Revert "rvv modify the vfredsum.vs behavior with e27 xlen=32"

This reverts commit c12f7f6510398d3d259cf52f810137504ff96f7a.

2 files changed, 9 insertions, 54 deletions

diff --git a/riscv/decode.h b/riscv/decode.h
index 4a72b06..64bb404 100644
--- a/riscv/decode.h
+++ b/riscv/decode.h
@@ -1662,6 +1662,7 @@ for (reg_t i = 0; i < vlmax && P.VU.vl != 0; ++i) { \
   vd_0 = vs1_0;\
   for (reg_t i=P.VU.vstart; i<vl; ++i){ \
     VI_LOOP_ELEMENT_SKIP(); \
+    int##width##_t &vd = P.VU.elt<int##width##_t>(rd_num, i, true); \
     float##width##_t vs2 = P.VU.elt<float##width##_t>(rs2_num, i); \
 
 #define VI_VFP_LOOP_WIDE_REDUCTION_BASE \
diff --git a/riscv/insns/vfredsum_vs.h b/riscv/insns/vfredsum_vs.h
index b354bdd..7b5cccc 100644
--- a/riscv/insns/vfredsum_vs.h
+++ b/riscv/insns/vfredsum_vs.h
@@ -1,54 +1,8 @@
-// vfredsum.vs  vd, vs2, vs1, vm # Unordered sum
-{	VI_CHECK_REDUCTION(false)
-	VI_VFP_COMMON
-    reg_t tmp_vl = ((vl>64)?128:(vl>32)?64:(vl>16)?32:(vl>8)?16:(vl>4)?8:(vl>2)?4:(vl>1)?2:1);
-    float32_t tmpValue[tmp_vl];
-    float32_t tmpZero;
-	tmpZero.v = 0;
-    reg_t i= 0;
-    while(i<P.VU.vstart){
-        tmpValue[i] =  tmpZero;
-        ++i;
-    }
-    for (i=P.VU.vstart; i<tmp_vl; ++i){
-        const int mlen = P.VU.vmlen;
-        const int midx = (mlen * i) / 32;
-        const int mpos = (mlen * i) % 32;
-        bool skip = ((P.VU.elt<uint32_t>(0, midx) >> mpos) & 0x1) == 0;
-        if ((insn.v_vm() == 0 && skip)|| i>=vl)
-            tmpValue[i].v =  0;
-        else
-            tmpValue[i] = P.VU.elt<float32_t>(rs2_num, i);
-    }
-    while(tmp_vl >4){
-        for (reg_t i=0; i< tmp_vl/2;i=i+4){
-            tmpValue[i+0] = f32_add(tmpValue[(i*2)+0], tmpValue[(i*2)+0+4]);
-            set_fp_exceptions;
-            tmpValue[i+1] = f32_add(tmpValue[(i*2)+1], tmpValue[(i*2)+1+4]);
-            set_fp_exceptions;
-            tmpValue[i+2] = f32_add(tmpValue[(i*2)+2], tmpValue[(i*2)+2+4]);
-            set_fp_exceptions;
-            tmpValue[i+3] = f32_add(tmpValue[(i*2)+3], tmpValue[(i*2)+3+4]);
-            set_fp_exceptions;
-        }
-        tmp_vl /=2;
-    }
-    if(tmp_vl>2){
-        tmpValue[0] = f32_add(tmpValue[0], tmpValue[2]);
-        set_fp_exceptions;
-        tmpValue[1] = f32_add(tmpValue[1], tmpValue[3]);
-        set_fp_exceptions;
-        tmp_vl /=2;
-    }
-    if(tmp_vl>1){
-        tmpValue[0] = f32_add(tmpValue[0], tmpValue[1]);
-        set_fp_exceptions;
-    }
-    P.VU.vstart = 0;
-    if (vl > 0) {
-        float32_t vd_0 = P.VU.elt<float32_t>(rs1_num, 0);
-        vd_0 = f32_add(vd_0, tmpValue[0]);
-        set_fp_exceptions;
-        P.VU.elt<type_sew_t<e32>::type>(rd_num, 0, true) = vd_0.v;
-    }
-}
+// vfredsum: vd[0] =  sum( vs2[*] , vs1[0] )
+VI_VFP_VV_LOOP_REDUCTION
+({
+  vd_0 = f32_add(vd_0, vs2);
+},
+{
+  vd_0 = f64_add(vd_0, vs2);
+})