[WebAssembly] Avoid dot for v16i8 partial_smla

[sparker-arm]

The sequence is shorter, by two extend operations, if we just use extmul and extadd_pairwise.

[llvmbot]

@llvm/pr-subscribers-backend-webassembly

Author: Sam Parker (sparker-arm)

Changes

The sequence is shorter, by two extend operations, if we just use extmul and extadd_pairwise.

---

Full diff: https://github.com/llvm/llvm-project/pull/163796.diff

2 Files Affected:

• (modified) llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td (+3-5)
• (modified) llvm/test/CodeGen/WebAssembly/int-mac-reduction-loops.ll (+4-6)

diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
index eedfdb309d289..29713b5d236da 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
@@ -1583,11 +1583,9 @@ def : Pat<(v4i32 (partial_reduce_umla (v4i32 V128:$acc), (v8i16 V128:$lhs),
 // MLA: v16i8 -> v4i32
 def : Pat<(v4i32 (partial_reduce_smla (v4i32 V128:$acc), (v16i8 V128:$lhs),
                                                          (v16i8 V128:$rhs))),
-          (ADD_I32x4 (ADD_I32x4 (DOT (extend_low_s_I16x8 $lhs),
-                                     (extend_low_s_I16x8 $rhs)),
-                                (DOT (extend_high_s_I16x8 $lhs),
-                                     (extend_high_s_I16x8 $rhs))),
-                      $acc)>;
+          (ADD_I32x4 (ADD_I32x4 (extadd_pairwise_s_I32x4 (EXTMUL_LOW_S_I16x8 $lhs, $rhs)),
+                                (extadd_pairwise_s_I32x4 (EXTMUL_HIGH_S_I16x8 $lhs, $rhs))),
+                     $acc)>;
 def : Pat<(v4i32 (partial_reduce_umla (v4i32 V128:$acc), (v16i8 V128:$lhs),
                                                          (v16i8 V128:$rhs))),
           (ADD_I32x4 (ADD_I32x4 (extadd_pairwise_u_I32x4 (EXTMUL_LOW_U_I16x8 $lhs, $rhs)),
diff --git a/llvm/test/CodeGen/WebAssembly/int-mac-reduction-loops.ll b/llvm/test/CodeGen/WebAssembly/int-mac-reduction-loops.ll
index 314e1b4fc69a1..91cd3dd1ca4e7 100644
--- a/llvm/test/CodeGen/WebAssembly/int-mac-reduction-loops.ll
+++ b/llvm/test/CodeGen/WebAssembly/int-mac-reduction-loops.ll
@@ -14,13 +14,11 @@ define hidden i32 @i32_mac_s8(ptr nocapture noundef readonly %a, ptr nocapture n
 ; CHECK:    i32x4.add
 
 ; MAX-BANDWIDTH: v128.load
-; MAX-BANDWIDTH: i16x8.extend_low_i8x16_s
 ; MAX-BANDWIDTH: v128.load
-; MAX-BANDWIDTH: i16x8.extend_low_i8x16_s
-; MAX-BANDWIDTH: i32x4.dot_i16x8_s
-; MAX-BANDWIDTH: i16x8.extend_high_i8x16_s
-; MAX-BANDWIDTH: i16x8.extend_high_i8x16_s
-; MAX-BANDWIDTH: i32x4.dot_i16x8_s
+; MAX-BANDWIDTH: i16x8.extmul_low_i8x16_s
+; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s
+; MAX-BANDWIDTH: i16x8.extmul_high_i8x16_s
+; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s
 ; MAX-BANDWIDTH: i32x4.add
 ; MAX-BANDWIDTH: i32x4.add