From 50122adb2cfca5f9f8b3a60967fbc5f5c56f6b98 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Sun, 8 Jun 2025 18:08:06 +0800
Subject: [PATCH 01/20] vbroadcast: vbroadcast: use opaque pointer

---
 src/llvm_intrin/vbroadcast.jl | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/src/llvm_intrin/vbroadcast.jl b/src/llvm_intrin/vbroadcast.jl
index 29cae45..ad8a01a 100644
--- a/src/llvm_intrin/vbroadcast.jl
+++ b/src/llvm_intrin/vbroadcast.jl
@@ -164,12 +164,10 @@ end
 ) where {W,T}
   isone(W) && return Expr(:block, Expr(:meta, :inline), :(vload(ptr)))
   typ = LLVM_TYPES[T]
-  ptyp = JULIAPOINTERTYPE
   vtyp = "<$W x $typ>"
   alignment = Base.datatype_alignment(T)
   instrs = """
-      %ptr = inttoptr $ptyp %0 to $typ*
-      %res = load $typ, $typ* %ptr, align $alignment
+      %res = load $typ, ptr %0, align $alignment
       %ie = insertelement $vtyp undef, $typ %res, i32 0
       %v = shufflevector $vtyp %ie, $vtyp undef, <$W x i32> zeroinitializer
       ret $vtyp %v

From 7499ff3c9cd081f3a6276764119be71cabbc428d Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Sun, 8 Jun 2025 18:40:55 +0800
Subject: [PATCH 02/20] memory_addr.expandload: use opaque pointer

---
 src/llvm_intrin/memory_addr.jl | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index 8ad9cf4..f58ac8f 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -2310,21 +2310,20 @@ end
 ) where {W,T<:NativeTypes,U<:Unsigned}
   typ = LLVM_TYPES[T]
   vtyp = "<$W x $typ>"
-  vptrtyp = "<$W x $typ*>"
   mtyp_input = LLVM_TYPES[U]
   mtyp_trunc = "i$W"
   instrs = String[]
-  push!(instrs, "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $typ*")
+  # push!(instrs, "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $typ*")
   if mtyp_input == mtyp_trunc
     push!(instrs, "%mask = bitcast $mtyp_input %1 to <$W x i1>")
   else
     push!(instrs, "%masktrunc = trunc $mtyp_input %1 to $mtyp_trunc")
     push!(instrs, "%mask = bitcast $mtyp_trunc %masktrunc to <$W x i1>")
   end
-  decl = "declare $vtyp @llvm.masked.expandload.$(suffix(W,T))($typ*, <$W x i1>, $vtyp)"
+  decl = "declare $vtyp @llvm.masked.expandload.$(suffix(W,T))(ptr, <$W x i1>, $vtyp)"
   push!(
     instrs,
-    "%res = call $vtyp @llvm.masked.expandload.$(suffix(W,T))($typ* %ptr, <$W x i1> %mask, $vtyp zeroinitializer)\nret $vtyp %res"
+    "%res = call $vtyp @llvm.masked.expandload.$(suffix(W,T))(ptr %0, <$W x i1> %mask, $vtyp zeroinitializer)\nret $vtyp %res"
   )
   llvmcall_expr(
     decl,

From 4c5756d468f63d090067e2bbccc422a0a9fc98d3 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Sun, 8 Jun 2025 18:41:13 +0800
Subject: [PATCH 03/20] memory_addr.compressstore: use opaque pointer

---
 src/llvm_intrin/memory_addr.jl | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index f58ac8f..fc4f71e 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -2284,12 +2284,13 @@ end
   vtyp = "<$W x $typ>"
   mtyp_input = LLVM_TYPES[U]
   mtyp_trunc = "i$W"
-  instrs = String["%ptr = inttoptr $JULIAPOINTERTYPE %1 to $typ*"]
+  instrs = String[]
+  # push!(instrs, "%ptr = inttoptr $JULIAPOINTERTYPE %1 to $typ*")
   truncate_mask!(instrs, '2', W, 0)
-  decl = "declare void @llvm.masked.compressstore.$(suffix(W,T))($vtyp, $typ*, <$W x i1>)"
+  decl = "declare void @llvm.masked.compressstore.$(suffix(W,T))($vtyp, ptr, <$W x i1>)"
   push!(
     instrs,
-    "call void @llvm.masked.compressstore.$(suffix(W,T))($vtyp %0, $typ* %ptr, <$W x i1> %mask.0)\nret void"
+    "call void @llvm.masked.compressstore.$(suffix(W,T))($vtyp %0, ptr %1, <$W x i1> %mask.0)\nret void"
   )
   llvmcall_expr(
     decl,

From e8f725369417a736e1baa8585a5ca9f08f01fdbc Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Sun, 8 Jun 2025 18:43:56 +0800
Subject: [PATCH 04/20] memory_addr.lifetime.start: use opaque pointer

---
 src/llvm_intrin/memory_addr.jl | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index fc4f71e..f59ce8f 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -2228,8 +2228,8 @@ end
 
 @generated function lifetime_start!(ptr::Ptr{T}, ::Val{L}) where {L,T}
   ptyp = LLVM_TYPES[T]
-  decl = "declare void @llvm.lifetime.start(i64, $ptyp* nocapture)"
-  instrs = "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $ptyp*\ncall void @llvm.lifetime.start(i64 $L, $ptyp* %ptr)\nret void"
+  decl = "declare void @llvm.lifetime.start(i64, ptr nocapture)"
+  instrs = "call void @llvm.lifetime.start(i64 $L, ptr %0)\nret void"
   llvmcall_expr(
     decl,
     instrs,

From 87fd6827639a5a022003cc1950da7bcdb12922e9 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Sun, 8 Jun 2025 18:44:11 +0800
Subject: [PATCH 05/20] memory_addr.lifetime.end: use opaque pointer

---
 src/llvm_intrin/memory_addr.jl | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index f59ce8f..eb025cb 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -2244,8 +2244,8 @@ end
 end
 @generated function lifetime_end!(ptr::Ptr{T}, ::Val{L}) where {L,T}
   ptyp = LLVM_TYPES[T]
-  decl = "declare void @llvm.lifetime.end(i64, $ptyp* nocapture)"
-  instrs = "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $ptyp*\ncall void @llvm.lifetime.end(i64 $L, $ptyp* %ptr)\nret void"
+  decl = "declare void @llvm.lifetime.end(i64, ptr nocapture)"
+  instrs = "call void @llvm.lifetime.end(i64 $L, ptr %0)\nret void"
   llvmcall_expr(
     decl,
     instrs,

From 12959244f80505cafa6ec2cc83804760cbdded18 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Sun, 8 Jun 2025 21:22:31 +0800
Subject: [PATCH 06/20] memory_addr.prefetch: use opaque pointer

---
 src/llvm_intrin/memory_addr.jl | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index eb025cb..7fb820e 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -2170,10 +2170,9 @@ end
       "Prefetch intrinsic requires a read/write argument of 0, 1, but received $R."
     )
   )
-  decl = "declare void @llvm.prefetch(i8*, i32, i32, i32)"
+  decl = "declare void @llvm.prefetch(ptr, i32, i32, i32)"
   instrs = """
-      %addr = inttoptr $JULIAPOINTERTYPE %0 to i8*
-      call void @llvm.prefetch(i8* %addr, i32 $R, i32 $L, i32 1)
+      call void @llvm.prefetch(ptr %0, i32 $R, i32 $L, i32 1)
       ret void
   """
   llvmcall_expr(

From 9f161a5e1d739897fb6bdba10b9a4e2568590bf7 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 00:27:01 +0800
Subject: [PATCH 07/20] memory_addr.offset_ptr: remove `inttoptr`

---
 src/llvm_intrin/memory_addr.jl | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index 7fb820e..a3063dc 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -168,7 +168,7 @@ function offset_ptr(
   if iszero(O)
     push!(
       instrs,
-      "%ptr.$(i) = inttoptr $(JULIAPOINTERTYPE) %0 to $(index_gep_typ)*"
+      "%ptr.$(i) = bitcast ptr %0 to ptr"  # do-nothing
     )
     i += 1
   else # !iszero(O)
@@ -181,7 +181,7 @@ function offset_ptr(
     end
     push!(
       instrs,
-      "%ptr.$(i) = inttoptr $(JULIAPOINTERTYPE) %0 to $(offset_gep_typ)*"
+      "%ptr.$(i) = bitcast ptr %0 to ptr"  # do-nothing
     )
     i += 1
     push!(

From 8bae6f53b4b6575b88d0c7424629beaca3f09934 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 00:28:36 +0800
Subject: [PATCH 08/20] memory_addr.offset_ptr: remove `ptrtoint`

---
 src/llvm_intrin/memory_addr.jl | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index a3063dc..d4fb09c 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -192,7 +192,7 @@ function offset_ptr(
     if forgep && iszero(M) && (iszero(X) || isone(X))
       push!(
         instrs,
-        "%ptr.$(i) = ptrtoint $(offset_gep_typ)* %ptr.$(i-1) to $(JULIAPOINTERTYPE)"
+        "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothiong
       )
       i += 1
       return instrs, i
@@ -224,7 +224,7 @@ function offset_ptr(
     if forgep
       push!(
         instrs,
-        "%ptr.$(i) = ptrtoint <$W x $index_gep_typ*> %ptr.$(i-1) to <$W x $JULIAPOINTERTYPE>"
+        "%ptr.$(i) = bitcast <$W x ptr> %ptr.$(i-1) to <$W x ptr>"  # do-nothing
       )
       i += 1
     elseif index_gep_typ != vtyp
@@ -305,7 +305,7 @@ function offset_ptr(
     if forgep
       push!(
         instrs,
-        "%ptr.$(i) = ptrtoint <$W x $typ*> %ptr.$(i-1) to <$W x $JULIAPOINTERTYPE>"
+        "%ptr.$(i) = bitcast <$W x ptr> %ptr.$(i-1) to <$W x ptr>"  # do-nothing
       )
       i += 1
     end
@@ -314,7 +314,7 @@ function offset_ptr(
   if forgep # if forgep, just return now
     push!(
       instrs,
-      "%ptr.$(i) = ptrtoint $(index_gep_typ)* %ptr.$(i-1) to $JULIAPOINTERTYPE"
+      "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
     )
     i += 1
   elseif index_gep_typ != vtyp

From 1b45d447cbcecf80afcb96c320d0b9a88cae67c1 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 00:30:11 +0800
Subject: [PATCH 09/20] memory_addr.offset_ptr: remove `ptrtoint`

---
 src/llvm_intrin/memory_addr.jl | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index d4fb09c..2174fc4 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -199,7 +199,7 @@ function offset_ptr(
     elseif offset_gep_typ != index_gep_typ
       push!(
         instrs,
-        "%ptr.$(i) = bitcast $(offset_gep_typ)* %ptr.$(i-1) to $(index_gep_typ)*"
+        "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
       )
       i += 1
     end
@@ -230,7 +230,7 @@ function offset_ptr(
     elseif index_gep_typ != vtyp
       push!(
         instrs,
-        "%ptr.$(i) = bitcast <$W x $index_gep_typ*> %ptr.$(i-1) to <$W x $typ*>"
+        "%ptr.$(i) = bitcast <$W x ptr> %ptr.$(i-1) to <$W x ptr>"  # do-nothing
       )
       i += 1
     end
@@ -293,7 +293,7 @@ function offset_ptr(
     if typ !== index_gep_typ
       push!(
         instrs,
-        "%ptr.$(i) = bitcast $(index_gep_typ)* %ptr.$(i-1) to $(typ)*"
+        "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
       )
       i += 1
     end
@@ -320,7 +320,7 @@ function offset_ptr(
   elseif index_gep_typ != vtyp
     push!(
       instrs,
-      "%ptr.$(i) = bitcast $(index_gep_typ)* %ptr.$(i-1) to $(vtyp)*"
+      "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
     )
     i += 1
   end

From 69d4855c21a06720c97e07a32a50fd011e9e6757 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 00:31:55 +0800
Subject: [PATCH 10/20] memory_addr.offset_ptr: fix `getelementptr`

---
 src/llvm_intrin/memory_addr.jl | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index 2174fc4..bffd505 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -186,7 +186,7 @@ function offset_ptr(
     i += 1
     push!(
       instrs,
-      "%ptr.$(i) = getelementptr inbounds $(offset_gep_typ), $(offset_gep_typ)* %ptr.$(i-1), i32 $(offset)"
+      "%ptr.$(i) = getelementptr inbounds $(offset_gep_typ), ptr %ptr.$(i-1), i32 $(offset)"
     )
     i += 1
     if forgep && iszero(M) && (iszero(X) || isone(X))
@@ -218,7 +218,7 @@ function offset_ptr(
     end
     push!(
       instrs,
-      "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), $(index_gep_typ)* %ptr.$(i-1), <$W x i$(ibits)> %$(indname)"
+      "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), ptr %ptr.$(i-1), <$W x i$(ibits)> %$(indname)"
     )
     i += 1
     if forgep
@@ -278,7 +278,7 @@ function offset_ptr(
     end
     push!(
       instrs,
-      "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), $(index_gep_typ)* %ptr.$(i-1), i$(ibits) %$(indname)"
+      "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), ptr %ptr.$(i-1), i$(ibits) %$(indname)"
     )
     i += 1
   end
@@ -299,7 +299,7 @@ function offset_ptr(
     end
     push!(
       instrs,
-      "%ptr.$(i) = getelementptr inbounds $(typ), $(typ)* %ptr.$(i-1), <$W x $(vityp)> <$vityp $vi>"
+      "%ptr.$(i) = getelementptr inbounds $(typ), ptr %ptr.$(i-1), <$W x $(vityp)> <$vityp $vi>"
     )
     i += 1
     if forgep

From a8b99642394487ca56f2f3cdb8a38134c47a3b26 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Sun, 8 Jun 2025 23:58:26 +0800
Subject: [PATCH 11/20] memory_addr: fix vstore_quote

---
 src/llvm_intrin/memory_addr.jl | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index bffd505..5860c4c 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -1254,33 +1254,33 @@ function vstore_quote(
       suffix(W, T_sym) *
       '.' *
       ptr_suffix(W, T_sym)
-    decl *= "declare $storeinstr($vtyp, <$W x $typ*>, i32, <$W x i1>)"
+    decl *= "declare $storeinstr($vtyp, <$W x ptr>, i32, <$W x i1>)"
     m = mask ? m = "%mask.0" : llvmconst(W, "i1 1")
     push!(
       instrs,
-      "call $storeinstr($vtyp $(argtostore), <$W x $typ*> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)" *
+      "call $storeinstr($vtyp $(argtostore), <$W x ptr> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)" *
       metadata
     )
-    # push!(instrs, "call $storeinstr($vtyp $(argtostore), <$W x $typ*> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)")
+    # push!(instrs, "call $storeinstr($vtyp $(argtostore), <$W x ptr> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)")
   elseif mask
     suff = suffix(W, T_sym)
     storeinstr = "void @llvm.masked.store." * suff * ".p0" * suff
-    decl *= "declare $storeinstr($vtyp, $vtyp*, i32, <$W x i1>)"
+    decl *= "declare $storeinstr($vtyp, ptr, i32, <$W x i1>)"
     push!(
       instrs,
-      "call $storeinstr($vtyp $(argtostore), $vtyp* %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0)" *
+      "call $storeinstr($vtyp $(argtostore), ptr %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0)" *
       metadata
     )
   elseif nontemporal
     push!(
       instrs,
-      "store $vtyp $(argtostore), $vtyp* %ptr.$(i-1), align $alignment, !nontemporal !{i32 1}" *
+      "store $vtyp $(argtostore), ptr %ptr.$(i-1), align $alignment, !nontemporal !{i32 1}" *
       metadata
     )
   else
     push!(
       instrs,
-      "store $vtyp $(argtostore), $vtyp* %ptr.$(i-1), align $alignment" *
+      "store $vtyp $(argtostore), ptr %ptr.$(i-1), align $alignment" *
       metadata
     )
   end
@@ -1298,7 +1298,7 @@ function vstore_quote(
   else
     Expr(:curly, :Tuple, ptrtyp, T_sym)
   end
-  largs = String[JULIAPOINTERTYPE, vtyp]
+  largs = String["ptr", vtyp]
   arg_syms = Union{Symbol,Expr}[:ptr, Expr(:call, :data, :v)]
   if dynamic_index
     push!(arg_syms, :(data(i)))

From 66451ef956950f44830211870e316bd784707550 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 00:00:10 +0800
Subject: [PATCH 12/20] memory_addr: fix vload_quote_llvmcall_core

---
 src/llvm_intrin/memory_addr.jl | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index 5860c4c..9a17f34 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -822,27 +822,27 @@ function vload_quote_llvmcall_core(
       suffix(W, T_sym) *
       '.' *
       ptr_suffix(W, T_sym)
-    decl *= "declare $loadinstr(<$W x $typ*>, i32, <$W x i1>, $vtyp)"
+    decl *= "declare $loadinstr(<$W x ptr>, i32, <$W x i1>, $vtyp)"
     m = mask ? m = "%mask.0" : llvmconst(W, "i1 1")
     passthrough = mask ? "zeroinitializer" : "undef"
     push!(
       instrs,
-      "%res = call $loadinstr(<$W x $typ*> %ptr.$(i-1), i32 $alignment, <$W x i1> $m, $vtyp $passthrough)" *
+      "%res = call $loadinstr(<$W x ptr> %ptr.$(i-1), i32 $alignment, <$W x i1> $m, $vtyp $passthrough)" *
       LOAD_SCOPE_TBAA_FLAGS
     )
   elseif mask
     suff = suffix(W, T_sym)
     loadinstr = "$vtyp @llvm.masked.load." * suff * ".p0" * suff
-    decl *= "declare $loadinstr($vtyp*, i32, <$W x i1>, $vtyp)"
+    decl *= "declare $loadinstr(ptr, i32, <$W x i1>, $vtyp)"
     push!(
       instrs,
-      "%res = call $loadinstr($vtyp* %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0, $vtyp zeroinitializer)" *
+      "%res = call $loadinstr(ptr %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0, $vtyp zeroinitializer)" *
       LOAD_SCOPE_TBAA_FLAGS
     )
   else
     push!(
       instrs,
-      "%res = load $vtyp, $vtyp* %ptr.$(i-1), align $alignment" *
+      "%res = load $vtyp, ptr %ptr.$(i-1), align $alignment" *
       LOAD_SCOPE_TBAA_FLAGS
     )
   end
@@ -886,7 +886,7 @@ function vload_quote_llvmcall_core(
     end
   end
   args = Expr(:curly, :Tuple, Expr(:curly, :Ptr, T_sym))
-  largs = String[JULIAPOINTERTYPE]
+  largs = String["ptr"]
   arg_syms = Union{Symbol,Expr}[:ptr]
   if dynamic_index
     push!(arg_syms, :(data(i)))

From 8b1a943f00d39326fc73c741e80ffd38287d0329 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 00:34:20 +0800
Subject: [PATCH 13/20] memory_addr.gep_quote: fix `lret` and `largs`

---
 src/llvm_intrin/memory_addr.jl | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index 9a17f34..ef07792 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -373,14 +373,14 @@ function gep_quote(
   # ::Type{T}, ind_type::Symbol, indargname = '1', ibytes::Int, W::Int = 1, X::Int = 1, M::Int = 1, O::Int = 0, forgep::Bool = false
   instrs, i = offset_ptr(T_sym, ind_type, '1', ibits, W, X, M, O, true, rs)
   ret = Expr(:curly, :Ptr, T_sym)
-  lret = JULIAPOINTERTYPE
+  lret = "ptr"
   if gep_returns_vector(W, X, M, ind_type)
     ret = Expr(:curly, :_Vec, W, ret)
     lret = "<$W x $lret>"
   end
 
   args = Expr(:curly, :Tuple, Expr(:curly, :Ptr, T_sym))
-  largs = String[JULIAPOINTERTYPE]
+  largs = String["ptr"]
   arg_syms = Union{Symbol,Expr}[:ptr]
 
   if !(iszero(M) || ind_type === :StaticInt)

From 34be75acd5c1c11fd8b41472cf316c3cfed08c1d Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 01:01:30 +0800
Subject: [PATCH 14/20] ci: Update nightly

---
 .github/workflows/ci-julia-nightly.yml | 23 ++++++-----------------
 1 file changed, 6 insertions(+), 17 deletions(-)

diff --git a/.github/workflows/ci-julia-nightly.yml b/.github/workflows/ci-julia-nightly.yml
index 266ad2b..ae43ddc 100644
--- a/.github/workflows/ci-julia-nightly.yml
+++ b/.github/workflows/ci-julia-nightly.yml
@@ -9,7 +9,7 @@ on:
     tags: '*'
 jobs:
   test-julia-nightly:
-    name: Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }}
+    name: Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ github.event_name }}
     runs-on: ${{ matrix.os }}
     strategy:
       fail-fast: false
@@ -18,29 +18,18 @@ jobs:
           - 'nightly'
         os:
           - ubuntu-latest
-          - macOS-latest
+          - macOS-latest  # Arm
           - windows-latest
-        arch:
-          - x64
+        # Use default arch
     steps:
       - uses: actions/checkout@v4
       - uses: julia-actions/setup-julia@v2
         with:
           version: ${{ matrix.version }}
-          arch: ${{ matrix.arch }}
-      - uses: actions/cache@v4
-        env:
-          cache-name: cache-artifacts
-        with:
-          path: ~/.julia/artifacts
-          key: ${{ runner.os }}-test-${{ env.cache-name }}-${{ hashFiles('**/Project.toml') }}
-          restore-keys: |
-            ${{ runner.os }}-test-${{ env.cache-name }}-
-            ${{ runner.os }}-test-
-            ${{ runner.os }}-
+      - uses: julia-actions/cache@v2
       - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
       - uses: julia-actions/julia-processcoverage@v1
-      - uses: codecov/codecov-action@v3
+      - uses: codecov/codecov-action@v5
         with:
-          file: lcov.info
+          files: lcov.info

From 5162645cea0e9592db3ed3b4a7cba2fec213ac4f Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 01:30:30 +0800
Subject: [PATCH 15/20] ci.Aqua: turn off piracy tests

---
 test/runtests.jl | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/test/runtests.jl b/test/runtests.jl
index 7afe72e..1f8363d 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -9,7 +9,10 @@ include("testsetup.jl")
   println("Aqua.test_all")
   t0 = time_ns()
   deps_compat = VERSION <= v"1.8" || isempty(VERSION.prerelease)
-  Aqua.test_all(VectorizationBase; deps_compat = deps_compat)
+  Aqua.test_all(VectorizationBase;
+    deps_compat = deps_compat,
+    piracies=false
+  )
   println("Aqua took $((time_ns() - t0)*1e-9) seconds")
   # @test isempty(detect_unbound_args(VectorizationBase))
   # @test isempty(detect_ambiguities(VectorizationBase))

From 54beef3de01ad213accf23e90594a9bf64f84fc2 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 01:47:16 +0800
Subject: [PATCH 16/20] doc: update jldoctest

---
 src/VectorizationBase.jl | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/VectorizationBase.jl b/src/VectorizationBase.jl
index 16b594e..cd6cb62 100644
--- a/src/VectorizationBase.jl
+++ b/src/VectorizationBase.jl
@@ -167,7 +167,7 @@ julia> rgbs = [
            B = Float32(i + 200) / 255
          ) for i = 0:7:49
        ]
-8-element Vector{NamedTuple{(:R, :G, :B), Tuple{Float32, Float32, Float32}}}:
+8-element Vector{@NamedTuple{R::Float32, G::Float32, B::Float32}}:
  (R = 0.0, G = 0.39215687, B = 0.78431374)
  (R = 0.02745098, G = 0.41960785, B = 0.8117647)
  (R = 0.05490196, G = 0.44705883, B = 0.8392157)

From f26e2088f419b8b961103944bed03faa7e4ccd0e Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Mon, 9 Jun 2025 01:47:28 +0800
Subject: [PATCH 17/20] ci: Move doc CI to `doc.yml`

---
 .github/workflows/ci.yml  | 43 -------------------------------
 .github/workflows/doc.yml | 53 +++++++++++++++++++++++++++++++++++++++
 docs/make.jl              |  3 ++-
 3 files changed, 55 insertions(+), 44 deletions(-)
 create mode 100644 .github/workflows/doc.yml

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 6d860d0..eac35b5 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -44,46 +44,3 @@ jobs:
       - uses: codecov/codecov-action@v3
         with:
           file: lcov.info
-  docs:
-    name: Documentation
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: 'nightly'
-      - run: julia --color=yes -e 'using Pkg; VERSION >= v"1.5-" && !isdir(joinpath(DEPOT_PATH[1], "registries", "General")) && Pkg.Registry.add("General")'
-        shell: bash
-        env:
-          JULIA_PKG_SERVER: ""
-      - run: |
-          julia --project=docs -e '
-            using Pkg
-            Pkg.develop(PackageSpec(path=pwd()))
-            Pkg.instantiate()'
-      - run: julia --project=docs docs/make.jl
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          DOCUMENTER_KEY: ${{ secrets.DOCUMENTER_KEY }}
-  doctests:
-    name: Doctests
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: 'nightly'
-      - run: julia --color=yes -e 'using Pkg; VERSION >= v"1.5-" && !isdir(joinpath(DEPOT_PATH[1], "registries", "General")) && Pkg.Registry.add("General")'
-        shell: bash
-        env:
-          JULIA_PKG_SERVER: ""
-      - run: |
-          julia --project=docs -e '
-            using Pkg
-            Pkg.develop(PackageSpec(path=pwd()))
-            Pkg.instantiate()'
-      - run: |
-          julia --project=docs -e '
-            using Documenter: doctest
-            using VectorizationBase
-            doctest(VectorizationBase)'
diff --git a/.github/workflows/doc.yml b/.github/workflows/doc.yml
new file mode 100644
index 0000000..2605c05
--- /dev/null
+++ b/.github/workflows/doc.yml
@@ -0,0 +1,53 @@
+name: Docs
+on:
+  pull_request:
+    branches:
+      - master
+  push:
+    branches:
+      - master
+    tags: '*'
+jobs:
+  doctests:
+    name: Doctests
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: 'nightly'
+      - uses: julia-actions/cache@v2
+      - uses: julia-actions/julia-buildpkg@v1
+      - name: Configure doc environment
+        shell: julia --project=docs --color=yes {0}
+        run: |
+          using Pkg
+          Pkg.develop(PackageSpec(path=pwd()))
+          Pkg.instantiate()
+      - name: Run doctest
+        shell: julia --color=yes --project=docs {0}
+        run: |
+          using Documenter: doctest
+          using VectorizationBase
+          doctest(VectorizationBase)
+  deploydocs:
+    name: Deploy Documentation
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: 'nightly'
+      - uses: julia-actions/cache@v2
+      - uses: julia-actions/julia-buildpkg@v1
+      - name: Configure doc environment
+        shell: julia --project=docs --color=yes {0}
+        run: |
+          using Pkg
+          Pkg.develop(PackageSpec(path=pwd()))
+          Pkg.instantiate()
+      - uses: julia-actions/julia-buildpkg@v1
+      - uses: julia-actions/julia-docdeploy@v1
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          DOCUMENTER_KEY: ${{ secrets.DOCUMENTER_KEY }}
diff --git a/docs/make.jl b/docs/make.jl
index 0383bc0..005be85 100644
--- a/docs/make.jl
+++ b/docs/make.jl
@@ -1,6 +1,8 @@
 using VectorizationBase
 using Documenter
 
+DocMeta.setdocmeta!(VectorizationBase, :DocTestSetup, :(using VectorizationBase); recursive=true)
+
 makedocs(;
   modules = [VectorizationBase],
   authors = "Chris Elrod",
@@ -11,7 +13,6 @@ makedocs(;
     canonical = "https://JuliaSIMD.github.io/VectorizationBase.jl"
   ),
   pages = ["Home" => "index.md"],
-  strict = false
 )
 
 deploydocs(; repo = "github.com/JuliaSIMD/VectorizationBase.jl")

From c679829909b6f77d9b53f06d4926411417f223a3 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Fri, 20 Jun 2025 00:57:34 +0800
Subject: [PATCH 18/20] llvmcall: fix compatibility with Julia1.6

---
 src/llvm_intrin/memory_addr.jl | 234 +++++++++++++++++++++++----------
 src/llvm_intrin/vbroadcast.jl  |  12 +-
 src/llvm_types.jl              |   8 ++
 3 files changed, 180 insertions(+), 74 deletions(-)

diff --git a/src/llvm_intrin/memory_addr.jl b/src/llvm_intrin/memory_addr.jl
index ef07792..8b08ffb 100644
--- a/src/llvm_intrin/memory_addr.jl
+++ b/src/llvm_intrin/memory_addr.jl
@@ -166,10 +166,12 @@ function offset_ptr(
   end
   # after this block, we will have a index_gep_typ pointer
   if iszero(O)
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = bitcast ptr %0 to ptr"  # do-nothing
-    )
+    else
+      "%ptr.$(i) = inttoptr $(JULIAPOINTERTYPE) %0 to $(index_gep_typ)*"
+    end
+    push!(instrs, instr)
     i += 1
   else # !iszero(O)
     if !iszero(O & (tzf - 1)) # then index_gep_typ works for the constant offset
@@ -179,28 +181,36 @@ function offset_ptr(
       offset_gep_typ = index_gep_typ
       offset = O >> tz
     end
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = bitcast ptr %0 to ptr"  # do-nothing
-    )
+    else
+      "%ptr.$(i) = inttoptr $(JULIAPOINTERTYPE) %0 to $(offset_gep_typ)*"
+    end
+    push!(instrs, instr)
     i += 1
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = getelementptr inbounds $(offset_gep_typ), ptr %ptr.$(i-1), i32 $(offset)"
-    )
+    else
+      "%ptr.$(i) = getelementptr inbounds $(offset_gep_typ), $(offset_gep_typ)* %ptr.$(i-1), i32 $(offset)"
+    end
+    push!(instrs, instr)
     i += 1
     if forgep && iszero(M) && (iszero(X) || isone(X))
-      push!(
-        instrs,
+      instr = @static if USE_OPAQUE_PTR
         "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothiong
-      )
+      else
+        "%ptr.$(i) = ptrtoint $(offset_gep_typ)* %ptr.$(i-1) to $(JULIAPOINTERTYPE)"
+      end
+      push!(instrs, instr)
       i += 1
       return instrs, i
     elseif offset_gep_typ != index_gep_typ
-      push!(
-        instrs,
+      instr = @static if USE_OPAQUE_PTR
         "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
-      )
+      else
+        "%ptr.$(i) = bitcast $(offset_gep_typ)* %ptr.$(i-1) to $(index_gep_typ)*"
+      end
+      push!(instrs, instr)
       i += 1
     end
   end
@@ -216,22 +226,28 @@ function offset_ptr(
         "%$(indname) = mul nsw <$W x i$(ibits)> %$(indargname), $(constmul)"
       )
     end
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), ptr %ptr.$(i-1), <$W x i$(ibits)> %$(indname)"
-    )
+    else
+      "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), $(index_gep_typ)* %ptr.$(i-1), <$W x i$(ibits)> %$(indname)"
+    end
+    push!(instrs, instr)
     i += 1
     if forgep
-      push!(
-        instrs,
+      instr = @static if USE_OPAQUE_PTR
         "%ptr.$(i) = bitcast <$W x ptr> %ptr.$(i-1) to <$W x ptr>"  # do-nothing
-      )
+      else
+        "%ptr.$(i) = ptrtoint <$W x $index_gep_typ*> %ptr.$(i-1) to <$W x $JULIAPOINTERTYPE>"
+      end
+      push!(instrs, instr)
       i += 1
     elseif index_gep_typ != vtyp
-      push!(
-        instrs,
+      instr = @static if USE_OPAQUE_PTR
         "%ptr.$(i) = bitcast <$W x ptr> %ptr.$(i-1) to <$W x ptr>"  # do-nothing
-      )
+      else
+        "%ptr.$(i) = bitcast <$W x $index_gep_typ*> %ptr.$(i-1) to <$W x $typ*>"
+      end
+      push!(instrs, instr)
       i += 1
     end
     return instrs, i
@@ -276,10 +292,12 @@ function offset_ptr(
       end
       # TODO: if X != 1 and X != 0, check if it is better to gep -> gep, or broadcast -> add -> gep
     end
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), ptr %ptr.$(i-1), i$(ibits) %$(indname)"
-    )
+    else
+      "%ptr.$(i) = getelementptr inbounds $(index_gep_typ), $(index_gep_typ)* %ptr.$(i-1), i$(ibits) %$(indname)"
+    end
+    push!(instrs, instr)
     i += 1
   end
   # ind_type === :Integer || ind_type === :StaticInt
@@ -291,37 +309,47 @@ function offset_ptr(
     vityp = "i$(8vibytes)"
     vi = join((X * w for w ∈ 0:W-1), ", $vityp ")
     if typ !== index_gep_typ
-      push!(
-        instrs,
+      instr = @static if USE_OPAQUE_PTR
         "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
-      )
+      else
+        "%ptr.$(i) = bitcast $(index_gep_typ)* %ptr.$(i-1) to $(typ)*"
+      end
+      push!(instrs, instr)
       i += 1
     end
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = getelementptr inbounds $(typ), ptr %ptr.$(i-1), <$W x $(vityp)> <$vityp $vi>"
-    )
+    else
+      "%ptr.$(i) = getelementptr inbounds $(typ), $(typ)* %ptr.$(i-1), <$W x $(vityp)> <$vityp $vi>"
+    end
+    push!(instrs, instr)
     i += 1
     if forgep
-      push!(
-        instrs,
+      instr = @static if USE_OPAQUE_PTR
         "%ptr.$(i) = bitcast <$W x ptr> %ptr.$(i-1) to <$W x ptr>"  # do-nothing
-      )
+      else
+        "%ptr.$(i) = ptrtoint <$W x $typ*> %ptr.$(i-1) to <$W x $JULIAPOINTERTYPE>"
+      end
+      push!(instrs, instr)
       i += 1
     end
     return instrs, i
   end
   if forgep # if forgep, just return now
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
-    )
+    else
+      "%ptr.$(i) = ptrtoint $(index_gep_typ)* %ptr.$(i-1) to $JULIAPOINTERTYPE"
+    end
+    push!(instrs, instr)
     i += 1
   elseif index_gep_typ != vtyp
-    push!(
-      instrs,
+    instr = @static if USE_OPAQUE_PTR
       "%ptr.$(i) = bitcast ptr %ptr.$(i-1) to ptr"  # do-nothing
-    )
+    else
+      "%ptr.$(i) = bitcast $(index_gep_typ)* %ptr.$(i-1) to $(vtyp)*"
+    end
+    push!(instrs, instr)
     i += 1
   end
   instrs, i
@@ -373,14 +401,22 @@ function gep_quote(
   # ::Type{T}, ind_type::Symbol, indargname = '1', ibytes::Int, W::Int = 1, X::Int = 1, M::Int = 1, O::Int = 0, forgep::Bool = false
   instrs, i = offset_ptr(T_sym, ind_type, '1', ibits, W, X, M, O, true, rs)
   ret = Expr(:curly, :Ptr, T_sym)
-  lret = "ptr"
+  lret = @static if USE_OPAQUE_PTR
+    "ptr"
+  else
+    JULIAPOINTERTYPE
+  end
   if gep_returns_vector(W, X, M, ind_type)
     ret = Expr(:curly, :_Vec, W, ret)
     lret = "<$W x $lret>"
   end
 
   args = Expr(:curly, :Tuple, Expr(:curly, :Ptr, T_sym))
-  largs = String["ptr"]
+  largs = @static if USE_OPAQUE_PTR
+    String["ptr"]
+  else
+    String[JULIAPOINTERTYPE]
+  end
   arg_syms = Union{Symbol,Expr}[:ptr]
 
   if !(iszero(M) || ind_type === :StaticInt)
@@ -816,33 +852,40 @@ function vload_quote_llvmcall_core(
       truncate_mask!(instrs, '1' + dynamic_index, W, 0, false)
     end
   end
+  @static if USE_OPAQUE_PTR
+    ptr_type = "ptr"
+    vptr_type = "ptr"
+  else
+    ptr_type = "$typ*"
+    vptr_type = "$vtyp*"
+  end
   if grv
     loadinstr =
       "$vtyp @llvm.masked.gather." *
       suffix(W, T_sym) *
       '.' *
       ptr_suffix(W, T_sym)
-    decl *= "declare $loadinstr(<$W x ptr>, i32, <$W x i1>, $vtyp)"
+    decl *= "declare $loadinstr(<$W x $ptr_type>, i32, <$W x i1>, $vtyp)"
     m = mask ? m = "%mask.0" : llvmconst(W, "i1 1")
     passthrough = mask ? "zeroinitializer" : "undef"
     push!(
       instrs,
-      "%res = call $loadinstr(<$W x ptr> %ptr.$(i-1), i32 $alignment, <$W x i1> $m, $vtyp $passthrough)" *
+      "%res = call $loadinstr(<$W x $ptr_type> %ptr.$(i-1), i32 $alignment, <$W x i1> $m, $vtyp $passthrough)" *
       LOAD_SCOPE_TBAA_FLAGS
     )
   elseif mask
     suff = suffix(W, T_sym)
     loadinstr = "$vtyp @llvm.masked.load." * suff * ".p0" * suff
-    decl *= "declare $loadinstr(ptr, i32, <$W x i1>, $vtyp)"
+    decl *= "declare $loadinstr($vptr_type, i32, <$W x i1>, $vtyp)"
     push!(
       instrs,
-      "%res = call $loadinstr(ptr %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0, $vtyp zeroinitializer)" *
+      "%res = call $loadinstr($vptr_type %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0, $vtyp zeroinitializer)" *
       LOAD_SCOPE_TBAA_FLAGS
     )
   else
     push!(
       instrs,
-      "%res = load $vtyp, ptr %ptr.$(i-1), align $alignment" *
+      "%res = load $vtyp, $vptr_type %ptr.$(i-1), align $alignment" *
       LOAD_SCOPE_TBAA_FLAGS
     )
   end
@@ -886,7 +929,11 @@ function vload_quote_llvmcall_core(
     end
   end
   args = Expr(:curly, :Tuple, Expr(:curly, :Ptr, T_sym))
-  largs = String["ptr"]
+  largs = @static if USE_OPAQUE_PTR
+    String["ptr"]
+  else
+    String[JULIAPOINTERTYPE]
+  end
   arg_syms = Union{Symbol,Expr}[:ptr]
   if dynamic_index
     push!(arg_syms, :(data(i)))
@@ -1248,39 +1295,46 @@ function vstore_quote(
   else
     argtostore = "%1"
   end
+  @static if USE_OPAQUE_PTR
+    ptr_type = "ptr"
+    vptr_type = "ptr"
+  else
+    ptr_type = "$typ*"
+    vptr_type = "$vtyp*"
+  end
   if grv
     storeinstr =
       "void @llvm.masked.scatter." *
       suffix(W, T_sym) *
       '.' *
       ptr_suffix(W, T_sym)
-    decl *= "declare $storeinstr($vtyp, <$W x ptr>, i32, <$W x i1>)"
+    decl *= "declare $storeinstr($vtyp, <$W x $ptr_type>, i32, <$W x i1>)"
     m = mask ? m = "%mask.0" : llvmconst(W, "i1 1")
     push!(
       instrs,
-      "call $storeinstr($vtyp $(argtostore), <$W x ptr> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)" *
+      "call $storeinstr($vtyp $(argtostore), <$W x $ptr_type> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)" *
       metadata
     )
-    # push!(instrs, "call $storeinstr($vtyp $(argtostore), <$W x ptr> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)")
+    # push!(instrs, "call $storeinstr($vtyp $(argtostore), <$W x $ptr_type> %ptr.$(i-1), i32 $alignment, <$W x i1> $m)")
   elseif mask
     suff = suffix(W, T_sym)
     storeinstr = "void @llvm.masked.store." * suff * ".p0" * suff
-    decl *= "declare $storeinstr($vtyp, ptr, i32, <$W x i1>)"
+    decl *= "declare $storeinstr($vtyp, $vptr_type, i32, <$W x i1>)"
     push!(
       instrs,
-      "call $storeinstr($vtyp $(argtostore), ptr %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0)" *
+      "call $storeinstr($vtyp $(argtostore), $vptr_type %ptr.$(i-1), i32 $alignment, <$W x i1> %mask.0)" *
       metadata
     )
   elseif nontemporal
     push!(
       instrs,
-      "store $vtyp $(argtostore), ptr %ptr.$(i-1), align $alignment, !nontemporal !{i32 1}" *
+      "store $vtyp $(argtostore), $vptr_type %ptr.$(i-1), align $alignment, !nontemporal !{i32 1}" *
       metadata
     )
   else
     push!(
       instrs,
-      "store $vtyp $(argtostore), ptr %ptr.$(i-1), align $alignment" *
+      "store $vtyp $(argtostore), $vptr_type %ptr.$(i-1), align $alignment" *
       metadata
     )
   end
@@ -1298,7 +1352,12 @@ function vstore_quote(
   else
     Expr(:curly, :Tuple, ptrtyp, T_sym)
   end
-  largs = String["ptr", vtyp]
+  largs = @static if USE_OPAQUE_PTR
+    String["ptr"]
+  else
+    String[JULIAPOINTERTYPE]
+  end
+  push!(largs, vtyp)
   arg_syms = Union{Symbol,Expr}[:ptr, Expr(:call, :data, :v)]
   if dynamic_index
     push!(arg_syms, :(data(i)))
@@ -2170,11 +2229,20 @@ end
       "Prefetch intrinsic requires a read/write argument of 0, 1, but received $R."
     )
   )
-  decl = "declare void @llvm.prefetch(ptr, i32, i32, i32)"
-  instrs = """
+  @static if USE_OPAQUE_PTR
+    decl = "declare void @llvm.prefetch(ptr, i32, i32, i32)"
+    instrs = """
       call void @llvm.prefetch(ptr %0, i32 $R, i32 $L, i32 1)
       ret void
-  """
+    """
+  else
+    decl = "declare void @llvm.prefetch(i8*, i32, i32, i32)"
+    instrs = """
+      %addr = inttoptr $JULIAPOINTERTYPE %0 to i8*
+      call void @llvm.prefetch(i8* %addr, i32 $R, i32 $L, i32 1)
+      ret void
+    """
+  end
   llvmcall_expr(
     decl,
     instrs,
@@ -2227,8 +2295,13 @@ end
 
 @generated function lifetime_start!(ptr::Ptr{T}, ::Val{L}) where {L,T}
   ptyp = LLVM_TYPES[T]
-  decl = "declare void @llvm.lifetime.start(i64, ptr nocapture)"
-  instrs = "call void @llvm.lifetime.start(i64 $L, ptr %0)\nret void"
+  @static if USE_OPAQUE_PTR
+    decl = "declare void @llvm.lifetime.start(i64, ptr nocapture)"
+    instrs = "call void @llvm.lifetime.start(i64 $L, ptr %0)\nret void"
+  else
+    decl = "declare void @llvm.lifetime.start(i64, $ptyp* nocapture)"
+    instrs = "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $ptyp*\ncall void @llvm.lifetime.start(i64 $L, $ptyp* %ptr)\nret void"
+  end
   llvmcall_expr(
     decl,
     instrs,
@@ -2243,8 +2316,13 @@ end
 end
 @generated function lifetime_end!(ptr::Ptr{T}, ::Val{L}) where {L,T}
   ptyp = LLVM_TYPES[T]
-  decl = "declare void @llvm.lifetime.end(i64, ptr nocapture)"
-  instrs = "call void @llvm.lifetime.end(i64 $L, ptr %0)\nret void"
+  @static if USE_OPAQUE_PTR
+    decl = "declare void @llvm.lifetime.end(i64, ptr nocapture)"
+    instrs = "call void @llvm.lifetime.end(i64 $L, ptr %0)\nret void"
+  else
+    decl = "declare void @llvm.lifetime.end(i64, $ptyp* nocapture)"
+    instrs = "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $ptyp*\ncall void @llvm.lifetime.end(i64 $L, $ptyp* %ptr)\nret void"
+  end
   llvmcall_expr(
     decl,
     instrs,
@@ -2283,13 +2361,18 @@ end
   vtyp = "<$W x $typ>"
   mtyp_input = LLVM_TYPES[U]
   mtyp_trunc = "i$W"
-  instrs = String[]
-  # push!(instrs, "%ptr = inttoptr $JULIAPOINTERTYPE %1 to $typ*")
+  @static if USE_OPAQUE_PTR
+    ptr_type = "ptr"
+    instrs = String["%ptr = bitcast ptr %1 to ptr"]
+  else
+    ptr_type = "$typ*"
+    instrs = String["%ptr = inttoptr $JULIAPOINTERTYPE %1 to $ptr_type"]
+  end
   truncate_mask!(instrs, '2', W, 0)
-  decl = "declare void @llvm.masked.compressstore.$(suffix(W,T))($vtyp, ptr, <$W x i1>)"
+  decl = "declare void @llvm.masked.compressstore.$(suffix(W,T))($vtyp, $ptr_type, <$W x i1>)"
   push!(
     instrs,
-    "call void @llvm.masked.compressstore.$(suffix(W,T))($vtyp %0, ptr %1, <$W x i1> %mask.0)\nret void"
+    "call void @llvm.masked.compressstore.$(suffix(W,T))($vtyp %0, $ptr_type %ptr, <$W x i1> %mask.0)\nret void"
   )
   llvmcall_expr(
     decl,
@@ -2310,20 +2393,27 @@ end
 ) where {W,T<:NativeTypes,U<:Unsigned}
   typ = LLVM_TYPES[T]
   vtyp = "<$W x $typ>"
+  vptrtyp = "<$W x $typ*>"
   mtyp_input = LLVM_TYPES[U]
   mtyp_trunc = "i$W"
   instrs = String[]
-  # push!(instrs, "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $typ*")
+  @static if USE_OPAQUE_PTR
+    ptr_type = "ptr"
+    push!(instrs, "%ptr = bitcast ptr %1 to ptr")
+  else
+    ptr_type = "$typ*"
+    push!(instrs, "%ptr = inttoptr $JULIAPOINTERTYPE %0 to $ptr_type")
+  end
   if mtyp_input == mtyp_trunc
     push!(instrs, "%mask = bitcast $mtyp_input %1 to <$W x i1>")
   else
     push!(instrs, "%masktrunc = trunc $mtyp_input %1 to $mtyp_trunc")
     push!(instrs, "%mask = bitcast $mtyp_trunc %masktrunc to <$W x i1>")
   end
-  decl = "declare $vtyp @llvm.masked.expandload.$(suffix(W,T))(ptr, <$W x i1>, $vtyp)"
+  decl = "declare $vtyp @llvm.masked.expandload.$(suffix(W,T))($ptr_type, <$W x i1>, $vtyp)"
   push!(
     instrs,
-    "%res = call $vtyp @llvm.masked.expandload.$(suffix(W,T))(ptr %0, <$W x i1> %mask, $vtyp zeroinitializer)\nret $vtyp %res"
+    "%res = call $vtyp @llvm.masked.expandload.$(suffix(W,T))($ptr_type %ptr, <$W x i1> %mask, $vtyp zeroinitializer)\nret $vtyp %res"
   )
   llvmcall_expr(
     decl,
diff --git a/src/llvm_intrin/vbroadcast.jl b/src/llvm_intrin/vbroadcast.jl
index ad8a01a..ef4aac4 100644
--- a/src/llvm_intrin/vbroadcast.jl
+++ b/src/llvm_intrin/vbroadcast.jl
@@ -164,10 +164,18 @@ end
 ) where {W,T}
   isone(W) && return Expr(:block, Expr(:meta, :inline), :(vload(ptr)))
   typ = LLVM_TYPES[T]
+  ptyp = JULIAPOINTERTYPE
   vtyp = "<$W x $typ>"
   alignment = Base.datatype_alignment(T)
-  instrs = """
-      %res = load $typ, ptr %0, align $alignment
+  instrs = @static if USE_OPAQUE_PTR
+    "%res = load $typ, ptr %0, align $alignment"
+  else
+    """
+    %ptr = inttoptr $ptyp %0 to $typ*
+    %res = load $typ, $typ* %ptr, align $alignment
+    """
+  end
+  instrs *= """
       %ie = insertelement $vtyp undef, $typ %res, i32 0
       %v = shufflevector $vtyp %ie, $vtyp undef, <$W x i32> zeroinitializer
       ret $vtyp %v
diff --git a/src/llvm_types.jl b/src/llvm_types.jl
index 266ceb2..0456428 100644
--- a/src/llvm_types.jl
+++ b/src/llvm_types.jl
@@ -114,6 +114,14 @@ function _get_alignment(W::Int, sym::Symbol)::Int
   end
 end
 
+"""
+use opaque pointer
+
+Ref:
+- Switch LLVM codegen of Ptr{T} to an actual pointer type.
+  https://github.com/JuliaLang/julia/pull/53687
+"""
+const USE_OPAQUE_PTR = VERSION >= v"1.12-DEV"
 const JULIAPOINTERTYPE = 'i' * string(8sizeof(Int))
 
 vtype(W, typ::String) = (isone(abs(W)) ? typ : "<$W x $typ>")::String

From 6c9dad56d62442fbf5d2bcb1a72fe978e8829581 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Fri, 20 Jun 2025 01:13:42 +0800
Subject: [PATCH 19/20] test: fix Auqa compat warn

---
 Project.toml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/Project.toml b/Project.toml
index 4673363..819539e 100644
--- a/Project.toml
+++ b/Project.toml
@@ -26,6 +26,7 @@ LinearAlgebra = "1"
 SIMDTypes = "0.1"
 Static = "0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 1"
 StaticArrayInterface = "1"
+Test = "<0.0.1, 1"
 julia = "1.6"
 
 [extras]

From e90e946a5bf450a7ad8a8eef60fbf8fc9ed85de5 Mon Sep 17 00:00:00 2001
From: Chengyu HAN <cyhan.dev@outlook.com>
Date: Fri, 20 Jun 2025 01:14:38 +0800
Subject: [PATCH 20/20] ci: test with macOS-13

---
 .github/workflows/ci.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index eac35b5..10c1326 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -18,7 +18,7 @@ jobs:
           - '^1.6.0-0'
         os:
           - ubuntu-latest
-          - macOS-latest
+          - macos-13  # Intel
           - windows-latest
         arch:
           - x64