rvv: add vfslide1[down|up].vf and refine checking rule

new features in spec 0.9

Signed-off-by: Chih-Min Chao <chihmin.chao@sifive.com>

11 files changed, 79 insertions, 27 deletions

diff --git a/riscv/decode.h b/riscv/decode.h
index feaf2d4..1e03b59 100644
--- a/riscv/decode.h
+++ b/riscv/decode.h
@@ -506,6 +506,15 @@ static inline bool is_overlapped(const int astart, const int asize,
   require((insn.rs2() & (P.VU.vlmul - 1)) == 0); \
   require(P.VU.vstart == 0); \
 
+#define VI_CHECK_SLIDE(is_over) \
+  require((insn.rs2() & (P.VU.vlmul - 1)) == 0); \
+  require((insn.rd() & (P.VU.vlmul - 1)) == 0); \
+  if (insn.v_vm() == 0 && P.VU.vlmul > 1) \
+    require(insn.rd() != 0); \
+  if (is_over) \
+    require(insn.rd() != insn.rs2()); \
+
+
 //
 // vector: loop header and end helper
 //
diff --git a/riscv/encoding.h b/riscv/encoding.h
index c052c6f..584bc27 100644
--- a/riscv/encoding.h
+++ b/riscv/encoding.h
@@ -932,6 +932,10 @@
 #define MASK_VFSGNJN_VF  0xfc00707f
 #define MATCH_VFSGNJX_VF 0x28005057
 #define MASK_VFSGNJX_VF  0xfc00707f
+#define MATCH_VFSLIDE1UP_VF 0x38005057
+#define MASK_VFSLIDE1UP_VF  0xfc00707f
+#define MATCH_VFSLIDE1DOWN_VF 0x3c005057
+#define MASK_VFSLIDE1DOWN_VF  0xfc00707f
 #define MATCH_VFMV_S_F 0x42005057
 #define MASK_VFMV_S_F  0xfff0707f
 #define MATCH_VFMERGE_VFM 0x5c005057
@@ -2209,6 +2213,8 @@ DECLARE_INSN(vfmax_vf, MATCH_VFMAX_VF, MASK_VFMAX_VF)
 DECLARE_INSN(vfsgnj_vf, MATCH_VFSGNJ_VF, MASK_VFSGNJ_VF)
 DECLARE_INSN(vfsgnjn_vf, MATCH_VFSGNJN_VF, MASK_VFSGNJN_VF)
 DECLARE_INSN(vfsgnjx_vf, MATCH_VFSGNJX_VF, MASK_VFSGNJX_VF)
+DECLARE_INSN(vfslide1up_vf, MATCH_VFSLIDE1UP_VF, MASK_VFSLIDE1UP_VF)
+DECLARE_INSN(vfslide1down_vf, MATCH_VFSLIDE1DOWN_VF, MASK_VFSLIDE1DOWN_VF)
 DECLARE_INSN(vfmv_s_f, MATCH_VFMV_S_F, MASK_VFMV_S_F)
 DECLARE_INSN(vfmerge_vfm, MATCH_VFMERGE_VFM, MASK_VFMERGE_VFM)
 DECLARE_INSN(vfmv_v_f, MATCH_VFMV_V_F, MASK_VFMV_V_F)
diff --git a/riscv/insns/vfslide1down_vf.h b/riscv/insns/vfslide1down_vf.h
new file mode 100644
index 0000000..b6c1735
--- /dev/null
+++ b/riscv/insns/vfslide1down_vf.h
@@ -0,0 +1,28 @@
+//vfslide1down.vf vd, vs2, rs1
+VI_CHECK_SLIDE(false);
+
+VI_VFP_LOOP_BASE
+if (i != vl - 1) {
+  switch (P.VU.vsew) {
+    case e32: {
+      VI_XI_SLIDEDOWN_PARAMS(e32, 1);
+      vd = vs2;
+    }
+    break;
+    case e64: {
+      VI_XI_SLIDEDOWN_PARAMS(e64, 1);
+      vd = vs2;
+    }
+    break;
+  }
+} else {
+  switch (P.VU.vsew) {
+    case e32:
+      P.VU.elt<float32_t>(rd_num, vl - 1) = f32(FRS1);
+      break;
+    case e64:
+      P.VU.elt<float64_t>(rd_num, vl - 1) = f64(FRS1);
+      break;
+  }
+}
+VI_VFP_LOOP_END
diff --git a/riscv/insns/vfslide1up_vf.h b/riscv/insns/vfslide1up_vf.h
new file mode 100644
index 0000000..40b26e1
--- /dev/null
+++ b/riscv/insns/vfslide1up_vf.h
@@ -0,0 +1,28 @@
+//vfslide1up.vf vd, vs2, rs1
+VI_CHECK_SLIDE(true);
+
+VI_VFP_LOOP_BASE
+if (i != 0) {
+  switch (P.VU.vsew) {
+    case e32: {
+      VI_XI_SLIDEUP_PARAMS(e32, 1);
+      vd = vs2;
+    }
+    break;
+    case e64: {
+      VI_XI_SLIDEUP_PARAMS(e64, 1);
+      vd = vs2;
+    }
+    break;
+  }
+} else {
+  switch (P.VU.vsew) {
+    case e32:
+      P.VU.elt<float32_t>(rd_num, 0) = f32(FRS1);
+      break;
+    case e64:
+      P.VU.elt<float64_t>(rd_num, 0) = f64(FRS1);
+      break;
+  }
+}
+VI_VFP_LOOP_END
diff --git a/riscv/insns/vslide1down_vx.h b/riscv/insns/vslide1down_vx.h
index 04e2540..1fe9d04 100644
--- a/riscv/insns/vslide1down_vx.h
+++ b/riscv/insns/vslide1down_vx.h
@@ -1,8 +1,5 @@
 //vslide1down.vx vd, vs2, rs1
-require((insn.rs2() & (P.VU.vlmul - 1)) == 0);
-require((insn.rd() & (P.VU.vlmul - 1)) == 0);
-if (P.VU.vlmul > 1 && insn.v_vm() == 0)
-  require(insn.rd() != 0);
+VI_CHECK_SLIDE(false);
 
 VI_LOOP_BASE
 if (i != vl - 1) {
diff --git a/riscv/insns/vslide1up_vx.h b/riscv/insns/vslide1up_vx.h
index 5154259..6f12528 100644
--- a/riscv/insns/vslide1up_vx.h
+++ b/riscv/insns/vslide1up_vx.h
@@ -1,9 +1,5 @@
 //vslide1up.vx vd, vs2, rs1
-require((insn.rs2() & (P.VU.vlmul - 1)) == 0);
-require((insn.rd() & (P.VU.vlmul - 1)) == 0);
-require(insn.rd() != insn.rs2());
-if (insn.v_vm() == 0 && P.VU.vlmul > 1)
-  require(insn.rd() != 0);
+VI_CHECK_SLIDE(true);
 
 VI_LOOP_BASE
 if (i != 0) {
diff --git a/riscv/insns/vslidedown_vi.h b/riscv/insns/vslidedown_vi.h
index dd58c1e..bc440cf 100644
--- a/riscv/insns/vslidedown_vi.h
+++ b/riscv/insns/vslidedown_vi.h
@@ -1,8 +1,5 @@
 // vslidedown.vi vd, vs2, rs1
-require((insn.rs2() & (P.VU.vlmul - 1)) == 0);
-require((insn.rd() & (P.VU.vlmul - 1)) == 0);
-if (P.VU.vlmul > 1 && insn.v_vm() == 0)
-  require(insn.rd() != 0);
+VI_CHECK_SLIDE(false);
 
 const reg_t sh = insn.v_zimm5();
 VI_LOOP_BASE
diff --git a/riscv/insns/vslidedown_vx.h b/riscv/insns/vslidedown_vx.h
index 744a7a5..074aa50 100644
--- a/riscv/insns/vslidedown_vx.h
+++ b/riscv/insns/vslidedown_vx.h
@@ -1,8 +1,5 @@
 //vslidedown.vx vd, vs2, rs1
-require((insn.rs2() & (P.VU.vlmul - 1)) == 0);
-require((insn.rd() & (P.VU.vlmul - 1)) == 0);
-if (P.VU.vlmul > 1 && insn.v_vm() == 0)
-  require(insn.rd() != 0);
+VI_CHECK_SLIDE(false);
 
 const uint128_t sh = RS1;
 VI_LOOP_BASE
diff --git a/riscv/insns/vslideup_vi.h b/riscv/insns/vslideup_vi.h
index 99d30bc..4265789 100644
--- a/riscv/insns/vslideup_vi.h
+++ b/riscv/insns/vslideup_vi.h
@@ -1,9 +1,5 @@
 // vslideup.vi vd, vs2, rs1
-require((insn.rs2() & (P.VU.vlmul - 1)) == 0);
-require((insn.rd() & (P.VU.vlmul - 1)) == 0);
-require(insn.rd() != insn.rs2());
-if (insn.v_vm() == 0 && P.VU.vlmul > 1)
-  require(insn.rd() != 0);
+VI_CHECK_SLIDE(true);
 
 const reg_t offset = insn.v_zimm5();
 VI_LOOP_BASE
diff --git a/riscv/insns/vslideup_vx.h b/riscv/insns/vslideup_vx.h
index 2d68a39..720d2ab 100644
--- a/riscv/insns/vslideup_vx.h
+++ b/riscv/insns/vslideup_vx.h
@@ -1,9 +1,5 @@
 //vslideup.vx vd, vs2, rs1
-require((insn.rs2() & (P.VU.vlmul - 1)) == 0);
-require((insn.rd() & (P.VU.vlmul - 1)) == 0);
-require(insn.rd() != insn.rs2());
-if (insn.v_vm() == 0 && P.VU.vlmul > 1)
-  require(insn.rd() != 0);
+VI_CHECK_SLIDE(true);
 
 const reg_t offset = RS1;
 VI_LOOP_BASE
diff --git a/riscv/riscv.mk.in b/riscv/riscv.mk.in
index 757c734..bbcba5e 100644
--- a/riscv/riscv.mk.in
+++ b/riscv/riscv.mk.in
@@ -565,6 +565,8 @@ riscv_insn_ext_v_alu_fp = \
 	vfsgnjx_vf \
 	vfsgnjx_vv \
 	vfsqrt_v \
+	vfslide1down_vf \
+	vfslide1up_vf \
 	vfsub_vf \
 	vfsub_vv \
 	vfwadd_vf \