[X86][Atom] Fix FP uops + port usage

Both ports are required in most cases. Update the uops counts + port usage based off the most recent llvm-exegesis captures (PR36895) and what Intel AoM / Agner / InstLatX64 reports as well.

Noticed while trying to improve fp costs for vectorization via the D103695 helper script.

diff --git a/llvm/lib/Target/X86/X86ScheduleAtom.td b/llvm/lib/Target/X86/X86ScheduleAtom.td
index 1087fda..1708226 100644 (file)
--- a/llvm/lib/Target/X86/X86ScheduleAtom.td
+++ b/llvm/lib/Target/X86/X86ScheduleAtom.td
@@ -237,15 +237,15 @@ defm : AtomWriteResPair<WriteFAddX,          [AtomPort1], [AtomPort0,AtomPort1],
 defm : X86WriteResPairUnsupported<WriteFAddY>;
 defm : X86WriteResPairUnsupported<WriteFAddZ>;
 defm : AtomWriteResPair<WriteFAdd64,         [AtomPort1], [AtomPort0,AtomPort1],  5,  5,  [1],  [1,1]>;
-defm : AtomWriteResPair<WriteFAdd64X,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  6,  7,  [5,5],  [6,6]>;
+defm : AtomWriteResPair<WriteFAdd64X,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  6,  7,  [5,5],  [6,6], 3, 4>;
 defm : X86WriteResPairUnsupported<WriteFAdd64Y>;
 defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
 defm : AtomWriteResPair<WriteFCmp,           [AtomPort1], [AtomPort0,AtomPort1],  5,  5,  [1],  [1,1]>;
-defm : AtomWriteResPair<WriteFCmpX,          [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  6,  7,  [5,5],  [6,6]>;
+defm : AtomWriteResPair<WriteFCmpX,          [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  6,  7,  [5,5],  [6,6], 3, 4>;
 defm : X86WriteResPairUnsupported<WriteFCmpY>;
 defm : X86WriteResPairUnsupported<WriteFCmpZ>;
 defm : AtomWriteResPair<WriteFCmp64,         [AtomPort1], [AtomPort0,AtomPort1],  5,  5,  [1],  [1,1]>;
-defm : AtomWriteResPair<WriteFCmp64X,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  6,  7,  [5,5],  [6,6]>;
+defm : AtomWriteResPair<WriteFCmp64X,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  6,  7,  [5,5],  [6,6], 3, 4>;
 defm : X86WriteResPairUnsupported<WriteFCmp64Y>;
 defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
 defm : AtomWriteResPair<WriteFCom,           [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
@@ -255,34 +255,34 @@ defm : AtomWriteResPair<WriteFMulX,          [AtomPort0],  [AtomPort0],  5,  5,
 defm : X86WriteResPairUnsupported<WriteFMulY>;
 defm : X86WriteResPairUnsupported<WriteFMulZ>;
 defm : AtomWriteResPair<WriteFMul64,         [AtomPort0],  [AtomPort0],  5,  5,  [2],  [2]>;
-defm : AtomWriteResPair<WriteFMul64X,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  9, 10,  [9,9],[10,10]>;
+defm : AtomWriteResPair<WriteFMul64X,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  9, 10,  [9,9],[10,10], 6, 7>;
 defm : X86WriteResPairUnsupported<WriteFMul64Y>;
 defm : X86WriteResPairUnsupported<WriteFMul64Z>;
 defm : AtomWriteResPair<WriteFRcp,           [AtomPort0],  [AtomPort0],  4,  4,  [4],  [4]>;
-defm : AtomWriteResPair<WriteFRcpX,         [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
+defm : AtomWriteResPair<WriteFRcpX,          [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  9, 10,  [9,9], [10,10], 5, 6>;
 defm : X86WriteResPairUnsupported<WriteFRcpY>;
 defm : X86WriteResPairUnsupported<WriteFRcpZ>;
 defm : AtomWriteResPair<WriteFRsqrt,         [AtomPort0],  [AtomPort0],  4,  4,  [4],  [4]>;
-defm : AtomWriteResPair<WriteFRsqrtX,       [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
+defm : AtomWriteResPair<WriteFRsqrtX,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],  9, 10,  [9,9], [10,10], 5, 6>;
 defm : X86WriteResPairUnsupported<WriteFRsqrtY>;
 defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
-defm : AtomWriteResPair<WriteFDiv,          [AtomPort01], [AtomPort01], 34, 34, [34], [34]>;
-defm : AtomWriteResPair<WriteFDivX,         [AtomPort01], [AtomPort01], 70, 70, [70], [70]>;
+defm : AtomWriteResPair<WriteFDiv,          [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 34, 34, [34,34], [34,34], 3, 4>;
+defm : AtomWriteResPair<WriteFDivX,         [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 70, 70, [70,70], [70,70], 6, 7>;
 defm : X86WriteResPairUnsupported<WriteFDivY>;
 defm : X86WriteResPairUnsupported<WriteFDivZ>;
-defm : AtomWriteResPair<WriteFDiv64,        [AtomPort01], [AtomPort01], 62, 62, [62], [62]>;
-defm : AtomWriteResPair<WriteFDiv64X,       [AtomPort01], [AtomPort01],125,125,[125],[125]>;
+defm : AtomWriteResPair<WriteFDiv64,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 62, 62,  [62,62],  [62,62], 3, 4>;
+defm : AtomWriteResPair<WriteFDiv64X,       [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],125,125,[125,125],[125,125], 6, 7>;
 defm : X86WriteResPairUnsupported<WriteFDiv64Y>;
 defm : X86WriteResPairUnsupported<WriteFDiv64Z>;
-defm : AtomWriteResPair<WriteFSqrt,         [AtomPort01], [AtomPort01], 34, 34, [34], [34]>;
-defm : AtomWriteResPair<WriteFSqrtX,        [AtomPort01], [AtomPort01], 70, 70, [70], [70]>;
+defm : AtomWriteResPair<WriteFSqrt,         [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 34, 34, [34,34], [34,34], 3, 4>;
+defm : AtomWriteResPair<WriteFSqrtX,        [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 70, 70, [70,70], [70,70], 5, 6>;
 defm : X86WriteResPairUnsupported<WriteFSqrtY>;
 defm : X86WriteResPairUnsupported<WriteFSqrtZ>;
-defm : AtomWriteResPair<WriteFSqrt64,       [AtomPort01], [AtomPort01], 62, 62, [62], [62]>;
-defm : AtomWriteResPair<WriteFSqrt64X,      [AtomPort01], [AtomPort01],125,125,[125],[125]>;
+defm : AtomWriteResPair<WriteFSqrt64,       [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 62, 62,  [62,62],  [62,62], 3, 4>;
+defm : AtomWriteResPair<WriteFSqrt64X,      [AtomPort0,AtomPort1], [AtomPort0,AtomPort1],125,125,[125,125],[125,125], 5, 6>;
 defm : X86WriteResPairUnsupported<WriteFSqrt64Y>;
 defm : X86WriteResPairUnsupported<WriteFSqrt64Z>;
-defm : AtomWriteResPair<WriteFSqrt80,       [AtomPort01], [AtomPort01], 71, 71, [71], [71]>;
+defm : AtomWriteResPair<WriteFSqrt80,        [AtomPort0],  [AtomPort0], 71, 71, [71], [71]>;
 defm : AtomWriteResPair<WriteFSign,          [AtomPort1],  [AtomPort1]>;
 defm : AtomWriteResPair<WriteFRnd,           [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : X86WriteResPairUnsupported<WriteFRndY>;
@@ -478,11 +478,11 @@ defm : X86WriteResPairUnsupported<WriteAESDecEnc>;
 // Horizontal add/sub  instructions.
 ////////////////////////////////////////////////////////////////////////////////
 
-defm : AtomWriteResPair<WriteFHAdd,  [AtomPort01], [AtomPort01], 8, 9, [8], [9]>;
-defm : AtomWriteResPair<WriteFHAddY, [AtomPort01], [AtomPort01], 8, 9, [8], [9]>;
-defm : AtomWriteResPair<WritePHAdd,  [AtomPort01], [AtomPort01], 3, 4, [3], [4]>;
-defm : AtomWriteResPair<WritePHAddX, [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
-defm : AtomWriteResPair<WritePHAddY, [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
+defm : AtomWriteResPair<WriteFHAdd,  [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 8, 9, [8,8], [9,9], 5, 6>;
+defm : X86WriteResPairUnsupported<WriteFHAddY>;
+defm : AtomWriteResPair<WritePHAdd,  [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 3, 4, [3,3], [4,4], 3, 4>;
+defm : AtomWriteResPair<WritePHAddX, [AtomPort0,AtomPort1], [AtomPort0,AtomPort1], 7, 8, [7,7], [8,8], 3, 4>;
+defm : X86WriteResPairUnsupported<WritePHAddY>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Carry-less multiplication instructions.
@@ -494,8 +494,8 @@ defm : X86WriteResPairUnsupported<WriteCLMul>;
 // Load/store MXCSR.
 ////////////////////////////////////////////////////////////////////////////////
 
-def : WriteRes<WriteLDMXCSR, [AtomPort01]> { let Latency = 5; let ResourceCycles = [5]; }
-def : WriteRes<WriteSTMXCSR, [AtomPort01]> { let Latency = 15; let ResourceCycles = [15]; }
+defm : X86WriteRes<WriteLDMXCSR, [AtomPort0,AtomPort1],  5,   [5,5], 4>;
+defm : X86WriteRes<WriteSTMXCSR, [AtomPort0,AtomPort1], 15, [15,15], 4>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Special Cases.

diff --git a/llvm/test/tools/llvm-mca/X86/Atom/resources-sse1.s b/llvm/test/tools/llvm-mca/X86/Atom/resources-sse1.s
index a75c2bb..a59deae 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/Atom/resources-sse1.s
+++ b/llvm/test/tools/llvm-mca/X86/Atom/resources-sse1.s
@@ -202,8 +202,8 @@ xorps       (%rax), %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   andnps (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50                        andps  %xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   andps  (%rax), %xmm2
-# CHECK-NEXT:  1      6     5.00                        cmpeqps        %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   cmpeqps        (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        cmpeqps        %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   cmpeqps        (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        cmpeqss        %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   cmpeqss        (%rax), %xmm2
 # CHECK-NEXT:  4      9     9.00                        comiss %xmm0, %xmm1
@@ -226,18 +226,18 @@ xorps       (%rax), %xmm2
 # CHECK-NEXT:  1      9     4.50                        cvttss2si      %xmm0, %rcx
 # CHECK-NEXT:  1      9     6.00    *                   cvttss2si      (%rax), %ecx
 # CHECK-NEXT:  1      10    5.00    *                   cvttss2si      (%rax), %rcx
-# CHECK-NEXT:  1      70    35.00                       divps  %xmm0, %xmm2
-# CHECK-NEXT:  1      70    35.00   *                   divps  (%rax), %xmm2
-# CHECK-NEXT:  1      34    17.00                       divss  %xmm0, %xmm2
-# CHECK-NEXT:  1      34    17.00   *                   divss  (%rax), %xmm2
-# CHECK-NEXT:  1      5     2.50    *      *      U     ldmxcsr        (%rax)
+# CHECK-NEXT:  6      70    70.00                       divps  %xmm0, %xmm2
+# CHECK-NEXT:  7      70    70.00   *                   divps  (%rax), %xmm2
+# CHECK-NEXT:  3      34    34.00                       divss  %xmm0, %xmm2
+# CHECK-NEXT:  4      34    34.00   *                   divss  (%rax), %xmm2
+# CHECK-NEXT:  4      5     5.00    *      *      U     ldmxcsr        (%rax)
 # CHECK-NEXT:  1      1     1.00    *      *      U     maskmovq       %mm0, %mm1
-# CHECK-NEXT:  1      6     5.00                        maxps  %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   maxps  (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        maxps  %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   maxps  (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        maxss  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   maxss  (%rax), %xmm2
-# CHECK-NEXT:  1      6     5.00                        minps  %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   minps  (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        minps  %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   minps  (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        minss  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   minss  (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50                        movaps %xmm0, %xmm2
@@ -290,22 +290,22 @@ xorps       (%rax), %xmm2
 # CHECK-NEXT:  1      4     1.00    *                   psadbw (%rax), %mm2
 # CHECK-NEXT:  1      1     1.00                        pshufw $1, %mm0, %mm2
 # CHECK-NEXT:  1      1     1.00    *                   pshufw $1, (%rax), %mm2
-# CHECK-NEXT:  1      9     4.50                        rcpps  %xmm0, %xmm2
-# CHECK-NEXT:  1      10    5.00    *                   rcpps  (%rax), %xmm2
+# CHECK-NEXT:  5      9     9.00                        rcpps  %xmm0, %xmm2
+# CHECK-NEXT:  6      10    10.00   *                   rcpps  (%rax), %xmm2
 # CHECK-NEXT:  1      4     4.00                        rcpss  %xmm0, %xmm2
 # CHECK-NEXT:  1      4     4.00    *                   rcpss  (%rax), %xmm2
-# CHECK-NEXT:  1      9     4.50                        rsqrtps        %xmm0, %xmm2
-# CHECK-NEXT:  1      10    5.00    *                   rsqrtps        (%rax), %xmm2
+# CHECK-NEXT:  5      9     9.00                        rsqrtps        %xmm0, %xmm2
+# CHECK-NEXT:  6      10    10.00   *                   rsqrtps        (%rax), %xmm2
 # CHECK-NEXT:  1      4     4.00                        rsqrtss        %xmm0, %xmm2
 # CHECK-NEXT:  1      4     4.00    *                   rsqrtss        (%rax), %xmm2
 # CHECK-NEXT:  1      1     1.00    *      *      U     sfence
 # CHECK-NEXT:  1      1     1.00                        shufps $1, %xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   shufps $1, (%rax), %xmm2
-# CHECK-NEXT:  1      70    35.00                       sqrtps %xmm0, %xmm2
-# CHECK-NEXT:  1      70    35.00   *                   sqrtps (%rax), %xmm2
-# CHECK-NEXT:  1      34    17.00                       sqrtss %xmm0, %xmm2
-# CHECK-NEXT:  1      34    17.00   *                   sqrtss (%rax), %xmm2
-# CHECK-NEXT:  1      15    7.50           *      U     stmxcsr        (%rax)
+# CHECK-NEXT:  5      70    70.00                       sqrtps %xmm0, %xmm2
+# CHECK-NEXT:  6      70    70.00   *                   sqrtps (%rax), %xmm2
+# CHECK-NEXT:  3      34    34.00                       sqrtss %xmm0, %xmm2
+# CHECK-NEXT:  4      34    34.00   *                   sqrtss (%rax), %xmm2
+# CHECK-NEXT:  4      15    15.00          *      U     stmxcsr        (%rax)
 # CHECK-NEXT:  1      5     1.00                        subps  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   subps  (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        subss  %xmm0, %xmm2
@@ -325,7 +325,7 @@ xorps       (%rax), %xmm2
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]
-# CHECK-NEXT: 475.00 430.00
+# CHECK-NEXT: 712.00 667.00
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    Instructions:
@@ -361,11 +361,11 @@ xorps       (%rax), %xmm2
 # CHECK-NEXT: 4.50   4.50   cvttss2si  %xmm0, %rcx
 # CHECK-NEXT: 6.00   6.00   cvttss2si  (%rax), %ecx
 # CHECK-NEXT: 5.00   5.00   cvttss2si  (%rax), %rcx
-# CHECK-NEXT: 35.00  35.00  divps      %xmm0, %xmm2
-# CHECK-NEXT: 35.00  35.00  divps      (%rax), %xmm2
-# CHECK-NEXT: 17.00  17.00  divss      %xmm0, %xmm2
-# CHECK-NEXT: 17.00  17.00  divss      (%rax), %xmm2
-# CHECK-NEXT: 2.50   2.50   ldmxcsr    (%rax)
+# CHECK-NEXT: 70.00  70.00  divps      %xmm0, %xmm2
+# CHECK-NEXT: 70.00  70.00  divps      (%rax), %xmm2
+# CHECK-NEXT: 34.00  34.00  divss      %xmm0, %xmm2
+# CHECK-NEXT: 34.00  34.00  divss      (%rax), %xmm2
+# CHECK-NEXT: 5.00   5.00   ldmxcsr    (%rax)
 # CHECK-NEXT: 1.00    -     maskmovq   %mm0, %mm1
 # CHECK-NEXT: 5.00   5.00   maxps      %xmm0, %xmm2
 # CHECK-NEXT: 6.00   6.00   maxps      (%rax), %xmm2
@@ -425,22 +425,22 @@ xorps       (%rax), %xmm2
 # CHECK-NEXT: 1.00    -     psadbw     (%rax), %mm2
 # CHECK-NEXT: 1.00    -     pshufw     $1, %mm0, %mm2
 # CHECK-NEXT: 1.00    -     pshufw     $1, (%rax), %mm2
-# CHECK-NEXT: 4.50   4.50   rcpps      %xmm0, %xmm2
-# CHECK-NEXT: 5.00   5.00   rcpps      (%rax), %xmm2
+# CHECK-NEXT: 9.00   9.00   rcpps      %xmm0, %xmm2
+# CHECK-NEXT: 10.00  10.00  rcpps      (%rax), %xmm2
 # CHECK-NEXT: 4.00    -     rcpss      %xmm0, %xmm2
 # CHECK-NEXT: 4.00    -     rcpss      (%rax), %xmm2
-# CHECK-NEXT: 4.50   4.50   rsqrtps    %xmm0, %xmm2
-# CHECK-NEXT: 5.00   5.00   rsqrtps    (%rax), %xmm2
+# CHECK-NEXT: 9.00   9.00   rsqrtps    %xmm0, %xmm2
+# CHECK-NEXT: 10.00  10.00  rsqrtps    (%rax), %xmm2
 # CHECK-NEXT: 4.00    -     rsqrtss    %xmm0, %xmm2
 # CHECK-NEXT: 4.00    -     rsqrtss    (%rax), %xmm2
 # CHECK-NEXT: 1.00    -     sfence
 # CHECK-NEXT: 1.00    -     shufps     $1, %xmm0, %xmm2
 # CHECK-NEXT: 1.00    -     shufps     $1, (%rax), %xmm2
-# CHECK-NEXT: 35.00  35.00  sqrtps     %xmm0, %xmm2
-# CHECK-NEXT: 35.00  35.00  sqrtps     (%rax), %xmm2
-# CHECK-NEXT: 17.00  17.00  sqrtss     %xmm0, %xmm2
-# CHECK-NEXT: 17.00  17.00  sqrtss     (%rax), %xmm2
-# CHECK-NEXT: 7.50   7.50   stmxcsr    (%rax)
+# CHECK-NEXT: 70.00  70.00  sqrtps     %xmm0, %xmm2
+# CHECK-NEXT: 70.00  70.00  sqrtps     (%rax), %xmm2
+# CHECK-NEXT: 34.00  34.00  sqrtss     %xmm0, %xmm2
+# CHECK-NEXT: 34.00  34.00  sqrtss     (%rax), %xmm2
+# CHECK-NEXT: 15.00  15.00  stmxcsr    (%rax)
 # CHECK-NEXT:  -     1.00   subps      %xmm0, %xmm2
 # CHECK-NEXT: 1.00   1.00   subps      (%rax), %xmm2
 # CHECK-NEXT:  -     1.00   subss      %xmm0, %xmm2

diff --git a/llvm/test/tools/llvm-mca/X86/Atom/resources-sse2.s b/llvm/test/tools/llvm-mca/X86/Atom/resources-sse2.s
index 525ba77..36b4c75 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/Atom/resources-sse2.s
+++ b/llvm/test/tools/llvm-mca/X86/Atom/resources-sse2.s
@@ -407,8 +407,8 @@ xorpd       (%rax), %xmm2
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      6     5.00                        addpd  %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   addpd  (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        addpd  %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   addpd  (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        addsd  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   addsd  (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50                        andnpd %xmm0, %xmm2
@@ -416,8 +416,8 @@ xorpd       (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50                        andpd  %xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   andpd  (%rax), %xmm2
 # CHECK-NEXT:  1      1     1.00    *      *      U     clflush        (%rax)
-# CHECK-NEXT:  1      6     5.00                        cmpeqpd        %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   cmpeqpd        (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        cmpeqpd        %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   cmpeqpd        (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        cmpeqsd        %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   cmpeqsd        (%rax), %xmm2
 # CHECK-NEXT:  4      9     9.00                        comisd %xmm0, %xmm1
@@ -460,19 +460,19 @@ xorpd       (%rax), %xmm2
 # CHECK-NEXT:  1      8     7.00                        cvttsd2si      %xmm0, %rcx
 # CHECK-NEXT:  1      9     6.00    *                   cvttsd2si      (%rax), %ecx
 # CHECK-NEXT:  1      9     6.00    *                   cvttsd2si      (%rax), %rcx
-# CHECK-NEXT:  1      125   62.50                       divpd  %xmm0, %xmm2
-# CHECK-NEXT:  1      125   62.50   *                   divpd  (%rax), %xmm2
-# CHECK-NEXT:  1      62    31.00                       divsd  %xmm0, %xmm2
-# CHECK-NEXT:  1      62    31.00   *                   divsd  (%rax), %xmm2
+# CHECK-NEXT:  6      125   125.00                      divpd  %xmm0, %xmm2
+# CHECK-NEXT:  7      125   125.00  *                   divpd  (%rax), %xmm2
+# CHECK-NEXT:  3      62    62.00                       divsd  %xmm0, %xmm2
+# CHECK-NEXT:  4      62    62.00   *                   divsd  (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50    *      *      U     lfence
 # CHECK-NEXT:  1      2     1.00    *      *      U     maskmovdqu     %xmm0, %xmm1
-# CHECK-NEXT:  1      6     5.00                        maxpd  %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   maxpd  (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        maxpd  %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   maxpd  (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        maxsd  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   maxsd  (%rax), %xmm2
 # CHECK-NEXT:  1      1     1.00    *      *      U     mfence
-# CHECK-NEXT:  1      6     5.00                        minpd  %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   minpd  (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        minpd  %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   minpd  (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        minsd  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   minsd  (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50                        movapd %xmm0, %xmm2
@@ -510,8 +510,8 @@ xorpd       (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50                        movupd %xmm0, %xmm2
 # CHECK-NEXT:  1      2     1.00           *            movupd %xmm0, (%rax)
 # CHECK-NEXT:  1      3     1.50    *                   movupd (%rax), %xmm2
-# CHECK-NEXT:  1      9     9.00                        mulpd  %xmm0, %xmm2
-# CHECK-NEXT:  1      10    10.00   *                   mulpd  (%rax), %xmm2
+# CHECK-NEXT:  6      9     9.00                        mulpd  %xmm0, %xmm2
+# CHECK-NEXT:  7      10    10.00   *                   mulpd  (%rax), %xmm2
 # CHECK-NEXT:  1      5     2.00                        mulsd  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     2.00    *                   mulsd  (%rax), %xmm2
 # CHECK-NEXT:  1      1     0.50                        orpd   %xmm0, %xmm2
@@ -658,12 +658,12 @@ xorpd       (%rax), %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   pxor   (%rax), %xmm2
 # CHECK-NEXT:  1      1     1.00                        shufpd $1, %xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   shufpd $1, (%rax), %xmm2
-# CHECK-NEXT:  1      125   62.50                       sqrtpd %xmm0, %xmm2
-# CHECK-NEXT:  1      125   62.50   *                   sqrtpd (%rax), %xmm2
-# CHECK-NEXT:  1      62    31.00                       sqrtsd %xmm0, %xmm2
-# CHECK-NEXT:  1      62    31.00   *                   sqrtsd (%rax), %xmm2
-# CHECK-NEXT:  1      6     5.00                        subpd  %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   subpd  (%rax), %xmm2
+# CHECK-NEXT:  5      125   125.00                      sqrtpd %xmm0, %xmm2
+# CHECK-NEXT:  6      125   125.00  *                   sqrtpd (%rax), %xmm2
+# CHECK-NEXT:  3      62    62.00                       sqrtsd %xmm0, %xmm2
+# CHECK-NEXT:  4      62    62.00   *                   sqrtsd (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        subpd  %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   subpd  (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        subsd  %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   subsd  (%rax), %xmm2
 # CHECK-NEXT:  4      9     9.00                        ucomisd        %xmm0, %xmm1
@@ -681,7 +681,7 @@ xorpd       (%rax), %xmm2
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]
-# CHECK-NEXT: 932.00 792.00
+# CHECK-NEXT: 1306.00 1166.00
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    Instructions:
@@ -738,10 +738,10 @@ xorpd       (%rax), %xmm2
 # CHECK-NEXT: 7.00   7.00   cvttsd2si  %xmm0, %rcx
 # CHECK-NEXT: 6.00   6.00   cvttsd2si  (%rax), %ecx
 # CHECK-NEXT: 6.00   6.00   cvttsd2si  (%rax), %rcx
-# CHECK-NEXT: 62.50  62.50  divpd      %xmm0, %xmm2
-# CHECK-NEXT: 62.50  62.50  divpd      (%rax), %xmm2
-# CHECK-NEXT: 31.00  31.00  divsd      %xmm0, %xmm2
-# CHECK-NEXT: 31.00  31.00  divsd      (%rax), %xmm2
+# CHECK-NEXT: 125.00 125.00 divpd      %xmm0, %xmm2
+# CHECK-NEXT: 125.00 125.00 divpd      (%rax), %xmm2
+# CHECK-NEXT: 62.00  62.00  divsd      %xmm0, %xmm2
+# CHECK-NEXT: 62.00  62.00  divsd      (%rax), %xmm2
 # CHECK-NEXT: 0.50   0.50   lfence
 # CHECK-NEXT: 1.00   1.00   maskmovdqu %xmm0, %xmm1
 # CHECK-NEXT: 5.00   5.00   maxpd      %xmm0, %xmm2
@@ -936,10 +936,10 @@ xorpd       (%rax), %xmm2
 # CHECK-NEXT: 1.00    -     pxor       (%rax), %xmm2
 # CHECK-NEXT: 1.00    -     shufpd     $1, %xmm0, %xmm2
 # CHECK-NEXT: 1.00    -     shufpd     $1, (%rax), %xmm2
-# CHECK-NEXT: 62.50  62.50  sqrtpd     %xmm0, %xmm2
-# CHECK-NEXT: 62.50  62.50  sqrtpd     (%rax), %xmm2
-# CHECK-NEXT: 31.00  31.00  sqrtsd     %xmm0, %xmm2
-# CHECK-NEXT: 31.00  31.00  sqrtsd     (%rax), %xmm2
+# CHECK-NEXT: 125.00 125.00 sqrtpd     %xmm0, %xmm2
+# CHECK-NEXT: 125.00 125.00 sqrtpd     (%rax), %xmm2
+# CHECK-NEXT: 62.00  62.00  sqrtsd     %xmm0, %xmm2
+# CHECK-NEXT: 62.00  62.00  sqrtsd     (%rax), %xmm2
 # CHECK-NEXT: 5.00   5.00   subpd      %xmm0, %xmm2
 # CHECK-NEXT: 6.00   6.00   subpd      (%rax), %xmm2
 # CHECK-NEXT:  -     1.00   subsd      %xmm0, %xmm2

diff --git a/llvm/test/tools/llvm-mca/X86/Atom/resources-sse3.s b/llvm/test/tools/llvm-mca/X86/Atom/resources-sse3.s
index 65fe282..0519fe9 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/Atom/resources-sse3.s
+++ b/llvm/test/tools/llvm-mca/X86/Atom/resources-sse3.s
@@ -43,18 +43,18 @@ mwait
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      6     5.00                        addsubpd       %xmm0, %xmm2
-# CHECK-NEXT:  1      7     6.00    *                   addsubpd       (%rax), %xmm2
+# CHECK-NEXT:  3      6     5.00                        addsubpd       %xmm0, %xmm2
+# CHECK-NEXT:  4      7     6.00    *                   addsubpd       (%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        addsubps       %xmm0, %xmm2
 # CHECK-NEXT:  1      5     1.00    *                   addsubps       (%rax), %xmm2
-# CHECK-NEXT:  1      8     4.00                        haddpd %xmm0, %xmm2
-# CHECK-NEXT:  1      9     4.50    *                   haddpd (%rax), %xmm2
-# CHECK-NEXT:  1      8     4.00                        haddps %xmm0, %xmm2
-# CHECK-NEXT:  1      9     4.50    *                   haddps (%rax), %xmm2
-# CHECK-NEXT:  1      8     4.00                        hsubpd %xmm0, %xmm2
-# CHECK-NEXT:  1      9     4.50    *                   hsubpd (%rax), %xmm2
-# CHECK-NEXT:  1      8     4.00                        hsubps %xmm0, %xmm2
-# CHECK-NEXT:  1      9     4.50    *                   hsubps (%rax), %xmm2
+# CHECK-NEXT:  5      8     8.00                        haddpd %xmm0, %xmm2
+# CHECK-NEXT:  6      9     9.00    *                   haddpd (%rax), %xmm2
+# CHECK-NEXT:  5      8     8.00                        haddps %xmm0, %xmm2
+# CHECK-NEXT:  6      9     9.00    *                   haddps (%rax), %xmm2
+# CHECK-NEXT:  5      8     8.00                        hsubpd %xmm0, %xmm2
+# CHECK-NEXT:  6      9     9.00    *                   hsubpd (%rax), %xmm2
+# CHECK-NEXT:  5      8     8.00                        hsubps %xmm0, %xmm2
+# CHECK-NEXT:  6      9     9.00    *                   hsubps (%rax), %xmm2
 # CHECK-NEXT:  1      3     1.50    *                   lddqu  (%rax), %xmm2
 # CHECK-NEXT:  1      45    22.50                 U     monitor
 # CHECK-NEXT:  1      1     1.00                        movddup        %xmm0, %xmm2
@@ -71,7 +71,7 @@ mwait
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]
-# CHECK-NEXT: 99.00  94.00
+# CHECK-NEXT: 133.00 128.00
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    Instructions:
@@ -79,14 +79,14 @@ mwait
 # CHECK-NEXT: 6.00   6.00   addsubpd   (%rax), %xmm2
 # CHECK-NEXT:  -     1.00   addsubps   %xmm0, %xmm2
 # CHECK-NEXT: 1.00   1.00   addsubps   (%rax), %xmm2
-# CHECK-NEXT: 4.00   4.00   haddpd     %xmm0, %xmm2
-# CHECK-NEXT: 4.50   4.50   haddpd     (%rax), %xmm2
-# CHECK-NEXT: 4.00   4.00   haddps     %xmm0, %xmm2
-# CHECK-NEXT: 4.50   4.50   haddps     (%rax), %xmm2
-# CHECK-NEXT: 4.00   4.00   hsubpd     %xmm0, %xmm2
-# CHECK-NEXT: 4.50   4.50   hsubpd     (%rax), %xmm2
-# CHECK-NEXT: 4.00   4.00   hsubps     %xmm0, %xmm2
-# CHECK-NEXT: 4.50   4.50   hsubps     (%rax), %xmm2
+# CHECK-NEXT: 8.00   8.00   haddpd     %xmm0, %xmm2
+# CHECK-NEXT: 9.00   9.00   haddpd     (%rax), %xmm2
+# CHECK-NEXT: 8.00   8.00   haddps     %xmm0, %xmm2
+# CHECK-NEXT: 9.00   9.00   haddps     (%rax), %xmm2
+# CHECK-NEXT: 8.00   8.00   hsubpd     %xmm0, %xmm2
+# CHECK-NEXT: 9.00   9.00   hsubpd     (%rax), %xmm2
+# CHECK-NEXT: 8.00   8.00   hsubps     %xmm0, %xmm2
+# CHECK-NEXT: 9.00   9.00   hsubps     (%rax), %xmm2
 # CHECK-NEXT: 1.50   1.50   lddqu      (%rax), %xmm2
 # CHECK-NEXT: 22.50  22.50  monitor
 # CHECK-NEXT: 1.00    -     movddup    %xmm0, %xmm2

diff --git a/llvm/test/tools/llvm-mca/X86/Atom/resources-ssse3.s b/llvm/test/tools/llvm-mca/X86/Atom/resources-ssse3.s
index b04be1a..905d755 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/Atom/resources-ssse3.s
+++ b/llvm/test/tools/llvm-mca/X86/Atom/resources-ssse3.s
@@ -122,30 +122,30 @@ psignw      (%rax), %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   palignr        $1, (%rax), %mm2
 # CHECK-NEXT:  1      1     1.00                        palignr        $1, %xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00    *                   palignr        $1, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.50                        phaddd %mm0, %mm2
-# CHECK-NEXT:  1      4     2.00    *                   phaddd (%rax), %mm2
+# CHECK-NEXT:  3      3     3.00                        phaddd %mm0, %mm2
+# CHECK-NEXT:  4      4     4.00    *                   phaddd (%rax), %mm2
 # CHECK-NEXT:  1      3     1.50                        phaddd %xmm0, %xmm2
 # CHECK-NEXT:  1      4     2.00    *                   phaddd (%rax), %xmm2
 # CHECK-NEXT:  1      5     2.50                        phaddsw        %mm0, %mm2
 # CHECK-NEXT:  1      6     3.00    *                   phaddsw        (%rax), %mm2
-# CHECK-NEXT:  1      7     3.50                        phaddsw        %xmm0, %xmm2
-# CHECK-NEXT:  1      8     4.00    *                   phaddsw        (%rax), %xmm2
+# CHECK-NEXT:  3      7     7.00                        phaddsw        %xmm0, %xmm2
+# CHECK-NEXT:  4      8     8.00    *                   phaddsw        (%rax), %xmm2
 # CHECK-NEXT:  1      5     2.50                        phaddw %mm0, %mm2
 # CHECK-NEXT:  1      6     3.00    *                   phaddw (%rax), %mm2
-# CHECK-NEXT:  1      7     3.50                        phaddw %xmm0, %xmm2
-# CHECK-NEXT:  1      8     4.00    *                   phaddw (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.50                        phsubd %mm0, %mm2
-# CHECK-NEXT:  1      4     2.00    *                   phsubd (%rax), %mm2
+# CHECK-NEXT:  3      7     7.00                        phaddw %xmm0, %xmm2
+# CHECK-NEXT:  4      8     8.00    *                   phaddw (%rax), %xmm2
+# CHECK-NEXT:  3      3     3.00                        phsubd %mm0, %mm2
+# CHECK-NEXT:  4      4     4.00    *                   phsubd (%rax), %mm2
 # CHECK-NEXT:  1      3     1.50                        phsubd %xmm0, %xmm2
 # CHECK-NEXT:  1      4     2.00    *                   phsubd (%rax), %xmm2
 # CHECK-NEXT:  1      5     2.50                        phsubsw        %mm0, %mm2
 # CHECK-NEXT:  1      6     3.00    *                   phsubsw        (%rax), %mm2
-# CHECK-NEXT:  1      7     3.50                        phsubsw        %xmm0, %xmm2
-# CHECK-NEXT:  1      8     4.00    *                   phsubsw        (%rax), %xmm2
+# CHECK-NEXT:  3      7     7.00                        phsubsw        %xmm0, %xmm2
+# CHECK-NEXT:  4      8     8.00    *                   phsubsw        (%rax), %xmm2
 # CHECK-NEXT:  1      5     2.50                        phsubw %mm0, %mm2
 # CHECK-NEXT:  1      6     3.00    *                   phsubw (%rax), %mm2
-# CHECK-NEXT:  1      7     3.50                        phsubw %xmm0, %xmm2
-# CHECK-NEXT:  1      8     4.00    *                   phsubw (%rax), %xmm2
+# CHECK-NEXT:  3      7     7.00                        phsubw %xmm0, %xmm2
+# CHECK-NEXT:  4      8     8.00    *                   phsubw (%rax), %xmm2
 # CHECK-NEXT:  1      4     1.00                        pmaddubsw      %mm0, %mm2
 # CHECK-NEXT:  1      4     1.00    *                   pmaddubsw      (%rax), %mm2
 # CHECK-NEXT:  1      5     2.00                        pmaddubsw      %xmm0, %xmm2
@@ -177,7 +177,7 @@ psignw      (%rax), %xmm2
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]
-# CHECK-NEXT: 109.00 79.00
+# CHECK-NEXT: 146.00 116.00
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    Instructions:
@@ -197,30 +197,30 @@ psignw      (%rax), %xmm2
 # CHECK-NEXT: 1.00    -     palignr    $1, (%rax), %mm2
 # CHECK-NEXT: 1.00    -     palignr    $1, %xmm0, %xmm2
 # CHECK-NEXT: 1.00    -     palignr    $1, (%rax), %xmm2
-# CHECK-NEXT: 1.50   1.50   phaddd     %mm0, %mm2
-# CHECK-NEXT: 2.00   2.00   phaddd     (%rax), %mm2
+# CHECK-NEXT: 3.00   3.00   phaddd     %mm0, %mm2
+# CHECK-NEXT: 4.00   4.00   phaddd     (%rax), %mm2
 # CHECK-NEXT: 1.50   1.50   phaddd     %xmm0, %xmm2
 # CHECK-NEXT: 2.00   2.00   phaddd     (%rax), %xmm2
 # CHECK-NEXT: 2.50   2.50   phaddsw    %mm0, %mm2
 # CHECK-NEXT: 3.00   3.00   phaddsw    (%rax), %mm2
-# CHECK-NEXT: 3.50   3.50   phaddsw    %xmm0, %xmm2
-# CHECK-NEXT: 4.00   4.00   phaddsw    (%rax), %xmm2
+# CHECK-NEXT: 7.00   7.00   phaddsw    %xmm0, %xmm2
+# CHECK-NEXT: 8.00   8.00   phaddsw    (%rax), %xmm2
 # CHECK-NEXT: 2.50   2.50   phaddw     %mm0, %mm2
 # CHECK-NEXT: 3.00   3.00   phaddw     (%rax), %mm2
-# CHECK-NEXT: 3.50   3.50   phaddw     %xmm0, %xmm2
-# CHECK-NEXT: 4.00   4.00   phaddw     (%rax), %xmm2
-# CHECK-NEXT: 1.50   1.50   phsubd     %mm0, %mm2
-# CHECK-NEXT: 2.00   2.00   phsubd     (%rax), %mm2
+# CHECK-NEXT: 7.00   7.00   phaddw     %xmm0, %xmm2
+# CHECK-NEXT: 8.00   8.00   phaddw     (%rax), %xmm2
+# CHECK-NEXT: 3.00   3.00   phsubd     %mm0, %mm2
+# CHECK-NEXT: 4.00   4.00   phsubd     (%rax), %mm2
 # CHECK-NEXT: 1.50   1.50   phsubd     %xmm0, %xmm2
 # CHECK-NEXT: 2.00   2.00   phsubd     (%rax), %xmm2
 # CHECK-NEXT: 2.50   2.50   phsubsw    %mm0, %mm2
 # CHECK-NEXT: 3.00   3.00   phsubsw    (%rax), %mm2
-# CHECK-NEXT: 3.50   3.50   phsubsw    %xmm0, %xmm2
-# CHECK-NEXT: 4.00   4.00   phsubsw    (%rax), %xmm2
+# CHECK-NEXT: 7.00   7.00   phsubsw    %xmm0, %xmm2
+# CHECK-NEXT: 8.00   8.00   phsubsw    (%rax), %xmm2
 # CHECK-NEXT: 2.50   2.50   phsubw     %mm0, %mm2
 # CHECK-NEXT: 3.00   3.00   phsubw     (%rax), %mm2
-# CHECK-NEXT: 3.50   3.50   phsubw     %xmm0, %xmm2
-# CHECK-NEXT: 4.00   4.00   phsubw     (%rax), %xmm2
+# CHECK-NEXT: 7.00   7.00   phsubw     %xmm0, %xmm2
+# CHECK-NEXT: 8.00   8.00   phsubw     (%rax), %xmm2
 # CHECK-NEXT: 1.00    -     pmaddubsw  %mm0, %mm2
 # CHECK-NEXT: 1.00    -     pmaddubsw  (%rax), %mm2
 # CHECK-NEXT: 2.00    -     pmaddubsw  %xmm0, %xmm2

diff --git a/llvm/test/tools/llvm-mca/X86/Atom/resources-x87.s b/llvm/test/tools/llvm-mca/X86/Atom/resources-x87.s
index 9952afd..07e8531 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/Atom/resources-x87.s
+++ b/llvm/test/tools/llvm-mca/X86/Atom/resources-x87.s
@@ -241,22 +241,22 @@ fyl2xp1
 # CHECK-NEXT:  1      9     4.50                  U     fcompi %st(3), %st
 # CHECK-NEXT:  1      174   87.00                 U     fcos
 # CHECK-NEXT:  1      1     0.50                  U     fdecstp
-# CHECK-NEXT:  1      34    17.00                 U     fdiv   %st, %st(1)
-# CHECK-NEXT:  1      34    17.00                 U     fdiv   %st(2), %st
-# CHECK-NEXT:  1      34    17.00   *             U     fdivs  (%ecx)
-# CHECK-NEXT:  1      34    17.00   *             U     fdivl  (%eax)
-# CHECK-NEXT:  1      34    17.00                 U     fdivp  %st, %st(1)
-# CHECK-NEXT:  1      34    17.00                 U     fdivp  %st, %st(2)
-# CHECK-NEXT:  1      34    17.00   *             U     fidivs (%ecx)
-# CHECK-NEXT:  1      34    17.00   *             U     fidivl (%eax)
-# CHECK-NEXT:  1      34    17.00                 U     fdivr  %st, %st(1)
-# CHECK-NEXT:  1      34    17.00                 U     fdivr  %st(2), %st
-# CHECK-NEXT:  1      34    17.00   *             U     fdivrs (%ecx)
-# CHECK-NEXT:  1      34    17.00   *             U     fdivrl (%eax)
-# CHECK-NEXT:  1      34    17.00                 U     fdivrp %st, %st(1)
-# CHECK-NEXT:  1      34    17.00                 U     fdivrp %st, %st(2)
-# CHECK-NEXT:  1      34    17.00   *             U     fidivrs        (%ecx)
-# CHECK-NEXT:  1      34    17.00   *             U     fidivrl        (%eax)
+# CHECK-NEXT:  3      34    34.00                 U     fdiv   %st, %st(1)
+# CHECK-NEXT:  3      34    34.00                 U     fdiv   %st(2), %st
+# CHECK-NEXT:  4      34    34.00   *             U     fdivs  (%ecx)
+# CHECK-NEXT:  4      34    34.00   *             U     fdivl  (%eax)
+# CHECK-NEXT:  3      34    34.00                 U     fdivp  %st, %st(1)
+# CHECK-NEXT:  3      34    34.00                 U     fdivp  %st, %st(2)
+# CHECK-NEXT:  4      34    34.00   *             U     fidivs (%ecx)
+# CHECK-NEXT:  4      34    34.00   *             U     fidivl (%eax)
+# CHECK-NEXT:  3      34    34.00                 U     fdivr  %st, %st(1)
+# CHECK-NEXT:  3      34    34.00                 U     fdivr  %st(2), %st
+# CHECK-NEXT:  4      34    34.00   *             U     fdivrs (%ecx)
+# CHECK-NEXT:  4      34    34.00   *             U     fdivrl (%eax)
+# CHECK-NEXT:  3      34    34.00                 U     fdivrp %st, %st(1)
+# CHECK-NEXT:  3      34    34.00                 U     fdivrp %st, %st(2)
+# CHECK-NEXT:  4      34    34.00   *             U     fidivrs        (%ecx)
+# CHECK-NEXT:  4      34    34.00   *             U     fidivrl        (%eax)
 # CHECK-NEXT:  1      1     0.50                  U     ffree  %st(0)
 # CHECK-NEXT:  1      5     5.00    *             U     ficoms (%ecx)
 # CHECK-NEXT:  1      5     5.00    *             U     ficoml (%eax)
@@ -307,7 +307,7 @@ fyl2xp1
 # CHECK-NEXT:  1      77    38.50                 U     fscale
 # CHECK-NEXT:  1      174   87.00                 U     fsin
 # CHECK-NEXT:  1      174   87.00                 U     fsincos
-# CHECK-NEXT:  1      71    35.50                 U     fsqrt
+# CHECK-NEXT:  1      71    71.00                 U     fsqrt
 # CHECK-NEXT:  1      2     1.00                  U     fst    %st(0)
 # CHECK-NEXT:  1      2     1.00           *      U     fsts   (%edx)
 # CHECK-NEXT:  1      2     1.00           *      U     fstl   (%ecx)
@@ -361,7 +361,7 @@ fyl2xp1
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]
-# CHECK-NEXT: 1500.00 1440.00
+# CHECK-NEXT: 1807.50 1676.50
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    Instructions:
@@ -400,22 +400,22 @@ fyl2xp1
 # CHECK-NEXT: 4.50   4.50   fcompi     %st(3), %st
 # CHECK-NEXT: 87.00  87.00  fcos
 # CHECK-NEXT: 0.50   0.50   fdecstp
-# CHECK-NEXT: 17.00  17.00  fdiv       %st, %st(1)
-# CHECK-NEXT: 17.00  17.00  fdiv       %st(2), %st
-# CHECK-NEXT: 17.00  17.00  fdivs      (%ecx)
-# CHECK-NEXT: 17.00  17.00  fdivl      (%eax)
-# CHECK-NEXT: 17.00  17.00  fdivp      %st, %st(1)
-# CHECK-NEXT: 17.00  17.00  fdivp      %st, %st(2)
-# CHECK-NEXT: 17.00  17.00  fidivs     (%ecx)
-# CHECK-NEXT: 17.00  17.00  fidivl     (%eax)
-# CHECK-NEXT: 17.00  17.00  fdivr      %st, %st(1)
-# CHECK-NEXT: 17.00  17.00  fdivr      %st(2), %st
-# CHECK-NEXT: 17.00  17.00  fdivrs     (%ecx)
-# CHECK-NEXT: 17.00  17.00  fdivrl     (%eax)
-# CHECK-NEXT: 17.00  17.00  fdivrp     %st, %st(1)
-# CHECK-NEXT: 17.00  17.00  fdivrp     %st, %st(2)
-# CHECK-NEXT: 17.00  17.00  fidivrs    (%ecx)
-# CHECK-NEXT: 17.00  17.00  fidivrl    (%eax)
+# CHECK-NEXT: 34.00  34.00  fdiv       %st, %st(1)
+# CHECK-NEXT: 34.00  34.00  fdiv       %st(2), %st
+# CHECK-NEXT: 34.00  34.00  fdivs      (%ecx)
+# CHECK-NEXT: 34.00  34.00  fdivl      (%eax)
+# CHECK-NEXT: 34.00  34.00  fdivp      %st, %st(1)
+# CHECK-NEXT: 34.00  34.00  fdivp      %st, %st(2)
+# CHECK-NEXT: 34.00  34.00  fidivs     (%ecx)
+# CHECK-NEXT: 34.00  34.00  fidivl     (%eax)
+# CHECK-NEXT: 34.00  34.00  fdivr      %st, %st(1)
+# CHECK-NEXT: 34.00  34.00  fdivr      %st(2), %st
+# CHECK-NEXT: 34.00  34.00  fdivrs     (%ecx)
+# CHECK-NEXT: 34.00  34.00  fdivrl     (%eax)
+# CHECK-NEXT: 34.00  34.00  fdivrp     %st, %st(1)
+# CHECK-NEXT: 34.00  34.00  fdivrp     %st, %st(2)
+# CHECK-NEXT: 34.00  34.00  fidivrs    (%ecx)
+# CHECK-NEXT: 34.00  34.00  fidivrl    (%eax)
 # CHECK-NEXT: 0.50   0.50   ffree      %st(0)
 # CHECK-NEXT: 5.00    -     ficoms     (%ecx)
 # CHECK-NEXT: 5.00    -     ficoml     (%eax)
@@ -466,7 +466,7 @@ fyl2xp1
 # CHECK-NEXT: 38.50  38.50  fscale
 # CHECK-NEXT: 87.00  87.00  fsin
 # CHECK-NEXT: 87.00  87.00  fsincos
-# CHECK-NEXT: 35.50  35.50  fsqrt
+# CHECK-NEXT: 71.00   -     fsqrt
 # CHECK-NEXT: 1.00   1.00   fst        %st(0)
 # CHECK-NEXT: 1.00   1.00   fsts       (%edx)
 # CHECK-NEXT: 1.00   1.00   fstl       (%ecx)