From dabe2bb4fbf47e64729591e896f7231bda0c42a7 Mon Sep 17 00:00:00 2001
From: Junyang Shao <shaojunyang@google.com>
Date: Sun, 14 Sep 2025 20:17:55 +0000
Subject: [PATCH] [dev.simd] cmd/compile: fix holes in mask peepholes

It turns out that ".Masked" is implemented by VPANDQ *and* VPANDD.
The shape of bitwise AND doesn't matter, the correctness of the rules is
guaranteed by the way the mask is generated.

This CL fix the holes in the peephole rules.

Change-Id: I2d15c4d17afed6fdbb2f3905a51b2c5c2f673348
Reviewed-on: https://go-review.googlesource.com/c/go/+/703257
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: David Chase <drchase@google.com>
---
 src/cmd/compile/internal/ssa/_gen/AMD64.rules |  4 ++
 src/cmd/compile/internal/ssa/rewriteAMD64.go  | 60 +++++++++++++++++++
 2 files changed, 64 insertions(+)

diff --git a/src/cmd/compile/internal/ssa/_gen/AMD64.rules b/src/cmd/compile/internal/ssa/_gen/AMD64.rules
index ad84ba75559..a5083958250 100644
--- a/src/cmd/compile/internal/ssa/_gen/AMD64.rules
+++ b/src/cmd/compile/internal/ssa/_gen/AMD64.rules
@@ -1768,6 +1768,10 @@
 (VPANDQ512 x (VPMOVMToVec32x16 k)) => (VMOVDQU32Masked512 x k)
 (VPANDQ512 x (VPMOVMToVec16x32 k)) => (VMOVDQU16Masked512 x k)
 (VPANDQ512 x (VPMOVMToVec8x64 k)) => (VMOVDQU8Masked512 x k)
+(VPANDD512 x (VPMOVMToVec64x8 k)) => (VMOVDQU64Masked512 x k)
+(VPANDD512 x (VPMOVMToVec32x16 k)) => (VMOVDQU32Masked512 x k)
+(VPANDD512 x (VPMOVMToVec16x32 k)) => (VMOVDQU16Masked512 x k)
+(VPANDD512 x (VPMOVMToVec8x64 k)) => (VMOVDQU8Masked512 x k)
 
 // Insert to zero of 32/64 bit floats and ints to a zero is just MOVS[SD]
 (VPINSRQ128 [0] (Zero128 <t>) y) && y.Type.IsFloat() => (VMOVSDf2v <types.TypeVec128> y)
diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go
index 01227793278..187b3ed9d64 100644
--- a/src/cmd/compile/internal/ssa/rewriteAMD64.go
+++ b/src/cmd/compile/internal/ssa/rewriteAMD64.go
@@ -34681,6 +34681,66 @@ func rewriteValueAMD64_OpAMD64VPADDQMasked512(v *Value) bool {
 func rewriteValueAMD64_OpAMD64VPANDD512(v *Value) bool {
 	v_1 := v.Args[1]
 	v_0 := v.Args[0]
+	// match: (VPANDD512 x (VPMOVMToVec64x8 k))
+	// result: (VMOVDQU64Masked512 x k)
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			x := v_0
+			if v_1.Op != OpAMD64VPMOVMToVec64x8 {
+				continue
+			}
+			k := v_1.Args[0]
+			v.reset(OpAMD64VMOVDQU64Masked512)
+			v.AddArg2(x, k)
+			return true
+		}
+		break
+	}
+	// match: (VPANDD512 x (VPMOVMToVec32x16 k))
+	// result: (VMOVDQU32Masked512 x k)
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			x := v_0
+			if v_1.Op != OpAMD64VPMOVMToVec32x16 {
+				continue
+			}
+			k := v_1.Args[0]
+			v.reset(OpAMD64VMOVDQU32Masked512)
+			v.AddArg2(x, k)
+			return true
+		}
+		break
+	}
+	// match: (VPANDD512 x (VPMOVMToVec16x32 k))
+	// result: (VMOVDQU16Masked512 x k)
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			x := v_0
+			if v_1.Op != OpAMD64VPMOVMToVec16x32 {
+				continue
+			}
+			k := v_1.Args[0]
+			v.reset(OpAMD64VMOVDQU16Masked512)
+			v.AddArg2(x, k)
+			return true
+		}
+		break
+	}
+	// match: (VPANDD512 x (VPMOVMToVec8x64 k))
+	// result: (VMOVDQU8Masked512 x k)
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			x := v_0
+			if v_1.Op != OpAMD64VPMOVMToVec8x64 {
+				continue
+			}
+			k := v_1.Args[0]
+			v.reset(OpAMD64VMOVDQU8Masked512)
+			v.AddArg2(x, k)
+			return true
+		}
+		break
+	}
 	// match: (VPANDD512 x l:(VMOVDQUload512 {sym} [off] ptr mem))
 	// cond: canMergeLoad(v, l) && clobber(l)
 	// result: (VPANDD512load {sym} [off] x ptr mem)