[llvm] [RISCV][VLOPT] Add support for integer multiply-add instructions (PR #112216)

[Craig Topper via llvm-commits]

================
@@ -1248,6 +1248,203 @@ define <vscale x 4 x i64> @vwmulu_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
   ret <vscale x 4 x i64> %2
 }
 
+define <vscale x 4 x i32> @vmacc_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmacc_vv:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, tu, ma
+; NOVLOPT-NEXT:    vmacc.vv v8, v8, v10
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vmul.vv v8, v8, v10
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vmacc_vv:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, tu, ma
+; VLOPT-NEXT:    vmacc.vv v8, v8, v10
+; VLOPT-NEXT:    vsetvli zero, zero, e32, m2, ta, ma
+; VLOPT-NEXT:    vmul.vv v8, v8, v10
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i32> @llvm.riscv.vmacc.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1, iXLen 0)
+  %2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %b, iXLen %vl)
+  ret <vscale x 4 x i32> %2
+}
+
+define <vscale x 4 x i32> @vmacc_vv_use(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmacc_vv_use:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vmul.vv v12, v8, v10
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, tu, ma
+; NOVLOPT-NEXT:    vmacc.vv v8, v12, v10
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vmacc_vv_use:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vmul.vv v12, v8, v10
+; VLOPT-NEXT:    vsetvli zero, zero, e32, m2, tu, ma
+; VLOPT-NEXT:    vmacc.vv v8, v12, v10
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
+  %2 = call <vscale x 4 x i32> @llvm.riscv.vmacc.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %1, <vscale x 4 x i32> %b, iXLen %vl, iXLen 0)
+  ret <vscale x 4 x i32> %2
+}
+
+define <vscale x 4 x i32> @vmacc_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmacc_vx:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vmv2r.v v10, v8
+; NOVLOPT-NEXT:    vsetvli a2, zero, e32, m2, tu, ma
+; NOVLOPT-NEXT:    vmacc.vx v10, a0, v8
+; NOVLOPT-NEXT:    vsetvli zero, a1, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vmul.vv v8, v10, v8
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vmacc_vx:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vmv2r.v v10, v8
+; VLOPT-NEXT:    vsetvli zero, a1, e32, m2, tu, ma
+; VLOPT-NEXT:    vmacc.vx v10, a0, v8
+; VLOPT-NEXT:    vsetvli zero, zero, e32, m2, ta, ma
+; VLOPT-NEXT:    vmul.vv v8, v10, v8
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i32> @llvm.riscv.vmacc.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, i32 %b, <vscale x 4 x i32> %a, iXLen -1, iXLen 0)
----------------
topperc wrote:

oh you're right. I forget that we keep leaving off the VL operand type. There are really 3 overloaded types.

https://github.com/llvm/llvm-project/pull/112216