[AArch64] Set the default streaming hazard size to 1024 for +sme,+sve

MacDue · MacDue · commit 70345f5d2cf2 · 2025-01-21T14:17:42.000Z
The default for all other feature combinations remains at zero (i.e.
no streaming hazards). This value may be adjusted in the future (e.g.
based on the processor family), for now, it is set conservatively.
diff --git a/llvm/lib/Target/AArch64/AArch64Subtarget.cpp b/llvm/lib/Target/AArch64/AArch64Subtarget.cpp
@@ -358,7 +358,10 @@ AArch64Subtarget::AArch64Subtarget(const Triple &TT, StringRef CPU,
       CustomCallSavedXRegs(AArch64::GPR64commonRegClass.getNumRegs()),
       IsLittle(LittleEndian), IsStreaming(IsStreaming),
       IsStreamingCompatible(IsStreamingCompatible),
-      StreamingHazardSize(AArch64StreamingHazardSize),
+      StreamingHazardSize(
+          AArch64StreamingHazardSize.getNumOccurrences() > 0
+              ? std::optional<unsigned>(AArch64StreamingHazardSize)
+              : std::nullopt),
       MinSVEVectorSizeInBits(MinSVEVectorSizeInBitsOverride),
       MaxSVEVectorSizeInBits(MaxSVEVectorSizeInBitsOverride), TargetTriple(TT),
       InstrInfo(initializeSubtargetDependencies(FS, CPU, TuneCPU, HasMinSize)),
diff --git a/llvm/lib/Target/AArch64/AArch64Subtarget.h b/llvm/lib/Target/AArch64/AArch64Subtarget.h
@@ -85,7 +85,7 @@ class AArch64Subtarget final : public AArch64GenSubtargetInfo {
 
   bool IsStreaming;
   bool IsStreamingCompatible;
-  unsigned StreamingHazardSize;
+  std::optional<unsigned> StreamingHazardSize;
   unsigned MinSVEVectorSizeInBits;
   unsigned MaxSVEVectorSizeInBits;
   unsigned VScaleForTuning = 1;
@@ -179,7 +179,12 @@ class AArch64Subtarget final : public AArch64GenSubtargetInfo {
 
   /// Returns the size of memory region that if accessed by both the CPU and
   /// the SME unit could result in a hazard. 0 = disabled.
-  unsigned getStreamingHazardSize() const { return StreamingHazardSize; }
+  unsigned getStreamingHazardSize() const {
+    // If StreamingHazardSize has been explicitly set to a value, use that.
+    // Otherwise, default to 1024 bytes when both SME and SVE are available,
+    // for all other configurations default to no streaming hazards.
+    return StreamingHazardSize.value_or(hasSME() && hasSVE() ? 1024 : 0);
+  }
 
   /// Returns true if the target has NEON and the function at runtime is known
   /// to have NEON enabled (e.g. the function is known not to be in streaming-SVE
diff --git a/llvm/test/CodeGen/AArch64/outlining-with-streaming-mode-changes.ll b/llvm/test/CodeGen/AArch64/outlining-with-streaming-mode-changes.ll
@@ -1,5 +1,5 @@
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -enable-machine-outliner -verify-machineinstrs < %s | FileCheck %s
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -enable-machine-outliner -verify-machineinstrs < %s | FileCheck %s -check-prefix=OUTLINER
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -enable-machine-outliner -aarch64-streaming-hazard-size=0 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -enable-machine-outliner -aarch64-streaming-hazard-size=0 -verify-machineinstrs < %s | FileCheck %s -check-prefix=OUTLINER
 
 declare void @callee();
 
diff --git a/llvm/test/CodeGen/AArch64/sme-callee-save-restore-pairs.ll b/llvm/test/CodeGen/AArch64/sme-callee-save-restore-pairs.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2 -mattr=+sve -aarch64-disable-multivector-spill-fill -verify-machineinstrs < %s | FileCheck %s --check-prefixes=NOPAIR
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme -mattr=+sve -verify-machineinstrs < %s | FileCheck %s --check-prefixes=NOPAIR
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2  -mattr=+sve -verify-machineinstrs < %s | FileCheck %s --check-prefixes=PAIR
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sme2 -mattr=+sve -aarch64-disable-multivector-spill-fill -verify-machineinstrs < %s | FileCheck %s --check-prefixes=NOPAIR
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sme -mattr=+sve -verify-machineinstrs < %s | FileCheck %s --check-prefixes=NOPAIR
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sme2  -mattr=+sve -verify-machineinstrs < %s | FileCheck %s --check-prefixes=PAIR
 
 declare void @my_func()
 declare void @my_func2(<vscale x 16 x i8> %v)
diff --git a/llvm/test/CodeGen/AArch64/sme-darwin-sve-vg.ll b/llvm/test/CodeGen/AArch64/sme-darwin-sve-vg.ll
@@ -1,4 +1,4 @@
-; RUN: llc -mtriple=aarch64-darwin -mattr=+sve -mattr=+sme -enable-aarch64-sme-peephole-opt=false -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-darwin -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme -enable-aarch64-sme-peephole-opt=false -verify-machineinstrs < %s | FileCheck %s
 
 declare void @normal_callee();
 
diff --git a/llvm/test/CodeGen/AArch64/sme-disable-gisel-fisel.ll b/llvm/test/CodeGen/AArch64/sme-disable-gisel-fisel.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -fast-isel=true -global-isel=false -fast-isel-abort=0 -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 < %s \
+; RUN: llc -fast-isel=true -aarch64-streaming-hazard-size=0 -global-isel=false -fast-isel-abort=0 -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 < %s \
 ; RUN:     | FileCheck %s --check-prefixes=CHECK-COMMON,CHECK-FISEL
-; RUN: llc -fast-isel=false -global-isel=true -global-isel-abort=0 -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 < %s \
+; RUN: llc -fast-isel=false -aarch64-streaming-hazard-size=0 -global-isel=true -global-isel-abort=0 -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 < %s \
 ; RUN:     | FileCheck %s --check-prefixes=CHECK-COMMON,CHECK-GISEL
 
 
diff --git a/llvm/test/CodeGen/AArch64/sme-lazy-save-call.ll b/llvm/test/CodeGen/AArch64/sme-lazy-save-call.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64 -mattr=+sve -mattr=+sme < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64 -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme < %s | FileCheck %s
 
 declare void @private_za_callee()
 declare float @llvm.cos.f32(float)
diff --git a/llvm/test/CodeGen/AArch64/sme-peephole-opts.ll b/llvm/test/CodeGen/AArch64/sme-peephole-opts.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve,+sme2 < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve,+sme2 < %s | FileCheck %s
 
 declare void @callee()
 declare void @callee_farg(float)
diff --git a/llvm/test/CodeGen/AArch64/sme-pstate-sm-changing-call-disable-coalescing.ll b/llvm/test/CodeGen/AArch64/sme-pstate-sm-changing-call-disable-coalescing.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
-; RUN: llc < %s | FileCheck %s
+; RUN: llc -aarch64-streaming-hazard-size=0 < %s | FileCheck %s
 
 target triple = "aarch64-unknown-unknown-eabi-elf"
 
diff --git a/llvm/test/CodeGen/AArch64/sme-streaming-body-streaming-compatible-interface.ll b/llvm/test/CodeGen/AArch64/sme-streaming-body-streaming-compatible-interface.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme -start-after=simplifycfg -enable-tail-merge=false -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme -start-after=simplifycfg -enable-tail-merge=false -verify-machineinstrs < %s | FileCheck %s
 
 declare void @normal_callee();
 declare void @streaming_callee() "aarch64_pstate_sm_enabled";
diff --git a/llvm/test/CodeGen/AArch64/sme-streaming-body.ll b/llvm/test/CodeGen/AArch64/sme-streaming-body.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme -start-after=simplifycfg -enable-tail-merge=false -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme -start-after=simplifycfg -enable-tail-merge=false -verify-machineinstrs < %s | FileCheck %s
 
 declare void @normal_callee();
 declare void @streaming_callee() "aarch64_pstate_sm_enabled";
diff --git a/llvm/test/CodeGen/AArch64/sme-streaming-compatible-interface.ll b/llvm/test/CodeGen/AArch64/sme-streaming-compatible-interface.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-linux-gnu -verify-machineinstrs -mattr=+sve -mattr=+sme < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -verify-machineinstrs -mattr=+sve -mattr=+sme < %s | FileCheck %s
 
 ; This file tests the following combinations related to streaming-enabled functions:
 ; [ ] N  ->  SC    (Normal -> Streaming-compatible)
diff --git a/llvm/test/CodeGen/AArch64/sme-streaming-interface.ll b/llvm/test/CodeGen/AArch64/sme-streaming-interface.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme -verify-machineinstrs < %s | FileCheck %s
 
 ; This file tests the following combinations related to streaming-enabled functions:
 ; [ ] N  ->  S    (Normal -> Streaming)
diff --git a/llvm/test/CodeGen/AArch64/sme-streaming-mode-changing-call-disable-stackslot-scavenging.ll b/llvm/test/CodeGen/AArch64/sme-streaming-mode-changing-call-disable-stackslot-scavenging.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
-; RUN: llc < %s | FileCheck %s
+; RUN: llc -aarch64-streaming-hazard-size=0 < %s | FileCheck %s
 
 target triple = "aarch64"
 
diff --git a/llvm/test/CodeGen/AArch64/sme-vg-to-stack.ll b/llvm/test/CodeGen/AArch64/sme-vg-to-stack.ll
@@ -1,7 +1,7 @@
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -verify-machineinstrs < %s | FileCheck %s
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -frame-pointer=non-leaf -verify-machineinstrs < %s | FileCheck %s --check-prefix=FP-CHECK
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme2 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme2 -frame-pointer=non-leaf -verify-machineinstrs < %s | FileCheck %s --check-prefix=FP-CHECK
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2 -frame-pointer=non-leaf -verify-machineinstrs < %s | FileCheck %s --check-prefix=NO-SVE-CHECK
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -verify-machineinstrs -enable-machine-outliner < %s | FileCheck %s --check-prefix=OUTLINER-CHECK
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme2 -verify-machineinstrs -enable-machine-outliner < %s | FileCheck %s --check-prefix=OUTLINER-CHECK
 
 declare void @callee();
 declare void @fixed_callee(<4 x i32>);
diff --git a/llvm/test/CodeGen/AArch64/stack-hazard-defaults.ll b/llvm/test/CodeGen/AArch64/stack-hazard-defaults.ll
@@ -0,0 +1,54 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=aarch64 -mattr=+sme -aarch64-stack-hazard-size=0 | FileCheck %s --check-prefix=CHECK0
+; RUN: llc < %s -mtriple=aarch64 -mattr=+sme -aarch64-stack-hazard-size=1024 | FileCheck %s --check-prefix=CHECK1024
+
+;; The following run lines check the default values for aarch64-stack-hazard-size/aarch64-streaming-hazard-size.
+
+;; When +sme,+sve is set the hazard size should default to 1024.
+; RUN: llc < %s -mtriple=aarch64 -mattr=+sme -mattr=+sve | FileCheck %s --check-prefix=CHECK1024
+
+;; The hazard size can still be overridden/disabled when +sme,+sve is set.
+; RUN: llc < %s -mtriple=aarch64 -mattr=+sme -mattr=+sve -aarch64-stack-hazard-size=0 | FileCheck %s --check-prefix=CHECK0
+
+;; When +sme is set (without +sve) the default hazard size should be 0.
+; RUN: llc < %s -mtriple=aarch64 -mattr=+sme | FileCheck %s --check-prefix=CHECK0
+
+define i32 @spill_fpr_with_gpr_stack_object(i64 %d) "aarch64_pstate_sm_compatible" {
+; CHECK0-LABEL: spill_fpr_with_gpr_stack_object:
+; CHECK0:       // %bb.0: // %entry
+; CHECK0-NEXT:    str d8, [sp, #-16]! // 8-byte Folded Spill
+; CHECK0-NEXT:    .cfi_def_cfa_offset 16
+; CHECK0-NEXT:    .cfi_offset b8, -16
+; CHECK0-NEXT:    mov x8, x0
+; CHECK0-NEXT:    mov w0, wzr
+; CHECK0-NEXT:    //APP
+; CHECK0-NEXT:    //NO_APP
+; CHECK0-NEXT:    str x8, [sp, #8]
+; CHECK0-NEXT:    ldr d8, [sp], #16 // 8-byte Folded Reload
+; CHECK0-NEXT:    ret
+;
+; CHECK1024-LABEL: spill_fpr_with_gpr_stack_object:
+; CHECK1024:       // %bb.0: // %entry
+; CHECK1024-NEXT:    sub sp, sp, #1040
+; CHECK1024-NEXT:    str d8, [sp] // 8-byte Folded Spill
+; CHECK1024-NEXT:    str x29, [sp, #1032] // 8-byte Folded Spill
+; CHECK1024-NEXT:    sub sp, sp, #1040
+; CHECK1024-NEXT:    .cfi_def_cfa_offset 2080
+; CHECK1024-NEXT:    .cfi_offset w29, -8
+; CHECK1024-NEXT:    .cfi_offset b8, -1040
+; CHECK1024-NEXT:    mov x8, x0
+; CHECK1024-NEXT:    mov w0, wzr
+; CHECK1024-NEXT:    //APP
+; CHECK1024-NEXT:    //NO_APP
+; CHECK1024-NEXT:    str x8, [sp, #8]
+; CHECK1024-NEXT:    add sp, sp, #1040
+; CHECK1024-NEXT:    ldr x29, [sp, #1032] // 8-byte Folded Reload
+; CHECK1024-NEXT:    ldr d8, [sp] // 8-byte Folded Reload
+; CHECK1024-NEXT:    add sp, sp, #1040
+; CHECK1024-NEXT:    ret
+entry:
+  %a = alloca i64
+  tail call void asm sideeffect "", "~{d8}"() #1
+  store i64 %d, ptr %a
+  ret i32 0
+}
diff --git a/llvm/test/CodeGen/AArch64/streaming-compatible-memory-ops.ll b/llvm/test/CodeGen/AArch64/streaming-compatible-memory-ops.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -verify-machineinstrs -aarch64-lower-to-sme-routines=false < %s | FileCheck %s -check-prefixes=CHECK-NO-SME-ROUTINES
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+sme2 -mattr=+mops -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK-MOPS
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme2 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme2 -verify-machineinstrs -aarch64-lower-to-sme-routines=false < %s | FileCheck %s -check-prefixes=CHECK-NO-SME-ROUTINES
+; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme2 -mattr=+mops -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK-MOPS
 
 @dst = global [512 x i8] zeroinitializer, align 1
 @src = global [512 x i8] zeroinitializer, align 1
diff --git a/llvm/test/CodeGen/AArch64/sve-stack-frame-layout.ll b/llvm/test/CodeGen/AArch64/sve-stack-frame-layout.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc < %s -mtriple=aarch64 -mattr=+sve2 | FileCheck %s --check-prefixes=CHECK
-; RUN: llc < %s -mtriple=aarch64 -mattr=+sve2 -pass-remarks-analysis=stack-frame-layout 2>&1 >/dev/null | FileCheck %s --check-prefixes=CHECK-FRAMELAYOUT
+; RUN: llc < %s -mtriple=aarch64 -mattr=+sve2 -aarch64-streaming-hazard-size=0 | FileCheck %s --check-prefixes=CHECK
+; RUN: llc < %s -mtriple=aarch64 -mattr=+sve2 -aarch64-streaming-hazard-size=0 -pass-remarks-analysis=stack-frame-layout 2>&1 >/dev/null | FileCheck %s --check-prefixes=CHECK-FRAMELAYOUT
 
 ; CHECK-FRAMELAYOUT-LABEL: Function: csr_d8_allocnxv4i32i32f64
 ; CHECK-FRAMELAYOUT-NEXT: Offset: [SP-8], Type: Spill, Align: 8, Size: 8

Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,4 @@`
`1`		`-; RUN: llc -mtriple=aarch64-darwin -mattr=+sve -mattr=+sme -enable-aarch64-sme-peephole-opt=false -verify-machineinstrs < %s \| FileCheck %s`
	`1`	`+; RUN: llc -mtriple=aarch64-darwin -aarch64-streaming-hazard-size=0 -mattr=+sve -mattr=+sme -enable-aarch64-sme-peephole-opt=false -verify-machineinstrs < %s \| FileCheck %s`
`2`	`2`
`3`	`3`	`declare void @normal_callee();`
`4`	`4`