[GlobalISel][AArch64] Legalize G_UADDSAT, G_SADDSAT, G_USUBSAT, and G…

[tschuett]

…_SSUBSAT

sve-int-imm.ll also tests saturation, but it has unsupported splats.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-llvm-globalisel

Author: Thorsten Schütt (tschuett)

Changes

…_SSUBSAT

---

Full diff: https://github.com/llvm/llvm-project/pull/114664.diff

3 Files Affected:

• (modified) llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp (+2)
• (modified) llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp (+1)
• (added) llvm/test/CodeGen/AArch64/sve-saturating-arith.ll (+171)

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp b/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
index b7541effafe5ce..3ea655f29fd24d 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
@@ -204,6 +204,8 @@ LegalityPredicate LegalityPredicates::memSizeInBytesNotPow2(unsigned MMOIdx) {
 LegalityPredicate LegalityPredicates::memSizeNotByteSizePow2(unsigned MMOIdx) {
   return [=](const LegalityQuery &Query) {
     const LLT MemTy = Query.MMODescrs[MMOIdx].MemoryTy;
+    if (Query.MMODescrs[MMOIdx].MemoryTy.isScalableVector())
+      return true;
     return !MemTy.isByteSized() ||
            !llvm::has_single_bit<uint32_t>(
                MemTy.getSizeInBytes().getKnownMinValue());
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index f162d1c2973cbc..8851ad7ac3503d 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -1280,6 +1280,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
 
   getActionDefinitionsBuilder({G_UADDSAT, G_SADDSAT, G_USUBSAT, G_SSUBSAT})
       .legalFor({v2s64, v2s32, v4s32, v4s16, v8s16, v8s8, v16s8})
+      .legalFor(HasSVE, {nxv2s64, nxv4s32, nxv8s16, nxv16s8})
       .clampNumElements(0, v8s8, v16s8)
       .clampNumElements(0, v4s16, v8s16)
       .clampNumElements(0, v2s32, v4s32)
diff --git a/llvm/test/CodeGen/AArch64/sve-saturating-arith.ll b/llvm/test/CodeGen/AArch64/sve-saturating-arith.ll
new file mode 100644
index 00000000000000..a14c86b3f3146f
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-saturating-arith.ll
@@ -0,0 +1,171 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64 -mattr=+sve -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SDAG
+; RUN: llc -mtriple=aarch64 -mattr=+sve -aarch64-enable-gisel-sve=1 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GISEL
+
+; SQADD
+define <vscale x 16 x i8> @sqadd_i8_low(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: sqadd_i8_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqadd z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 16 x i8> @llvm.sadd.sat.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @sqadd_i16_low(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: sqadd_i16_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqadd z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 8 x i16> @llvm.sadd.sat.nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @sqadd_i32_low(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: sqadd_i32_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqadd z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 4 x i32> @llvm.sadd.sat.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @sqadd_i64_low(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: sqadd_i64_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqadd z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 2 x i64> @llvm.sadd.sat.nxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b)
+  ret <vscale x 2 x i64> %res
+}
+
+; UQADD
+define <vscale x 16 x i8> @uqadd_i8_low(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: uqadd_i8_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqadd z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 16 x i8> @llvm.uadd.sat.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @uqadd_i16_low(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: uqadd_i16_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqadd z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 8 x i16> @llvm.uadd.sat.nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @uqadd_i32_low(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: uqadd_i32_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqadd z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 4 x i32> @llvm.uadd.sat.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @uqadd_i64_low(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: uqadd_i64_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqadd z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 2 x i64> @llvm.uadd.sat.nxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b)
+  ret <vscale x 2 x i64> %res
+}
+
+; SQSUB
+define <vscale x 16 x i8> @sqsub_i8_low(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: sqsub_i8_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqsub z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 16 x i8> @llvm.ssub.sat.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @sqsub_i16_low(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: sqsub_i16_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqsub z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 8 x i16> @llvm.ssub.sat.nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @sqsub_i32_low(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: sqsub_i32_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqsub z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 4 x i32> @llvm.ssub.sat.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @sqsub_i64_low(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: sqsub_i64_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sqsub z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 2 x i64> @llvm.ssub.sat.nxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b)
+  ret <vscale x 2 x i64> %res
+}
+
+; UQSUB
+define <vscale x 16 x i8> @uqsub_i8_low(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: uqsub_i8_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqsub z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 16 x i8> @llvm.usub.sat.nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @uqsub_i16_low(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: uqsub_i16_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqsub z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 8 x i16> @llvm.usub.sat.nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @uqsub_i32_low(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: uqsub_i32_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqsub z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 4 x i32> @llvm.usub.sat.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @uqsub_i64_low(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: uqsub_i64_low:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uqsub z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %res =  call <vscale x 2 x i64> @llvm.usub.sat.nxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b)
+  ret <vscale x 2 x i64> %res
+}
+
+declare <vscale x 16 x i8> @llvm.sadd.sat.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>)
+declare <vscale x 8 x i16> @llvm.sadd.sat.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>)
+declare <vscale x 4 x i32> @llvm.sadd.sat.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>)
+declare <vscale x 2 x i64> @llvm.sadd.sat.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 16 x i8> @llvm.uadd.sat.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>)
+declare <vscale x 8 x i16> @llvm.uadd.sat.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>)
+declare <vscale x 4 x i32> @llvm.uadd.sat.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>)
+declare <vscale x 2 x i64> @llvm.uadd.sat.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 16 x i8> @llvm.ssub.sat.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>)
+declare <vscale x 8 x i16> @llvm.ssub.sat.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>)
+declare <vscale x 4 x i32> @llvm.ssub.sat.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>)
+declare <vscale x 2 x i64> @llvm.ssub.sat.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 16 x i8> @llvm.usub.sat.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>)
+declare <vscale x 8 x i16> @llvm.usub.sat.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>)
+declare <vscale x 4 x i32> @llvm.usub.sat.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>)
+declare <vscale x 2 x i64> @llvm.usub.sat.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>)
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK-GISEL: {{.*}}
+; CHECK-SDAG: {{.*}}

[davemgreen]

This sounds OK, but is the memsize part needed for this patch?

[davemgreen]

Thanks. LGTM