[X86][BtVer2] X86ISD::VPERMILPV has local forwarding disabled

[RKSimon]

Summary

Similar to horizontal ops on D56777, the vpermilpd/vpermilps variable mask ops has local forwarding disabled, adding +1cy to the use latency for the result.

[spatel]

LGTM

lib/Target/X86/X86ScheduleBtVer2.td

 defm : JWriteResYMMPair<WriteFLogicY,     [JFPU01, JFPX],  1, [2, 2], 2>;
 defm : X86WriteResPairUnsupported<WriteFLogicZ>;
 defm : JWriteResFpuPair<WriteFTest,       [JFPU0, JFPA, JALU0], 3>;
 defm : JWriteResYMMPair<WriteFTestY ,     [JFPU01, JFPX, JFPA, JALU0], 4, [2, 2, 2, 1], 3>;
 defm : X86WriteResPairUnsupported<WriteFTestZ>;
 defm : JWriteResFpuPair<WriteFShuffle,    [JFPU01, JFPX],  1>;
 defm : JWriteResYMMPair<WriteFShuffleY,   [JFPU01, JFPX],  1, [2, 2], 2>;
 defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
-defm : JWriteResFpuPair<WriteFVarShuffle, [JFPU01, JFPX],  2, [1, 4], 3>;
-defm : JWriteResYMMPair<WriteFVarShuffleY,[JFPU01, JFPX],  3, [2, 6], 6>;
+defm : JWriteResFpuPair<WriteFVarShuffle, [JFPU01, JFPX],  3, [1, 4], 3>; // +1cy latency.
+defm : JWriteResYMMPair<WriteFVarShuffleY,[JFPU01, JFPX],  4, [2, 6], 6>; // +1cy latency.
 defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
 defm : JWriteResFpuPair<WriteFBlend,      [JFPU01, JFPX],  1>;
 defm : JWriteResYMMPair<WriteFBlendY,     [JFPU01, JFPX],  1, [2, 2], 2>;
 defm : X86WriteResPairUnsupported<WriteFBlendZ>;
 defm : JWriteResFpuPair<WriteFVarBlend,   [JFPU01, JFPX],  2, [4, 4], 3>;
 defm : JWriteResYMMPair<WriteFVarBlendY,  [JFPU01, JFPX],  3, [6, 6], 6>;
 defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
 defm : JWriteResFpuPair<WriteFShuffle256, [JFPU01, JFPX],  1, [2, 2], 2>;

test/CodeGen/X86/avx-schedule.ll

 ; BDVER2-LABEL: test_permilvarpd:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vpermilpd %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
 ; BDVER2-NEXT:    vpermilpd (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
 ; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarpd:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    vpermilpd %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
-; BTVER2-NEXT:    vpermilpd (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
+; BTVER2-NEXT:    vpermilpd %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BTVER2-NEXT:    vpermilpd (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
 ; ZNVER1-LABEL: test_permilvarpd:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vpermilpd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
 ; ZNVER1-NEXT:    vpermilpd (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
 ; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %1 = call <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> %a0, <2 x i64> %a1)

 ; BDVER2-LABEL: test_permilvarpd_ymm:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
 ; BDVER2-NEXT:    vpermilpd (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
 ; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarpd_ymm:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
-; BTVER2-NEXT:    vpermilpd (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
+; BTVER2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0 # sched: [4:3.00]
+; BTVER2-NEXT:    vpermilpd (%rdi), %ymm0, %ymm0 # sched: [9:3.00]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
 ; ZNVER1-LABEL: test_permilvarpd_ymm:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
 ; ZNVER1-NEXT:    vpermilpd (%rdi), %ymm0, %ymm0 # sched: [8:0.50]
 ; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %1 = call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %a0, <4 x i64> %a1)

 ; BDVER2-LABEL: test_permilvarps:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vpermilps %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
 ; BDVER2-NEXT:    vpermilps (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
 ; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarps:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    vpermilps %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
-; BTVER2-NEXT:    vpermilps (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
+; BTVER2-NEXT:    vpermilps %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BTVER2-NEXT:    vpermilps (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
 ; ZNVER1-LABEL: test_permilvarps:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vpermilps %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
 ; ZNVER1-NEXT:    vpermilps (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
 ; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %1 = call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> %a1)

 ; BDVER2-LABEL: test_permilvarps_ymm:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vpermilps %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
 ; BDVER2-NEXT:    vpermilps (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
 ; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarps_ymm:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    vpermilps %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
-; BTVER2-NEXT:    vpermilps (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
+; BTVER2-NEXT:    vpermilps %ymm1, %ymm0, %ymm0 # sched: [4:3.00]
+; BTVER2-NEXT:    vpermilps (%rdi), %ymm0, %ymm0 # sched: [9:3.00]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
 ; ZNVER1-LABEL: test_permilvarps_ymm:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
 ; ZNVER1-NEXT:    vpermilps (%rdi), %ymm0, %ymm0 # sched: [8:0.50]
 ; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %1 = call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> %a1)

test/tools/llvm-mca/X86/BtVer2/resources-avx1.s

 # CHECK-NEXT:  3      7     2.00                        vpcmpistri  $1, %xmm0, %xmm2
 # CHECK-NEXT:  3      12    2.00    *                   vpcmpistri  $1, (%rax), %xmm2
 # CHECK-NEXT:  3      8     2.00                        vpcmpistrm  $1, %xmm0, %xmm2
 # CHECK-NEXT:  3      13    2.00    *                   vpcmpistrm  $1, (%rax), %xmm2
 # CHECK-NEXT:  2      1     1.00                        vperm2f128  $1, %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      6     1.00    *                   vperm2f128  $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      1     0.50                        vpermilpd $1, %xmm0, %xmm2
 # CHECK-NEXT:  1      6     1.00    *                   vpermilpd $1, (%rax), %xmm2
-# CHECK-NEXT:  3      2     2.00                        vpermilpd %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      7     2.00    *                   vpermilpd (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  3      3     2.00                        vpermilpd %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  3      8     2.00    *                   vpermilpd (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  2      1     1.00                        vpermilpd $1, %ymm0, %ymm2
 # CHECK-NEXT:  2      6     2.00    *                   vpermilpd $1, (%rax), %ymm2
-# CHECK-NEXT:  6      3     3.00                        vpermilpd %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  6      8     3.00    *                   vpermilpd (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  6      4     3.00                        vpermilpd %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  6      9     3.00    *                   vpermilpd (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      1     0.50                        vpermilps $1, %xmm0, %xmm2
 # CHECK-NEXT:  1      6     1.00    *                   vpermilps $1, (%rax), %xmm2
-# CHECK-NEXT:  3      2     2.00                        vpermilps %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      7     2.00    *                   vpermilps (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  3      3     2.00                        vpermilps %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  3      8     2.00    *                   vpermilps (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  2      1     1.00                        vpermilps $1, %ymm0, %ymm2
 # CHECK-NEXT:  2      6     2.00    *                   vpermilps $1, (%rax), %ymm2
-# CHECK-NEXT:  6      3     3.00                        vpermilps %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  6      8     3.00    *                   vpermilps (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  6      4     3.00                        vpermilps %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  6      9     3.00    *                   vpermilps (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      3     1.00                        vpextrb $1, %xmm0, %ecx
 # CHECK-NEXT:  1      3     1.00           *            vpextrb $1, %xmm0, (%rax)
 # CHECK-NEXT:  1      3     1.00                        vpextrd $1, %xmm0, %ecx
 # CHECK-NEXT:  1      3     1.00           *            vpextrd $1, %xmm0, (%rax)
 # CHECK-NEXT:  1      3     1.00                        vpextrq $1, %xmm0, %rcx
 # CHECK-NEXT:  1      3     1.00           *            vpextrq $1, %xmm0, (%rax)
 # CHECK-NEXT:  1      3     1.00                        vpextrw $1, %xmm0, %ecx
 # CHECK-NEXT:  1      3     1.00           *            vpextrw $1, %xmm0, (%rax)