Use correct LLVM intrinsic for vmull and vaddv

Author: sayantn

crates/core_arch/src/aarch64/neon/generated.rs

@@ -623,7 +623,7 @@ pub fn vaddv_s8(a: int8x8_t) -> i8 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.saddv.i32.v8i8"
+            link_name = "llvm.aarch64.neon.saddv.i8.v8i8"
         )]
         fn _vaddv_s8(a: int8x8_t) -> i8;
     }
@@ -639,7 +639,7 @@ pub fn vaddvq_s8(a: int8x16_t) -> i8 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.saddv.i32.v16i8"
+            link_name = "llvm.aarch64.neon.saddv.i8.v16i8"
         )]
         fn _vaddvq_s8(a: int8x16_t) -> i8;
     }
@@ -655,7 +655,7 @@ pub fn vaddv_s16(a: int16x4_t) -> i16 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.saddv.i32.v4i16"
+            link_name = "llvm.aarch64.neon.saddv.i16.v4i16"
         )]
         fn _vaddv_s16(a: int16x4_t) -> i16;
     }
@@ -671,7 +671,7 @@ pub fn vaddvq_s16(a: int16x8_t) -> i16 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.saddv.i32.v8i16"
+            link_name = "llvm.aarch64.neon.saddv.i16.v8i16"
         )]
         fn _vaddvq_s16(a: int16x8_t) -> i16;
     }
@@ -719,7 +719,7 @@ pub fn vaddv_u8(a: uint8x8_t) -> u8 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.uaddv.i32.v8i8"
+            link_name = "llvm.aarch64.neon.uaddv.i8.v8i8"
         )]
         fn _vaddv_u8(a: uint8x8_t) -> u8;
     }
@@ -735,7 +735,7 @@ pub fn vaddvq_u8(a: uint8x16_t) -> u8 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.uaddv.i32.v16i8"
+            link_name = "llvm.aarch64.neon.uaddv.i8.v16i8"
         )]
         fn _vaddvq_u8(a: uint8x16_t) -> u8;
     }
@@ -751,7 +751,7 @@ pub fn vaddv_u16(a: uint16x4_t) -> u16 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.uaddv.i32.v4i16"
+            link_name = "llvm.aarch64.neon.uaddv.i16.v4i16"
         )]
         fn _vaddv_u16(a: uint16x4_t) -> u16;
     }
@@ -767,7 +767,7 @@ pub fn vaddvq_u16(a: uint16x8_t) -> u16 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.uaddv.i32.v8i16"
+            link_name = "llvm.aarch64.neon.uaddv.i16.v8i16"
         )]
         fn _vaddvq_u16(a: uint16x8_t) -> u16;
     }

crates/core_arch/src/arm_shared/neon/generated.rs

@@ -34126,9 +34126,9 @@ pub fn vmull_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.smull.v4i16"
+            link_name = "llvm.aarch64.neon.smull.v4i32"
         )]
-        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmulls.v4i16")]
+        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmulls.v4i32")]
         fn _vmull_s16(a: int16x4_t, b: int16x4_t) -> int32x4_t;
     }
     unsafe { _vmull_s16(a, b) }
@@ -34155,9 +34155,9 @@ pub fn vmull_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.smull.v2i32"
+            link_name = "llvm.aarch64.neon.smull.v2i64"
         )]
-        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmulls.v2i32")]
+        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmulls.v2i64")]
         fn _vmull_s32(a: int32x2_t, b: int32x2_t) -> int64x2_t;
     }
     unsafe { _vmull_s32(a, b) }
@@ -34184,9 +34184,9 @@ pub fn vmull_s8(a: int8x8_t, b: int8x8_t) -> int16x8_t {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.smull.v8i8"
+            link_name = "llvm.aarch64.neon.smull.v8i16"
         )]
-        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmulls.v8i8")]
+        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmulls.v8i16")]
         fn _vmull_s8(a: int8x8_t, b: int8x8_t) -> int16x8_t;
     }
     unsafe { _vmull_s8(a, b) }
@@ -34213,9 +34213,9 @@ pub fn vmull_u8(a: uint8x8_t, b: uint8x8_t) -> uint16x8_t {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.umull.v8i8"
+            link_name = "llvm.aarch64.neon.umull.v8i16"
         )]
-        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmullu.v8i8")]
+        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmullu.v8i16")]
         fn _vmull_u8(a: uint8x8_t, b: uint8x8_t) -> uint16x8_t;
     }
     unsafe { _vmull_u8(a, b) }
@@ -34242,9 +34242,9 @@ pub fn vmull_u16(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.umull.v4i16"
+            link_name = "llvm.aarch64.neon.umull.v4i32"
         )]
-        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmullu.v4i16")]
+        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmullu.v4i32")]
         fn _vmull_u16(a: uint16x4_t, b: uint16x4_t) -> uint32x4_t;
     }
     unsafe { _vmull_u16(a, b) }
@@ -34271,9 +34271,9 @@ pub fn vmull_u32(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.umull.v2i32"
+            link_name = "llvm.aarch64.neon.umull.v2i64"
         )]
-        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmullu.v2i32")]
+        #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vmullu.v2i64")]
         fn _vmull_u32(a: uint32x2_t, b: uint32x2_t) -> uint64x2_t;
     }
     unsafe { _vmull_u32(a, b) }

crates/stdarch-gen-arm/spec/neon/aarch64.spec.yml

@@ -13183,16 +13183,16 @@ intrinsics:
     assert_instr: [addv]
     safety: safe
     types:
-      - [int8x8_t, i8, i32]
-      - [int16x4_t, i16, i32]
-      - [int8x16_t, i8, i32]
-      - [int16x8_t, i16, i32]
-      - [int32x4_t, i32, i32]
+      - [int8x8_t, i8]
+      - [int16x4_t, i16]
+      - [int8x16_t, i8]
+      - [int16x8_t, i16]
+      - [int32x4_t, i32]
     compose:
       - LLVMLink:
           name: "vaddv{neon_type[0].no}"
           links:
-            - link: "llvm.aarch64.neon.saddv.{type[2]}.{neon_type[0]}"
+            - link: "llvm.aarch64.neon.saddv.{type[1]}.{neon_type[0]}"
               arch: aarch64,arm64ec
 
   - name: "vaddv{neon_type[0].no}"
@@ -13238,16 +13238,16 @@ intrinsics:
     assert_instr: [addv]
     safety: safe
     types:
-      - [uint8x8_t,  u8, i32]
-      - [uint16x4_t, u16, i32]
-      - [uint8x16_t, u8, i32]
-      - [uint16x8_t, u16, i32]
-      - [uint32x4_t, u32, i32]
+      - [uint8x8_t,  u8]
+      - [uint16x4_t, u16]
+      - [uint8x16_t, u8]
+      - [uint16x8_t, u16]
+      - [uint32x4_t, u32]
     compose:
       - LLVMLink:
           name: "vaddv{neon_type[0].no}"
           links:
-            - link: "llvm.aarch64.neon.uaddv.{type[2]}.{neon_type[0]}"
+            - link: "llvm.aarch64.neon.uaddv.{type[1]}.{neon_type[0]}"
               arch: aarch64,arm64ec
 
   - name: "vaddv{neon_type[0].no}"

crates/stdarch-gen-arm/spec/neon/arm_shared.spec.yml

@@ -6427,9 +6427,9 @@ intrinsics:
       - LLVMLink:
           name: "smull.{neon_type[1]}"
           links:
-            - link: "llvm.aarch64.neon.smull.{neon_type[1]}"
+            - link: "llvm.aarch64.neon.smull.{neon_type[2]}"
               arch: aarch64,arm64ec
-            - link: "llvm.arm.neon.vmulls.{neon_type[1]}"
+            - link: "llvm.arm.neon.vmulls.{neon_type[2]}"
               arch: arm
 
   - name: "vmull{neon_type[1].no}"
@@ -6451,9 +6451,9 @@ intrinsics:
       - LLVMLink:
           name: "smull.{neon_type[1]}"
           links:
-            - link: "llvm.aarch64.neon.umull.{neon_type[1]}"
+            - link: "llvm.aarch64.neon.umull.{neon_type[2]}"
               arch: aarch64,arm64ec
-            - link: "llvm.arm.neon.vmullu.{neon_type[1]}"
+            - link: "llvm.arm.neon.vmullu.{neon_type[2]}"
               arch: arm
 
   - name: "vmull{neon_type[1].no}"