Add zext, sext and trunc

Author: davemgreen

codesize.html

@@ -9,7 +9,7 @@
                 .then(response => {
                     if (!response.ok) {
                         fetch("data-" + label + ".json")
-                            .then(reposonse => reposonse.json())
+                            .then(response => response.json())
                             .then(json => drawTable(json, label))
                             .catch(reason => console.log(reason));
                         return Promise.reject()
@@ -23,7 +23,8 @@
         function reload(type) {
             load("int", type)
             load("fp", type)
-            load("cast", type)
+            load("castint", type)
+            load("castfp", type)
             load("vec", type)
         }
 
@@ -43,8 +44,7 @@
             });
 
             const table = document.getElementById('table-'+label)
-            while (table.childNodes.length != 0)
-                table.removeChild(table.childNodes[0]);
+            table.innerHTML = ""
 
             head = table.createTHead();
             tr = head.insertRow(0);
@@ -157,10 +157,15 @@ <h3>FP diffs</h3>
         <div width="60%" style="flex: 0 0 60%;"><table id="table-fp" class="dataframe" border="1"><th>Loading</th></table></div>
         <div width="40%" style="position: sticky; top: 10px; align-self: flex-start; white-space: pre; padding: 10px;" id="details-fp"><h3>Details</h3></div>
     </div>
-    <h3>Cast diffs</h3>
+    <h3>Integer-Cast diffs</h3>
     <div style="display: flex;">
-        <div width="60%" style="flex: 0 0 60%;"><table id="table-cast" class="dataframe" border="1"><th>Loading</th></table></div>
-        <div width="40%" style="position: sticky; top: 10px; align-self: flex-start; white-space: pre; padding: 10px;" id="details-cast"><h3>Details</h3></div>
+        <div width="60%" style="flex: 0 0 60%;"><table id="table-castint" class="dataframe" border="1"><th>Loading</th></table></div>
+        <div width="40%" style="position: sticky; top: 10px; align-self: flex-start; white-space: pre; padding: 10px;" id="details-castint"><h3>Details</h3></div>
+    </div>
+    <h3>FP-Cast diffs</h3>
+    <div style="display: flex;">
+        <div width="60%" style="flex: 0 0 60%;"><table id="table-castfp" class="dataframe" border="1"><th>Loading</th></table></div>
+        <div width="40%" style="position: sticky; top: 10px; align-self: flex-start; white-space: pre; padding: 10px;" id="details-castfp"><h3>Details</h3></div>
     </div>
     <h3>Vec diffs</h3>
     <div style="display: flex;">

data-cast-fullfp16.json renamed to data-castfp-fullfp16.json

@@ -7528,10 +7528,10 @@
   "lat": 4,
   "sizelat": 4,
   "size": 2,
-  "gisize": -1,
+  "gisize": 22,
   "extrasize": 0,
   "asm": "index z1.d, #7, #-1\nptrue p0.d, vl2\nudiv z0.d, p0/m, z0.d, z1.d\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %178:_(s64), %179:_(s64), %180:_(s64), %181:_(s64) = G_UNMERGE_VALUES %103:_(<2 x s128>) (in function: test)",
+  "giasm": "mov x8, #9363                       // =0x2493\nfmov x9, d0\nmov x10, #-6148914691236517206      // =0xaaaaaaaaaaaaaaaa\nmovk x8, #37449, lsl #16\nmovk x10, #43691\nmovk x8, #18724, lsl #32\nmovk x8, #9362, lsl #48\numulh x8, x9, x8\nmov x9, v0.d[1]\numulh x9, x9, x10\nmov v1.d[0], x8\nmov x8, #-9223372036854775808       // =0x8000000000000000\nmov v1.d[1], x9\nsub v0.2d, v0.2d, v1.2d\nfmov x9, d0\nmov x10, v0.d[1]\numulh x8, x9, x8\numulh x9, x10, xzr\nmov v0.d[0], x8\nmov v0.d[1], x9\nadd v0.2d, v0.2d, v1.2d\nushr v0.2d, v0.2d, #2\nret",
   "ll": "define <2 x i64> @test(<2 x i64> %a) {\n  %r = udiv <2 x i64> %a, <i64 7, i64 6>\n  ret <2 x i64> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 4 for instruction:   %r = udiv <2 x i64> %a, <i64 7, i64 6>\nCost Model: Found an estimated cost of 1 for instruction:   ret <2 x i64> %r"
  },
@@ -7560,10 +7560,10 @@
   "lat": 4,
   "sizelat": 4,
   "size": 3,
-  "gisize": -1,
+  "gisize": 13,
   "extrasize": 0,
   "asm": "mov w8, #7                          // =0x7\nptrue p0.d, vl2\ndup v1.2d, x8\nudiv z0.d, p0/m, z0.d, z1.d\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %110:_(s64), %111:_(s64), %112:_(s64), %113:_(s64) = G_UNMERGE_VALUES %33:_(<2 x s128>) (in function: test)",
+  "giasm": "mov x8, #9363                       // =0x2493\nfmov x9, d0\nmov x10, v0.d[1]\nmovk x8, #37449, lsl #16\nmovk x8, #18724, lsl #32\nmovk x8, #9362, lsl #48\numulh x9, x9, x8\numulh x8, x10, x8\nmov v1.d[0], x9\nmov v1.d[1], x8\nsub v0.2d, v0.2d, v1.2d\nusra v1.2d, v0.2d, #1\nushr v0.2d, v1.2d, #2\nret",
   "ll": "define <2 x i64> @test(<2 x i64> %a) {\n  %r = udiv <2 x i64> %a, splat (i64 7)\n  ret <2 x i64> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 4 for instruction:   %r = udiv <2 x i64> %a, splat (i64 7)\nCost Model: Found an estimated cost of 1 for instruction:   ret <2 x i64> %r"
  },
@@ -7592,10 +7592,10 @@
   "lat": 4,
   "sizelat": 4,
   "size": 3,
-  "gisize": -1,
+  "gisize": 37,
   "extrasize": 0,
   "asm": "index z2.d, #7, #-1\nptrue p0.d, vl2\nudiv z0.d, p0/m, z0.d, z2.d\nudiv z1.d, p0/m, z1.d, z2.d\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %0:_(<4 x s64>) = G_CONCAT_VECTORS %1:_(<2 x s64>), %2:_(<2 x s64>) (in function: test)",
+  "giasm": "mov x8, #9363                       // =0x2493\nfmov x9, d0\nfmov x11, d1\nmovk x8, #37449, lsl #16\nmov x10, v0.d[1]\nmov x12, v1.d[1]\nmovk x8, #18724, lsl #32\nmovk x8, #9362, lsl #48\numulh x9, x9, x8\numulh x8, x11, x8\nmov x11, #-6148914691236517206      // =0xaaaaaaaaaaaaaaaa\nmovk x11, #43691\numulh x10, x10, x11\nmov v2.d[0], x9\numulh x9, x12, x11\nmov v3.d[0], x8\nmov x8, #-9223372036854775808       // =0x8000000000000000\nmov v2.d[1], x10\nmov v3.d[1], x9\nsub v0.2d, v0.2d, v2.2d\nsub v1.2d, v1.2d, v3.2d\nfmov x9, d0\nmov x10, v0.d[1]\nfmov x11, d1\nmov x12, v1.d[1]\numulh x9, x9, x8\numulh x8, x11, x8\numulh x10, x10, xzr\nmov v0.d[0], x9\numulh x11, x12, xzr\nmov v1.d[0], x8\nmov v0.d[1], x10\nmov v1.d[1], x11\nadd v0.2d, v0.2d, v2.2d\nadd v1.2d, v1.2d, v3.2d\nushr v0.2d, v0.2d, #2\nushr v1.2d, v1.2d, #2\nret",
   "ll": "define <4 x i64> @test(<4 x i64> %a) {\n  %r = udiv <4 x i64> %a, <i64 7, i64 6, i64 7, i64 6>\n  ret <4 x i64> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 4 for instruction:   %r = udiv <4 x i64> %a, <i64 7, i64 6, i64 7, i64 6>\nCost Model: Found an estimated cost of 1 for instruction:   ret <4 x i64> %r"
  },
@@ -7624,10 +7624,10 @@
   "lat": 4,
   "sizelat": 4,
   "size": 4,
-  "gisize": -1,
+  "gisize": 22,
   "extrasize": 0,
   "asm": "mov w8, #7                          // =0x7\nptrue p0.d, vl2\ndup v2.2d, x8\nudiv z0.d, p0/m, z0.d, z2.d\nudiv z1.d, p0/m, z1.d, z2.d\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %0:_(<4 x s64>) = G_CONCAT_VECTORS %1:_(<2 x s64>), %2:_(<2 x s64>) (in function: test)",
+  "giasm": "mov x8, #9363                       // =0x2493\nfmov x9, d0\nfmov x11, d1\nmovk x8, #37449, lsl #16\nmov x10, v0.d[1]\nmov x12, v1.d[1]\nmovk x8, #18724, lsl #32\nmovk x8, #9362, lsl #48\numulh x9, x9, x8\numulh x11, x11, x8\numulh x10, x10, x8\nmov v2.d[0], x9\numulh x8, x12, x8\nmov v3.d[0], x11\nmov v2.d[1], x10\nmov v3.d[1], x8\nsub v0.2d, v0.2d, v2.2d\nsub v1.2d, v1.2d, v3.2d\nusra v2.2d, v0.2d, #1\nusra v3.2d, v1.2d, #1\nushr v0.2d, v2.2d, #2\nushr v1.2d, v3.2d, #2\nret",
   "ll": "define <4 x i64> @test(<4 x i64> %a) {\n  %r = udiv <4 x i64> %a, splat (i64 7)\n  ret <4 x i64> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 4 for instruction:   %r = udiv <4 x i64> %a, splat (i64 7)\nCost Model: Found an estimated cost of 1 for instruction:   ret <4 x i64> %r"
  },
@@ -32936,10 +32936,10 @@
   "lat": 160,
   "sizelat": 160,
   "size": 7,
-  "gisize": -1,
+  "gisize": 7,
   "extrasize": 0,
   "asm": "movi v2.16b, #1\nsub v3.16b, v0.16b, v2.16b\nsub v2.16b, v1.16b, v2.16b\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\ncnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<32 x s8>) = G_CTPOP %10:_(<32 x s8>) (in function: test)",
+  "giasm": "movi v2.2d, #0xffffffffffffffff\nadd v3.16b, v0.16b, v2.16b\nadd v2.16b, v1.16b, v2.16b\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\ncnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nret",
   "ll": "define <32 x i8> @test(<32 x i8> %a) {\n  %r = call <32 x i8> @llvm.cttz(<32 x i8> %a, i1 0)\n  ret <32 x i8> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 160 for instruction:   %r = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)\nCost Model: Found an estimated cost of 1 for instruction:   ret <32 x i8> %r"
  },
@@ -32984,10 +32984,10 @@
   "lat": 80,
   "sizelat": 80,
   "size": 10,
-  "gisize": -1,
+  "gisize": 9,
   "extrasize": 0,
   "asm": "movi v2.8h, #1\nsub v3.8h, v0.8h, v2.8h\nsub v2.8h, v1.8h, v2.8h\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\nmovi v2.8h, #16\nclz v0.8h, v0.8h\nclz v1.8h, v1.8h\nsub v0.8h, v2.8h, v0.8h\nsub v1.8h, v2.8h, v1.8h\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<16 x s16>) = G_CTPOP %10:_(<16 x s16>) (in function: test)",
+  "giasm": "movi v2.2d, #0xffffffffffffffff\nadd v3.8h, v0.8h, v2.8h\nadd v2.8h, v1.8h, v2.8h\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\ncnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nret",
   "ll": "define <16 x i16> @test(<16 x i16> %a) {\n  %r = call <16 x i16> @llvm.cttz(<16 x i16> %a, i1 0)\n  ret <16 x i16> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 80 for instruction:   %r = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)\nCost Model: Found an estimated cost of 1 for instruction:   ret <16 x i16> %r"
  },
@@ -33032,10 +33032,10 @@
   "lat": 40,
   "sizelat": 40,
   "size": 10,
-  "gisize": -1,
+  "gisize": 11,
   "extrasize": 0,
   "asm": "movi v2.4s, #1\nsub v3.4s, v0.4s, v2.4s\nsub v2.4s, v1.4s, v2.4s\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\nmovi v2.4s, #32\nclz v0.4s, v0.4s\nclz v1.4s, v1.4s\nsub v0.4s, v2.4s, v0.4s\nsub v1.4s, v2.4s, v1.4s\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<8 x s32>) = G_CTPOP %10:_(<8 x s32>) (in function: test)",
+  "giasm": "movi v2.2d, #0xffffffffffffffff\nadd v3.4s, v0.4s, v2.4s\nadd v2.4s, v1.4s, v2.4s\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\ncnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nuaddlp v0.4s, v0.8h\nuaddlp v1.4s, v1.8h\nret",
   "ll": "define <8 x i32> @test(<8 x i32> %a) {\n  %r = call <8 x i32> @llvm.cttz(<8 x i32> %a, i1 0)\n  ret <8 x i32> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 40 for instruction:   %r = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)\nCost Model: Found an estimated cost of 1 for instruction:   ret <8 x i32> %r"
  },
@@ -33064,10 +33064,10 @@
   "lat": 20,
   "sizelat": 20,
   "size": 14,
-  "gisize": -1,
+  "gisize": 13,
   "extrasize": 0,
   "asm": "mov w8, #1                          // =0x1\ndup v2.2d, x8\nsub v3.2d, v0.2d, v2.2d\nsub v2.2d, v1.2d, v2.2d\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\ncnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nuaddlp v0.4s, v0.8h\nuaddlp v1.4s, v1.8h\nuaddlp v0.2d, v0.4s\nuaddlp v1.2d, v1.4s\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<4 x s64>) = G_CTPOP %10:_(<4 x s64>) (in function: test)",
+  "giasm": "movi v2.2d, #0xffffffffffffffff\nadd v3.2d, v0.2d, v2.2d\nadd v2.2d, v1.2d, v2.2d\nbic v0.16b, v3.16b, v0.16b\nbic v1.16b, v2.16b, v1.16b\ncnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nuaddlp v0.4s, v0.8h\nuaddlp v1.4s, v1.8h\nuaddlp v0.2d, v0.4s\nuaddlp v1.2d, v1.4s\nret",
   "ll": "define <4 x i64> @test(<4 x i64> %a) {\n  %r = call <4 x i64> @llvm.cttz(<4 x i64> %a, i1 0)\n  ret <4 x i64> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 20 for instruction:   %r = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)\nCost Model: Found an estimated cost of 1 for instruction:   ret <4 x i64> %r"
  },
@@ -33352,10 +33352,10 @@
   "lat": 2,
   "sizelat": 2,
   "size": 2,
-  "gisize": -1,
+  "gisize": 2,
   "extrasize": 0,
   "asm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<32 x s8>) = G_CTPOP %0:_(<32 x s8>) (in function: test)",
+  "giasm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nret",
   "ll": "define <32 x i8> @test(<32 x i8> %a) {\n  %r = call <32 x i8> @llvm.ctpop(<32 x i8> %a)\n  ret <32 x i8> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 2 for instruction:   %r = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a)\nCost Model: Found an estimated cost of 1 for instruction:   ret <32 x i8> %r"
  },
@@ -33400,10 +33400,10 @@
   "lat": 4,
   "sizelat": 4,
   "size": 4,
-  "gisize": -1,
+  "gisize": 4,
   "extrasize": 0,
   "asm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<16 x s16>) = G_CTPOP %0:_(<16 x s16>) (in function: test)",
+  "giasm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nret",
   "ll": "define <16 x i16> @test(<16 x i16> %a) {\n  %r = call <16 x i16> @llvm.ctpop(<16 x i16> %a)\n  ret <16 x i16> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 4 for instruction:   %r = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a)\nCost Model: Found an estimated cost of 1 for instruction:   ret <16 x i16> %r"
  },
@@ -33448,10 +33448,10 @@
   "lat": 6,
   "sizelat": 6,
   "size": 6,
-  "gisize": -1,
+  "gisize": 6,
   "extrasize": 0,
   "asm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nuaddlp v0.4s, v0.8h\nuaddlp v1.4s, v1.8h\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<8 x s32>) = G_CTPOP %0:_(<8 x s32>) (in function: test)",
+  "giasm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nuaddlp v0.4s, v0.8h\nuaddlp v1.4s, v1.8h\nret",
   "ll": "define <8 x i32> @test(<8 x i32> %a) {\n  %r = call <8 x i32> @llvm.ctpop(<8 x i32> %a)\n  ret <8 x i32> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 6 for instruction:   %r = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a)\nCost Model: Found an estimated cost of 1 for instruction:   ret <8 x i32> %r"
  },
@@ -33480,10 +33480,10 @@
   "lat": 8,
   "sizelat": 8,
   "size": 8,
-  "gisize": -1,
+  "gisize": 8,
   "extrasize": 0,
   "asm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nuaddlp v0.4s, v0.8h\nuaddlp v1.4s, v1.8h\nuaddlp v0.2d, v0.4s\nuaddlp v1.2d, v1.4s\nret",
-  "giasm": "LLVM ERROR: unable to legalize instruction: %3:_(<4 x s64>) = G_CTPOP %0:_(<4 x s64>) (in function: test)",
+  "giasm": "cnt v0.16b, v0.16b\ncnt v1.16b, v1.16b\nuaddlp v0.8h, v0.16b\nuaddlp v1.8h, v1.16b\nuaddlp v0.4s, v0.8h\nuaddlp v1.4s, v1.8h\nuaddlp v0.2d, v0.4s\nuaddlp v1.2d, v1.4s\nret",
   "ll": "define <4 x i64> @test(<4 x i64> %a) {\n  %r = call <4 x i64> @llvm.ctpop(<4 x i64> %a)\n  ret <4 x i64> %r\n}",
   "costoutput": "Printing analysis 'Cost Model Analysis' for function 'test':\nCost Model: Found an estimated cost of 8 for instruction:   %r = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a)\nCost Model: Found an estimated cost of 1 for instruction:   ret <4 x i64> %r"
  },