Add support for some packed multiplication for new x64 backend

Adds support for i32x4, and i16x8 and lowering for pmuludq in preperation for i64x2.
2020-08-17 13:44:10 -07:00
parent 81b3450114
commit a31336996c
4 changed files with 80 additions and 44 deletions
--- a/cranelift/codegen/src/isa/x64/inst/emit_tests.rs
+++ b/cranelift/codegen/src/isa/x64/inst/emit_tests.rs
@@ -3062,6 +3062,24 @@ fn test_x64_emit() {
        "psubq   %xmm8, %xmm1",
    ));

+    insns.push((
+        Inst::xmm_rm_r(SseOpcode::Pmulld, RegMem::reg(xmm15), w_xmm6),
+        "66410F3840F7",
+        "pmulld  %xmm15, %xmm6",
+    ));
+
+    insns.push((
+        Inst::xmm_rm_r(SseOpcode::Pmullw, RegMem::reg(xmm14), w_xmm1),
+        "66410FD5CE",
+        "pmullw  %xmm14, %xmm1",
+    ));
+
+    insns.push((
+        Inst::xmm_rm_r(SseOpcode::Pmuludq, RegMem::reg(xmm8), w_xmm9),
+        "66450FF4C8",
+        "pmuludq %xmm8, %xmm9",
+    ));
+
    // XMM_Mov_R_M: float stores
    insns.push((
        Inst::xmm_mov_r_m(SseOpcode::Movss, xmm15, Amode::imm_reg(128, r12), None),