEXT:     f32mm               FEAT_F32MM                                             Enable Matrix Multiply FP32 Extension
 // CHECK-NEXT:     f64mm               FEAT_F64MM                                             Enable Matrix Multiply FP64 Extension
 // CHECK-NEXT:     f8f16mm             FEAT_F8F16MM                                           Enable Armv9.6-A FP8 to Half-Precision Matrix Multiplication
@@ -95,6 +96,7 @@
 // CHECK-NEXT:     sve-aes             FEAT_SVE_AES, FEAT_SVE_PMULL128                        Enable SVE AES and quadword SVE polynomial multiply instructions
 // CHECK-NEXT:     sve-aes2            FEAT_SVE_AES2                                          Enable Armv9.6-A SVE multi-vector AES and multi-vector quadword polynomial multiply instructions
 // CHECK-NEXT:     sve-b16b16          FEAT_SVE_B16B16                                        Enable SVE2 non-widening and SME2 Z-targeting non-widening BFloat16 instructions
+// CHECK-NEXT:     sve-b16mm           FEAT_SVE_B16MM                                         Enable Armv9.7-A SVE non-widening BFloat16 matrix multiply-accumulate
 // CHECK-NEXT:     sve-bfscale         FEAT_SVE_BFSCALE                                       Enable Armv9.6-A SVE BFloat16 scaling instructions
 // CHECK-NEXT:     sve-bitperm         FEAT_SVE_BitPerm                                       Enable bit permutation SVE2 instructions
 // CHECK-NEXT:     sve-f16f32mm        FEAT_SVE_F16F32MM                                      Enable Armv9.6-A FP16 to FP32 Matrix Multiply
diff --git a/llvm/lib/Target/AArch64/AArch64.td b/llvm/lib/Target/AArch64/AArch64.td
index 463465310c95b..a4529a5844ab7 100644
--- a/llvm/lib/Target/AArch64/AArch64.td
+++ b/llvm/lib/Target/AArch64/AArch64.td
@@ -74,7 +74,7 @@ def SVEUnsupported : AArch64Unsupported {
 }
 
 def SME2p3Unsupported : AArch64Unsupported {
-  let F = [HasSVE2p3_or_SME2p3];
+  let F = [HasSVE2p3_or_SME2p3, HasSVE_B16MM];
 }
 
 def SME2p2Unsupported : AArch64Unsupported {
diff --git a/llvm/lib/Target/AArch64/AArch64Features.td b/llvm/lib/Target/AArch64/AArch64Features.td
index 22e1e6ea3de4a..5f943d39321f9 100644
--- a/llvm/lib/Target/AArch64/AArch64Features.td
+++ b/llvm/lib/Target/AArch64/AArch64Features.td
@@ -613,6 +613,12 @@ def FeatureSVE2p3 : ExtensionWithMArch<"sve2p3", "SVE2p3", "FEAT_SVE2p3",
 def FeatureSME2p3 : ExtensionWithMArch<"sme2p3", "SME2p3", "FEAT_SME2p3",
   "Enable Armv9.7-A Scalable Matrix Extension 2.3 instructions", [FeatureSME2p2]>;
 
+def FeatureSVE_B16MM : ExtensionWithMArch<"sve-b16mm", "SVE_B16MM", "FEAT_SVE_B16MM",
+  "Enable Armv9.7-A SVE non-widening BFloat16 matrix multiply-accumulate", [FeatureSVE]>;
+
+def FeatureF16MM : ExtensionWithMArch<"f16mm", "F16MM", "FEAT_F16MM",
+  "Enable Armv9.7-A non-widening half-precision matrix multiply-accumulate", [FeatureFullFP16]>;
+
 //===----------------------------------------------------------------------===//
 //  Other Features
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 6bdfbecaa2759..881bb882de351 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -246,6 +246,12 @@ def HasCMH          : Predicate<"Subtarget->hasCMH()">,
                                  AssemblerPredicateWithAll<(all_of FeatureCMH), "cmh">;
 def HasLSCP         : Predicate<"Subtarget->hasLSCP()">,
                                  AssemblerPredicateWithAll<(all_of FeatureLSCP), "lscp">;
+def HasSVE2p2       : Predicate<"Subtarget->hasSVE2p2()">,
+                                 AssemblerPredicateWithAll<(all_of FeatureSVE2p2), "sve2p2">;
+def HasSVE_B16MM    : Predicate<"Subtarget->isSVEAvailable() &&ubtarget->hasSVE_B16MM()">,
+                                 AssemblerPredicateWithAll<(all_of FeatureSVE_B16MM), "sve-b16mm">;
+def HasF16MM        : Predicate<"Subtarget->isSVEAvailable() && Subtarget->hasF16MM()">,
+                                 AssemblerPredicateWithAll<(all_of FeatureF16MM), "f16mm">;
 
 // A subset of SVE(2) instructions are legal in Streaming SVE execution mode,
 // they should be enabled if either has been specified.
diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 85aebe18ee169..3783daed5d61b 100644
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -2569,7 +2569,7 @@ let Predicates = [HasBF16, HasSVE_or_SME] in {
 } // End HasBF16, HasSVE_or_SME
 
 let Predicates = [HasBF16, HasSVE] in {
-  defm BFMMLA_ZZZ_HtoS : sve_fp_matrix_mla<0b01, "bfmmla", ZPR32, ZPR16, int_aarch64_sve_bfmmla, nxv4f32, nxv8bf16>;
+  defm BFMMLA_ZZZ_HtoS : sve_fp_matrix_mla<0b011, "bfmmla", ZPR32, ZPR16, int_aarch64_sve_bfmmla, nxv4f32, nxv8bf16>;
 } // End HasBF16, HasSVE
 
 let Predicates = [HasBF16, HasSVE_or_SME] in {
@@ -3680,15 +3680,15 @@ let Predicates = [HasSVE_or_SME, HasMatMulInt8] in {
 } // End HasSVE_or_SME, HasMatMulInt8
 
 let Predicates = [HasSVE, HasMatMulFP32] in {
-  defm FMMLA_ZZZ_S : sve_fp_matrix_mla<0b10, "fmmla", ZPR32, ZPR32, int_aarch64_sve_fmmla, nxv4f32, nxv4f32>;
+  defm FMMLA_ZZZ_S : sve_fp_matrix_mla<0b101, "fmmla", ZPR32, ZPR32, int_aarch64_sve_fmmla, nxv4f32, nxv4f32>;
 } // End HasSVE, HasMatMulFP32
 
 let Predicates = [HasSVE_F16F32MM] in {
-  def FMLLA_ZZZ_HtoS : sve_fp_matrix_mla<0b00, "fmmla", ZPR32, ZPR16>;
+  def FMLLA_ZZZ_HtoS : sve_fp_matrix_mla<0b001, "fmmla", ZPR32, ZPR16>;
 } // End HasSVE_F16F32MM
 
 let Predicates = [HasSVE, HasMatMulFP64] in {
-  defm FMMLA_ZZZ_D : sve_fp_matrix_mla<0b11, "fmmla", ZPR64, ZPR64, int_aarch64_sve_fmmla, nxv2f64, nxv2f64>;
+  defm FMMLA_ZZZ_D : sve_fp_matrix_mla<0b111, "fmmla", ZPR64, ZPR64, int_aarch64_sve_fmmla, nxv2f64, nxv2f64>;
   defm LD1RO_B_IMM : sve_mem_ldor_si<0b00, "ld1rob", Z_b, ZPR8,  nxv16i8, nxv16i1, AArch64ld1ro_z>;
   defm LD1RO_H_IMM : sve_mem_ldor_si<0b01, "ld1roh", Z_h, ZPR16, nxv8i16, nxv8i1,  AArch64ld1ro_z>;
   defm LD1RO_W_IMM : sve_mem_ldor_si<0b10, "ld1row", Z_s, ZPR32, nxv4i32, nxv4i1,  AArch64ld1ro_z>;
@@ -4631,8 +4631,30 @@ let Predicates = [HasSVE2p3_or_SME2p3] in {
   defm SABAL_ZZZ : sve2_int_two_way_absdiff_accum_long<0b0, "sabal">;
   defm UABAL_ZZZ : sve2_int_two_way_absdiff_accum_long<0b1, "uabal">;
 
+  // SVE2 integer dot product
+  def SDOT_ZZZ_BtoH : sve_intx_dot<0b01, 0b00000, 0b0, "sdot", ZPR16, ZPR8>;
+  def UDOT_ZZZ_BtoH : sve_intx_dot<0b01, 0b00000, 0b1, "udot", ZPR16, ZPR8>;
+
+  // SVE2 integer indexed dot product
+  def SDOT_ZZZI_BtoH : sve_intx_dot_by_indexed_elem_x<0b0, "sdot">;
+  def UDOT_ZZZI_BtoH : sve_intx_dot_by_indexed_elem_x<0b1, "udot">;
+
 } // End HasSME2p2orSVE2p2
 
+//===----------------------------------------------------------------------===//
+// SVE_B16MM Instructions
+//===----------------------------------------------------------------------===//
+let Predicates = [HasSVE_B16MM] in {
+  def BFMMLA_ZZZ_H : sve_fp_matrix_mla<0b110, "bfmmla", ZPR16, ZPR16>;
+}
+
+//===----------------------------------------------------------------------===//
+// F16MM Instructions
+//===----------------------------------------------------------------------===//
+let Predicates = [HasSVE2p2, HasF16MM] in {
+  def FMMLA_ZZZ_H : sve_fp_matrix_mla<0b100, "fmmla", ZPR16, ZPR16>;
+}
+
 //===----------------------------------------------------------------------===//
 // SME2.2 or SVE2.2 instructions - Legal in streaming mode iff target has SME2p2
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp b/llvm/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp
index 4bfc74f7aa984..d9f3c4ffd5226 100644
--- a/llvm/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp
+++ b/llvm/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp
@@ -3892,6 +3892,8 @@ static const struct Extension {
     {"gcie", {AArch64::FeatureGCIE}},
     {"sme2p3", {AArch64::FeatureSME2p3}},
     {"sve2p3", {AArch64::FeatureSVE2p3}},
+    {"sve-b16mm", {AArch64::FeatureSVE_B16MM}},
+    {"f16mm", {AArch64::FeatureF16MM}},
 };
 
 static void setRequiredFeatureString(FeatureBitset FBS, std::string &Str) {
diff --git a/llvm/lib/Target/AArch64/SVEInstrFormats.td b/llvm/lib/Target/AArch64/SVEInstrFormats.td
index 337c6b694a505..21b9310f69159 100644
--- a/llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -3787,7 +3787,7 @@ multiclass sve2p1_two_way_dot_vv<string mnemonic, bit u, SDPatternOperator intri
 // SVE Integer Dot Product Group - Indexed Group
 //===----------------------------------------------------------------------===//
 
-class sve_intx_dot_by_indexed_elem<bit sz, bit U, string asm,
+class sve_intx_dot_by_indexed_elem<bit U, string asm,
                                    ZPRRegOp zprty1, ZPRRegOp zprty2,
                                    ZPRRegOp zprty3, Operand itype>
 : I<(outs zprty1:$Zda), (ins zprty1:$_Zda, zprty2:$Zn, zprty3:$Zm, itype:$iop),
@@ -3795,8 +3795,7 @@ class sve_intx_dot_by_indexed_elem<bit sz, bit U, string asm,
   "", []>, Sched<[]> {
   bits<5> Zda;
   bits<5> Zn;
-  let Inst{31-23} = 0b010001001;
-  let Inst{22}    = sz;
+  let Inst{31-24} = 0b01000100;
   let Inst{21}    = 0b1;
   let Inst{15-11} = 0;
   let Inst{10}    = U;
@@ -3810,16 +3809,18 @@ class sve_intx_dot_by_indexed_elem<bit sz, bit U, string asm,
 
 multiclass sve_intx_dot_by_indexed_elem<bit opc, string asm,
                                         SDPatternOperator op> {
-  def _BtoS : sve_intx_dot_by_indexed_elem<0b0, opc, asm, ZPR32, ZPR8, ZPR3b8, VectorIndexS32b_timm> {
+  def _BtoS : sve_intx_dot_by_indexed_elem<opc, asm, ZPR32, ZPR8, ZPR3b8, VectorIndexS32b_timm> {
     bits<2> iop;
     bits<3> Zm;
+    let Inst{23-22} = 0b10;
     let Inst{20-19} = iop;
     let Inst{18-16} = Zm;
   }
-  def _HtoD : sve_intx_dot_by_indexed_elem<0b1, opc, asm, ZPR64, ZPR16, ZPR4b16, VectorIndexD32b_timm> {
+  def _HtoD : sve_intx_dot_by_indexed_elem<opc, asm, ZPR64, ZPR16, ZPR4b16, VectorIndexD32b_timm> {
     bits<1> iop;
     bits<4> Zm;
-    let Inst{20} = iop;
+    let Inst{23-22} = 0b11;
+    let Inst{20}    = iop;
     let Inst{19-16} = Zm;
   }
 
@@ -3827,6 +3828,16 @@ multiclass sve_intx_dot_by_indexed_elem<bit opc, string asm,
   def : SVE_4_Op_Imm_Pat<nxv2i64, op, nxv2i64, nxv8i16, nxv8i16, i32, VectorIndexD32b_timm, !cast<Instruction>(NAME # _HtoD)>;
 }
 
+class sve_intx_dot_by_indexed_elem_x<bit opc, string asm>
+: sve_intx_dot_by_indexed_elem<opc, asm, ZPR16, ZPR8, ZPR3b8, VectorIndexH32b_timm> {
+ bits<3> iop;
+ bits<3> Zm;
+ let Inst{23}    = 0b0;
+ let Inst{22}    = iop{2-2};
+ let Inst{20-19} = iop{1-0};
+ let Inst{18-16} = Zm;
+}
+
 //===----------------------------------------------------------------------===//
 // SVE2 Complex Integer Dot Product Group
 //===----------------------------------------------------------------------===//
@@ -9616,17 +9627,18 @@ multiclass sve_int_dot_mixed_indexed<bit U, string asm, SDPatternOperator op> {
 // SVE Floating Point Matrix Multiply Accumulate Group
 //===----------------------------------------------------------------------===//
 
-class sve_fp_matrix_mla<bits<2> opc, string asm, ZPRRegOp zda_ty, ZPRRegOp reg_ty>
+class sve_fp_matrix_mla<bits<3> opc, string asm, ZPRRegOp zda_ty, ZPRRegOp reg_ty>
 : I<(outs zda_ty:$Zda), (ins zda_ty:$_Zda, reg_ty:$Zn, reg_ty:$Zm),
     asm, "\t$Zda, $Zn, $Zm", "", []>, Sched<[]> {
   bits<5> Zda;
   bits<5> Zn;
   bits<5> Zm;
   let Inst{31-24} = 0b01100100;
-  let Inst{23-22} = opc;
+  let Inst{23-22} = opc{2-1};
   let Inst{21}    = 1;
   let Inst{20-16} = Zm;
-  let Inst{15-10} = 0b111001;
+  let Inst{15-11} = 0b11100;
+  let Inst{10}    = opc{0};
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zda;
 
@@ -9636,10 +9648,12 @@ class sve_fp_matrix_mla<bits<2> opc, string asm, ZPRRegOp zda_ty, ZPRRegOp reg_t
   let mayRaiseFPException = 1;
 }
 
-multiclass sve_fp_matrix_mla<bits<2> opc, string asm, ZPRRegOp zda_ty, ZPRRegOp reg_ty, SDPatternOperator op, ValueType zda_vt, ValueType reg_vt> {
+multiclass sve_fp_matrix_mla<bits<3> opc, string asm, ZPRRegOp zda_ty,
+                             ZPRRegOp reg_ty, SDPatternOperator op,
+                             ValueType zda_vt, ValueType reg_vt> {
   def NAME : sve_fp_matrix_mla<opc, asm, zda_ty, reg_ty>;
 
-  def : SVE_3_Op_Pat<zda_vt, op , zda_vt, reg_vt, reg_vt, !cast<Instruction>(NAME)>;
+  def : SVE_3_Op_Pat<zda_vt, op, zda_vt, reg_vt, reg_vt, !cast<Instruction>(NAME)>;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/llvm/test/MC/AArch64/SVE/bfmmla-diagnostics.s b/llvm/test/MC/AArch64/SVE/bfmmla-diagnostics.s
index 409c2c5741cab..4695b050a681e 100644
--- a/llvm/test/MC/AArch64/SVE/bfmmla-diagnostics.s
+++ b/llvm/test/MC/AArch64/SVE/bfmmla-diagnostics.s
@@ -5,11 +5,6 @@ bfmmla z0.s, z1.s, z2.h
 // CHECK-NEXT: bfmmla z0.s, z1.s, z2.h
 // CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
 
-bfmmla z0.h, z1.h, z2.h
-// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
-// CHECK-NEXT: bfmmla z0.h, z1.h, z2.h
-// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
-
 bfmmla z0.s, z1.h, z2.s
 // CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
 // CHECK-NEXT: bfmmla z0.s, z1.h, z2.s
diff --git a/llvm/test/MC/AArch64/SVE2p1/sdot-diagnostics.s b/llvm/test/MC/AArch64/SVE2p1/sdot-diagnostics.s
index 344abb1fe9082..42c6ae7da156b 100644
--- a/llvm/test/MC/AArch64/SVE2p1/sdot-diagnostics.s
+++ b/llvm/test/MC/AArch64/SVE2p1/sdot-diagnostics.s
@@ -29,7 +29,7 @@ sdot z0.s, z0.h, z0.h[-1]
 // Invalid vector suffix
 
 sdot z0.h, z0.s, z0.s
-// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
 // CHECK-NEXT: sdot z0.h, z0.s, z0.s
 // CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
 
diff --git a/llvm/test/MC/AArch64/SVE2p1/udot-diagnostics.s b/llvm/test/MC/AArch64/SVE2p1/udot-diagnostics.s
index 0debddf82b37c..aa2ec7f4d3204 100644
--- a/llvm/test/MC/AArch64/SVE2p1/udot-diagnostics.s
+++ b/llvm/test/MC/AArch64/SVE2p1/udot-diagnostics.s
@@ -29,11 +29,11 @@ udot z0.s, z0.h, z0.h[-1]
 // Invalid vector suffix
 
 udot z0.h, z0.s, z0.s
-// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
 // CHECK-NEXT: udot z0.h, z0.s, z0.s
 // CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
 
 udot z0.h, z0.s, z0.s[1]
-// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
 // CHECK-NEXT: udot z0.h, z0.s, z0.s[1]
 // CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
diff --git a/llvm/test/MC/AArch64/SVE2p2/fmmla-diagnostics.s b/llvm/test/MC/AArch64/SVE2p2/fmmla-diagnostics.s
new file mode 100644
index 0000000000000..1a48582cfc742
--- /dev/null
+++ b/llvm/test/MC/AArch64/SVE2p2/fmmla-diagnostics.s
@@ -0,0 +1,24 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2p2,+f16mm 2>&1 < %s| FileCheck %s
+
+// --------------------------------------------------------------------------//
+// Invalid element width
+
+fmmla z0.b, z0.b, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: fmmla z0.b, z0.b, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+// --------------------------------------------------------------------------//
+// Negative tests for instructions that are incompatible with movprfx
+
+movprfx z0.h, p0/m, z7.h
+fmmla z0.h, z0.h, z0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a movprfx and destination also used as non-destructive source
+// CHECK-NEXT: fmmla z0.h, z0.h, z0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+movprfx z0, z7
+fmmla z0.h, z0.h, z0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a movprfx and destination also used as non-destructive source
+// CHECK-NEXT: fmmla z0.h, z0.h, z0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
diff --git a/llvm/test/MC/AArch64/SVE2p2/fmmla.s b/llvm/test/MC/AArch64/SVE2p2/fmmla.s
new file mode 100644
index 0000000000000..19929a921dba6
--- /dev/null
+++ b/llvm/test/MC/AArch64/SVE2p2/fmmla.s
@@ -0,0 +1,45 @@
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2p2,+f16mm < %s \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+// RUN: not llvm-mc -triple=aarch64 -show-encoding < %s 2>&1 \
+// RUN:        | FileCheck %s --check-prefix=CHECK-ERROR
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve2p2,+f16mm < %s \
+// RUN:        | llvm-objdump -d --mattr=+sve2p2,+f16mm --no-print-imm-hex - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve2p2,+f16mm < %s \
+// RUN:        | llvm-objdump -d --mattr=-sve2p2,-f16mm --no-print-imm-hex - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+// Disassemble encoding and check the re-encoding (-show-encoding) matches.
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2p2,+f16mm < %s \
+// RUN:        | sed '/.text/d' | sed 's/.*encoding: //g' \
+// RUN:        | llvm-mc -triple=aarch64 -mattr=+sve2p2,+f16mm -disassemble -show-encoding \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+
+fmmla z0.h, z0.h, z0.h
+// CHECK-INST: fmmla z0.h, z0.h, z0.h
+// CHECK-ENCODING: encoding: [0x00,0xe0,0xa0,0x64]
+// CHECK-ERROR: instruction requires: f16mm sve2p2
+// CHECK-UNKNOWN: 64a0e000 <unknown>
+
+fmmla z10.h, z10.h, z10.h
+// CHECK-INST: fmmla z10.h, z10.h, z10.h
+// CHECK-ENCODING: encoding: [0x4a,0xe1,0xaa,0x64]
+// CHECK-ERROR: instruction requires: f16mm sve2p2
+// CHECK-UNKNOWN: 64aae14a <unknown>
+
+fmmla z21.h, z21.h, z21.h
+// CHECK-INST: fmmla z21.h, z21.h, z21.h
+// CHECK-ENCODING: encoding: [0xb5,0xe2,0xb5,0x64]
+// CHECK-ERROR: instruction requires: f16mm sve2p2
+// CHECK-UNKNOWN: 64b5e2...
[truncated]

jthackray · 2025-10-23T12:32:13Z

Merge activity

Oct 23, 12:32 PM UTC: A user started a stack merge that includes this pull request via Graphite.
Oct 23, 12:46 PM UTC: The Graphite merge of this pull request was cancelled.
Oct 23, 8:46 PM UTC: A user started a stack merge that includes this pull request via Graphite.
Oct 23, 10:03 PM UTC: The Graphite merge of this pull request was cancelled.

Add instructions for SVE2p3 DOT and MLA operations: - BFMMLA (non-widening) - FMMLA (non-widening) - SDOT (2-way, vectors) - SDOT (2-way, indexed) - UDOT (2-way, vectors) - UDOT (2-way, indexed) as documented here: * https://developer.arm.com/documentation/ddi0602/2025-09/ * https://developer.arm.com/documentation/109697/2025_09/2025-Architecture-Extensions

…operations Fix accidental typo

…operations Remove tests which aren't useful after CR comments

…operations Add diagnostic for test following predicated movprfx

…ons (llvm#163161) Add instructions for SVE2p3 DOT and MLA operations: - BFMMLA (non-widening) - FMMLA (non-widening) - SDOT (2-way, vectors) - SDOT (2-way, indexed) - UDOT (2-way, vectors) - UDOT (2-way, indexed) as documented here: * https://developer.arm.com/documentation/ddi0602/2025-09/ * https://developer.arm.com/documentation/109697/2025_09/2025-Architecture-Extensions

jthackray requested review from CarolineConcatto, Lukacma, amilendra, kmclaughlin-arm and rgwott October 13, 2025 09:38

jthackray marked this pull request as ready for review October 13, 2025 09:40

llvmbot added clang Clang issues not falling into any other category backend:AArch64 clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' labels Oct 13, 2025

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from 6d421c4 to f484bf2 Compare October 13, 2025 16:49

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch 2 times, most recently from 6c85eb7 to 44312cd Compare October 13, 2025 17:59

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from a743313 to 5e8d2c0 Compare October 13, 2025 17:59

jthackray mentioned this pull request Oct 15, 2025

[AArch64] (NFC) Tidy up alignment/formatting in AArch64/AArch64InstrInfo.td #163645

Merged

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from 44312cd to 815a3a4 Compare October 16, 2025 11:01

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from 5e8d2c0 to 5992744 Compare October 16, 2025 11:01

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from 611d999 to d39a1ff Compare October 23, 2025 10:36

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from 1c0febe to c192485 Compare October 23, 2025 10:36

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from d39a1ff to a36d3d6 Compare October 23, 2025 10:42

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from c192485 to 9357db7 Compare October 23, 2025 10:42

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from 9357db7 to f5bd6dc Compare October 23, 2025 12:47

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch 2 times, most recently from 6e8d1df to e08414c Compare October 23, 2025 22:14

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch 2 times, most recently from 3584efb to 76bacd7 Compare October 23, 2025 22:27

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from e08414c to 035e54b Compare October 23, 2025 22:27

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from 76bacd7 to 425049e Compare October 23, 2025 22:32

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from 035e54b to 3304047 Compare October 23, 2025 22:32

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from 425049e to 2e03194 Compare October 23, 2025 22:36

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from 3304047 to 341e61d Compare October 23, 2025 22:36

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from 2e03194 to a8d3f49 Compare October 23, 2025 22:40

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from 341e61d to 971ece9 Compare October 23, 2025 22:40

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from a8d3f49 to c03c938 Compare October 23, 2025 22:44

jthackray force-pushed the users/jthackray/armv9.7a-sve-arithmetic branch from 971ece9 to 16ee161 Compare October 23, 2025 22:44

Base automatically changed from users/jthackray/armv9.7a-sve-arithmetic to main October 23, 2025 22:47

jthackray added 4 commits October 23, 2025 23:48

fixup! [AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA …

1bf2e06

…operations Fix accidental typo

fixup! [AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA …

b90ed8f

…operations Remove tests which aren't useful after CR comments

fixup! [AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA …

c9b795c

…operations Add diagnostic for test following predicated movprfx

jthackray force-pushed the users/jthackray/armv9.7a-sve-dot-mla branch from c03c938 to c9b795c Compare October 23, 2025 22:48

jthackray merged commit bfae15a into main Oct 23, 2025
4 of 5 checks passed

jthackray deleted the users/jthackray/armv9.7a-sve-dot-mla branch October 23, 2025 22:51

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA operations #163161

[AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA operations #163161

Uh oh!

jthackray commented Oct 13, 2025

Uh oh!

jthackray commented Oct 13, 2025 •

edited

Loading

Uh oh!

llvmbot commented Oct 13, 2025 •

edited

Loading

Uh oh!

jthackray commented Oct 23, 2025 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

[AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA operations #163161

[AArch64][llvm] Armv9.7-A: Add support for SVE2p3 DOT and MLA operations #163161

Uh oh!

Conversation

jthackray commented Oct 13, 2025

Uh oh!

jthackray commented Oct 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Oct 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jthackray commented Oct 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merge activity

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

jthackray commented Oct 13, 2025 •

edited

Loading

llvmbot commented Oct 13, 2025 •

edited

Loading

jthackray commented Oct 23, 2025 •

edited

Loading