[X86] Improve LOOP/LOOPE/LOOPNE schedule on SandyBridge model

D138359 was reporting that this override was superfluous, but it had never been setup - I took the numbers from uops.info (I couldn't find an estimate in Intel docs).

diff --git a/llvm/lib/Target/X86/X86SchedSandyBridge.td b/llvm/lib/Target/X86/X86SchedSandyBridge.td
index 9d7069a..9ceb8db 100644 (file)
--- a/llvm/lib/Target/X86/X86SchedSandyBridge.td
+++ b/llvm/lib/Target/X86/X86SchedSandyBridge.td
@@ -615,7 +615,6 @@ def SBWriteResGroup2 : SchedWriteRes<[SBPort5]> {
 }
 def: InstRW<[SBWriteResGroup2], (instrs FDECSTP, FINCSTP, FFREE, FFREEP, FNOP,
                                         LD_Frr, ST_Frr, ST_FPrr)>;
-def: InstRW<[SBWriteResGroup2], (instrs LOOP, LOOPE, LOOPNE)>; // FIXME: This seems wrong compared to other Intel CPUs.
 def: InstRW<[SBWriteResGroup2], (instrs RET64)>;
 
 def SBWriteResGroup4 : SchedWriteRes<[SBPort05]> {
@@ -736,6 +735,20 @@ def SBWriteResGroup29_2 : SchedWriteRes<[SBPort5,SBPort015]> {
 }
 def: InstRW<[SBWriteResGroup29_2], (instrs PAUSE)>;
 
+def SBWriteResGroup30 : SchedWriteRes<[SBPort1,SBPort5,SBPort015]> {
+  let Latency = 3;
+  let NumMicroOps = 8;
+  let ResourceCycles = [1,3,4];
+}
+def: InstRW<[SBWriteResGroup30], (instrs LOOP)>;
+
+def SBWriteResGroup31 : SchedWriteRes<[SBPort1,SBPort5,SBPort015,SBPort05]> {
+  let Latency = 4;
+  let NumMicroOps = 12;
+  let ResourceCycles = [1,3,6,2];
+}
+def: InstRW<[SBWriteResGroup31], (instrs LOOPE, LOOPNE)>;
+
 def SBWriteResGroup76 : SchedWriteRes<[SBPort05]> {
   let Latency = 5;
   let NumMicroOps = 8;

diff --git a/llvm/test/tools/llvm-mca/X86/Barcelona/resources-x86_64.s b/llvm/test/tools/llvm-mca/X86/Barcelona/resources-x86_64.s
index 358a1aa..37ccff2 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/Barcelona/resources-x86_64.s
+++ b/llvm/test/tools/llvm-mca/X86/Barcelona/resources-x86_64.s
@@ -1394,9 +1394,9 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  3      7     0.67                  U     lodsw  (%rsi), %ax
 # CHECK-NEXT:  2      6     0.50                  U     lodsl  (%rsi), %eax
 # CHECK-NEXT:  2      6     0.50                  U     lodsq  (%rsi), %rax
-# CHECK-NEXT:  1      1     1.00                  U     loop   0
-# CHECK-NEXT:  1      1     1.00                  U     loope  0
-# CHECK-NEXT:  1      1     1.00                  U     loopne 0
+# CHECK-NEXT:  8      3     3.00                  U     loop   0
+# CHECK-NEXT:  12     4     4.00                  U     loope  0
+# CHECK-NEXT:  12     4     4.00                  U     loopne 0
 # CHECK-NEXT:  5      8     1.00                  U     movsb  (%rsi), %es:(%rdi)
 # CHECK-NEXT:  5      8     1.00                  U     movsw  (%rsi), %es:(%rdi)
 # CHECK-NEXT:  5      8     1.00                  U     movsl  (%rsi), %es:(%rdi)
@@ -1960,7 +1960,7 @@ xorq (%rax), %rdi
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT: 160.00  -     658.17 310.67 361.00 686.17 455.50 455.50
+# CHECK-NEXT: 160.00  -     665.50 319.00 361.00 699.50 455.50 455.50
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
@@ -2324,9 +2324,9 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  -      -     0.67   0.67    -     0.67   0.50   0.50   lodsw    (%rsi), %ax
 # CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsl    (%rsi), %eax
 # CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsq    (%rsi), %rax
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loop     0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loope    0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loopne   0
+# CHECK-NEXT:  -      -     1.33   2.33    -     4.33    -      -     loop     0
+# CHECK-NEXT:  -      -     3.00   3.00    -     6.00    -      -     loope    0
+# CHECK-NEXT:  -      -     3.00   3.00    -     6.00    -      -     loopne   0
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsb    (%rsi), %es:(%rdi)
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsw    (%rsi), %es:(%rdi)
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsl    (%rsi), %es:(%rdi)

diff --git a/llvm/test/tools/llvm-mca/X86/Generic/resources-x86_64.s b/llvm/test/tools/llvm-mca/X86/Generic/resources-x86_64.s
index 358a1aa..37ccff2 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/Generic/resources-x86_64.s
+++ b/llvm/test/tools/llvm-mca/X86/Generic/resources-x86_64.s
@@ -1394,9 +1394,9 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  3      7     0.67                  U     lodsw  (%rsi), %ax
 # CHECK-NEXT:  2      6     0.50                  U     lodsl  (%rsi), %eax
 # CHECK-NEXT:  2      6     0.50                  U     lodsq  (%rsi), %rax
-# CHECK-NEXT:  1      1     1.00                  U     loop   0
-# CHECK-NEXT:  1      1     1.00                  U     loope  0
-# CHECK-NEXT:  1      1     1.00                  U     loopne 0
+# CHECK-NEXT:  8      3     3.00                  U     loop   0
+# CHECK-NEXT:  12     4     4.00                  U     loope  0
+# CHECK-NEXT:  12     4     4.00                  U     loopne 0
 # CHECK-NEXT:  5      8     1.00                  U     movsb  (%rsi), %es:(%rdi)
 # CHECK-NEXT:  5      8     1.00                  U     movsw  (%rsi), %es:(%rdi)
 # CHECK-NEXT:  5      8     1.00                  U     movsl  (%rsi), %es:(%rdi)
@@ -1960,7 +1960,7 @@ xorq (%rax), %rdi
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT: 160.00  -     658.17 310.67 361.00 686.17 455.50 455.50
+# CHECK-NEXT: 160.00  -     665.50 319.00 361.00 699.50 455.50 455.50
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
@@ -2324,9 +2324,9 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  -      -     0.67   0.67    -     0.67   0.50   0.50   lodsw    (%rsi), %ax
 # CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsl    (%rsi), %eax
 # CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsq    (%rsi), %rax
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loop     0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loope    0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loopne   0
+# CHECK-NEXT:  -      -     1.33   2.33    -     4.33    -      -     loop     0
+# CHECK-NEXT:  -      -     3.00   3.00    -     6.00    -      -     loope    0
+# CHECK-NEXT:  -      -     3.00   3.00    -     6.00    -      -     loopne   0
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsb    (%rsi), %es:(%rdi)
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsw    (%rsi), %es:(%rdi)
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsl    (%rsi), %es:(%rdi)

diff --git a/llvm/test/tools/llvm-mca/X86/SandyBridge/resources-x86_64.s b/llvm/test/tools/llvm-mca/X86/SandyBridge/resources-x86_64.s
index c56257b..c046cbd 100644 (file)
--- a/llvm/test/tools/llvm-mca/X86/SandyBridge/resources-x86_64.s
+++ b/llvm/test/tools/llvm-mca/X86/SandyBridge/resources-x86_64.s
@@ -1394,9 +1394,9 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  3      7     0.67                  U     lodsw  (%rsi), %ax
 # CHECK-NEXT:  2      6     0.50                  U     lodsl  (%rsi), %eax
 # CHECK-NEXT:  2      6     0.50                  U     lodsq  (%rsi), %rax
-# CHECK-NEXT:  1      1     1.00                  U     loop   0
-# CHECK-NEXT:  1      1     1.00                  U     loope  0
-# CHECK-NEXT:  1      1     1.00                  U     loopne 0
+# CHECK-NEXT:  8      3     3.00                  U     loop   0
+# CHECK-NEXT:  12     4     4.00                  U     loope  0
+# CHECK-NEXT:  12     4     4.00                  U     loopne 0
 # CHECK-NEXT:  5      8     1.00                  U     movsb  (%rsi), %es:(%rdi)
 # CHECK-NEXT:  5      8     1.00                  U     movsw  (%rsi), %es:(%rdi)
 # CHECK-NEXT:  5      8     1.00                  U     movsl  (%rsi), %es:(%rdi)
@@ -1960,7 +1960,7 @@ xorq (%rax), %rdi
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT: 160.00  -     658.17 310.67 361.00 686.17 455.50 455.50
+# CHECK-NEXT: 160.00  -     665.50 319.00 361.00 699.50 455.50 455.50
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
@@ -2324,9 +2324,9 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  -      -     0.67   0.67    -     0.67   0.50   0.50   lodsw    (%rsi), %ax
 # CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsl    (%rsi), %eax
 # CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsq    (%rsi), %rax
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loop     0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loope    0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     loopne   0
+# CHECK-NEXT:  -      -     1.33   2.33    -     4.33    -      -     loop     0
+# CHECK-NEXT:  -      -     3.00   3.00    -     6.00    -      -     loope    0
+# CHECK-NEXT:  -      -     3.00   3.00    -     6.00    -      -     loopne   0
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsb    (%rsi), %es:(%rdi)
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsw    (%rsi), %es:(%rdi)
 # CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsl    (%rsi), %es:(%rdi)