[llvm] 3a766dc - [AArch64] Fix throughout of 64-bit SVE gather loads (#168572)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Nov 28 07:16:49 PST 2025
Author: Asher Dobrescu
Date: 2025-11-28T15:16:45Z
New Revision: 3a766dcbc462946ad34c1306be70cbbdc1dc868b
URL: https://github.com/llvm/llvm-project/commit/3a766dcbc462946ad34c1306be70cbbdc1dc868b
DIFF: https://github.com/llvm/llvm-project/commit/3a766dcbc462946ad34c1306be70cbbdc1dc868b.diff
LOG: [AArch64] Fix throughout of 64-bit SVE gather loads (#168572)
In the Neoverse N3 Software Optimisation Guide, SVE non termporal gather
load, vector+scalar 64-bit element size and gather load, vector + imm,
64-bit element size both show throughput of 4/5. However, it currently
shows as 2/3. This patch adds a new resource group in order to show the
correct throughput.
Added:
Modified:
llvm/lib/Target/AArch64/AArch64SchedNeoverseN3.td
llvm/test/tools/llvm-mca/AArch64/Neoverse/N3-sve-instructions.s
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/AArch64SchedNeoverseN3.td b/llvm/lib/Target/AArch64/AArch64SchedNeoverseN3.td
index 78bbf3f923c26..22e6d1107a337 100644
--- a/llvm/lib/Target/AArch64/AArch64SchedNeoverseN3.td
+++ b/llvm/lib/Target/AArch64/AArch64SchedNeoverseN3.td
@@ -49,6 +49,12 @@ def N3UnitM : ProcResGroup<[N3UnitM0, N3UnitM1]>;
def N3UnitL : ProcResGroup<[N3UnitL01, N3UnitL2]>;
def N3UnitI : ProcResGroup<[N3UnitS, N3UnitM0, N3UnitM1]>;
+// Group required for modelling SVE gather loads throughput
+def N3UnitVL : ProcResGroup<[N3UnitL01, N3UnitV0, N3UnitV1]>;
+// Unused group to fix: "error: proc resource group overlaps with N3UnitVL but
+// no supergroup contains both."
+def : ProcResGroup<[N3UnitL01, N3UnitL2, N3UnitV0, N3UnitV1]>;
+
//===----------------------------------------------------------------------===//
def : ReadAdvance<ReadI, 0>;
@@ -321,6 +327,12 @@ def N3Write_6c_2I_2L : SchedWriteRes<[N3UnitI, N3UnitI, N3UnitL, N3UnitL]> {
let NumMicroOps = 4;
}
+def N3Write_6c_2L01_2V : SchedWriteRes<[N3UnitVL]> {
+ let Latency = 6;
+ let NumMicroOps = 4;
+ let ReleaseAtCycles = [5];
+}
+
def N3Write_6c_4V0 : SchedWriteRes<[N3UnitV0, N3UnitV0, N3UnitV0, N3UnitV0]> {
let Latency = 6;
let NumMicroOps = 4;
@@ -2270,8 +2282,8 @@ def : InstRW<[N3Write_7c_4L], (instregex "^LDNT1[BHW]_ZZR_S$",
"^LDNT1S[BH]_ZZR_S$")>;
// Non temporal gather load, vector + scalar 64-bit element size
-def : InstRW<[N3Write_6c_2L], (instregex "^LDNT1S?[BHW]_ZZR_D$")>;
-def : InstRW<[N3Write_6c_2L], (instrs LDNT1D_ZZR_D)>;
+def : InstRW<[N3Write_6c_2L01_2V], (instregex "^LDNT1S?[BHW]_ZZR_D$")>;
+def : InstRW<[N3Write_6c_2L01_2V], (instrs LDNT1D_ZZR_D)>;
// Contiguous first faulting load, scalar + scalar
def : InstRW<[N3Write_6c_1L], (instregex "^LDFF1[BHWD]$",
@@ -2320,11 +2332,11 @@ def : InstRW<[N3Write_7c_4L], (instregex "^GLD(FF)?1S?[BH]_S_IMM$",
"^GLD(FF)?1W_IMM$")>;
// Gather load, vector + imm, 64-bit element size
-def : InstRW<[N3Write_6c_2L], (instregex "^GLD(FF)?1S?[BHW]_D_IMM$",
+def : InstRW<[N3Write_6c_2L01_2V], (instregex "^GLD(FF)?1S?[BHW]_D_IMM$",
"^GLD(FF)?1D_IMM$")>;
// Gather load, 64-bit element size
-def : InstRW<[N3Write_6c_2L],
+def : InstRW<[N3Write_6c_2L01_2V],
(instregex "^GLD(FF)?1S?[BHW]_D_[SU]XTW(_SCALED)?$",
"^GLD(FF)?1S?[BHW]_D(_SCALED)?$",
"^GLD(FF)?1D_[SU]XTW(_SCALED)?$",
diff --git a/llvm/test/tools/llvm-mca/AArch64/Neoverse/N3-sve-instructions.s b/llvm/test/tools/llvm-mca/AArch64/Neoverse/N3-sve-instructions.s
index 0093b28a756ee..38893c37b33d7 100644
--- a/llvm/test/tools/llvm-mca/AArch64/Neoverse/N3-sve-instructions.s
+++ b/llvm/test/tools/llvm-mca/AArch64/Neoverse/N3-sve-instructions.s
@@ -4410,7 +4410,7 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: 1 6 0.33 * ld1b { z0.b }, p0/z, [x0, x0]
# CHECK-NEXT: 1 6 0.33 * ld1b { z0.b }, p0/z, [x0]
# CHECK-NEXT: 1 6 0.33 * ld1b { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ld1b { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * ld1b { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * ld1b { z0.h }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ld1b { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * ld1b { z0.s }, p0/z, [x0, z0.s, uxtw]
@@ -4418,53 +4418,53 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: 4 7 1.33 * ld1b { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: 1 6 0.33 * ld1b { z21.b }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1b { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1b { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * ld1b { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1b { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1b { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1b { z21.h }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1b { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1b { z21.s }, p5/z, [x10, x21]
# CHECK-NEXT: 1 6 0.33 * ld1b { z23.d }, p3/z, [x13, x8]
# CHECK-NEXT: 1 6 0.33 * ld1b { z31.b }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1b { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1b { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: 2 6 0.67 * ld1b { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: 4 6 1.25 * ld1b { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * ld1b { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: 1 6 0.33 * ld1b { z31.h }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1b { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 4 7 1.33 * ld1b { z31.s }, p7/z, [z31.s, #31]
# CHECK-NEXT: 1 6 0.33 * ld1b { z5.h }, p3/z, [x17, x16]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
# CHECK-NEXT: 1 6 0.33 * ld1d { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * ld1d { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1d { z23.d }, p3/z, [sp, x8, lsl #3]
# CHECK-NEXT: 1 6 0.33 * ld1d { z23.d }, p3/z, [x13, x8, lsl #3]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
# CHECK-NEXT: 1 6 0.33 * ld1d { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: 2 6 0.67 * ld1d { z31.d }, p7/z, [z31.d, #248]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * ld1d { z31.d }, p7/z, [z31.d, #248]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
# CHECK-NEXT: 1 6 0.33 * ld1h { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * ld1h { z0.h }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ld1h { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * ld1h { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1h { z0.s }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ld1h { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: 1 6 0.33 * ld1h { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1h { z21.h }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1h { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1h { z21.s }, p5/z, [x10, x21, lsl #1]
# CHECK-NEXT: 1 6 0.33 * ld1h { z23.d }, p3/z, [x13, x8, lsl #1]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
# CHECK-NEXT: 1 6 0.33 * ld1h { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: 2 6 0.67 * ld1h { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * ld1h { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: 1 6 0.33 * ld1h { z31.h }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1h { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 4 7 1.33 * ld1h { z31.s }, p7/z, [sp, z31.s, sxtw #1]
@@ -4525,7 +4525,7 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: 1 6 0.33 * ld1rw { z31.d }, p7/z, [sp, #252]
# CHECK-NEXT: 1 6 0.33 * ld1rw { z31.s }, p7/z, [sp, #252]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ld1sb { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * ld1sb { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z0.h }, p0/z, [sp, x0]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z0.h }, p0/z, [x0, x0]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z0.h }, p0/z, [x0]
@@ -4533,73 +4533,73 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: 1 6 0.33 * ld1sb { z0.s }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ld1sb { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * ld1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z21.h }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z21.s }, p5/z, [x10, x21]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z23.d }, p3/z, [x13, x8]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1sb { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: 2 6 0.67 * ld1sb { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: 4 6 1.25 * ld1sb { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * ld1sb { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z31.h }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1sb { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 4 7 1.33 * ld1sb { z31.s }, p7/z, [z31.s, #31]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 4 7 1.33 * ld1sh { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * ld1sh { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z0.s }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ld1sh { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z21.s }, p5/z, [sp, x21, lsl #1]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z21.s }, p5/z, [x10, x21, lsl #1]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z23.d }, p3/z, [x13, x8, lsl #1]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: 2 6 0.67 * ld1sh { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * ld1sh { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: 1 6 0.33 * ld1sh { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 4 7 1.33 * ld1sh { z31.s }, p7/z, [sp, z31.s, sxtw #1]
# CHECK-NEXT: 4 7 1.33 * ld1sh { z31.s }, p7/z, [sp, z31.s, uxtw #1]
# CHECK-NEXT: 4 7 1.33 * ld1sh { z31.s }, p7/z, [z31.s, #62]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
# CHECK-NEXT: 1 6 0.33 * ld1sw { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * ld1sw { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1sw { z23.d }, p3/z, [sp, x8, lsl #2]
# CHECK-NEXT: 1 6 0.33 * ld1sw { z23.d }, p3/z, [x13, x8, lsl #2]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
# CHECK-NEXT: 1 6 0.33 * ld1sw { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: 2 6 0.67 * ld1sw { z31.d }, p7/z, [z31.d, #124]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * ld1sw { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
# CHECK-NEXT: 1 6 0.33 * ld1w { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 4 7 1.33 * ld1w { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * ld1w { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1w { z0.s }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ld1w { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: 1 6 0.33 * ld1w { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * ld1w { z21.s }, p5/z, [sp, x21, lsl #2]
# CHECK-NEXT: 1 6 0.33 * ld1w { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: 1 6 0.33 * ld1w { z21.s }, p5/z, [x10, x21, lsl #2]
# CHECK-NEXT: 1 6 0.33 * ld1w { z23.d }, p3/z, [x13, x8, lsl #2]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
# CHECK-NEXT: 1 6 0.33 * ld1w { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: 2 6 0.67 * ld1w { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * ld1w { z31.d }, p7/z, [z31.d, #124]
# CHECK-NEXT: 1 6 0.33 * ld1w { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 4 7 1.33 * ld1w { z31.s }, p7/z, [sp, z31.s, sxtw #2]
# CHECK-NEXT: 4 7 1.33 * ld1w { z31.s }, p7/z, [sp, z31.s, uxtw #2]
@@ -4665,108 +4665,108 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: 9 12 2.50 * ld4w { z23.s - z26.s }, p3/z, [x13, #-32, mul vl]
# CHECK-NEXT: 14 13 2.50 * ld4w { z5.s - z8.s }, p3/z, [x17, x16, lsl #2]
# CHECK-NEXT: 1 6 0.33 * U ldff1b { z0.d }, p0/z, [x0, x0]
-# CHECK-NEXT: 2 6 0.67 * U ldff1b { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1b { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1b { z0.h }, p0/z, [x0, x0]
# CHECK-NEXT: 1 6 0.33 * U ldff1b { z0.s }, p0/z, [x0, x0]
# CHECK-NEXT: 4 7 1.33 * U ldff1b { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1b { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1b { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: 2 6 0.67 * U ldff1b { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1b { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1b { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1b { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: 1 6 0.33 * U ldff1b { z31.b }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1b { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1b { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1b { z31.d }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1b { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: 4 6 1.25 * U ldff1b { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: 1 6 0.33 * U ldff1b { z31.h }, p7/z, [sp]
# CHECK-NEXT: 1 6 0.33 * U ldff1b { z31.s }, p7/z, [sp]
# CHECK-NEXT: 4 7 1.33 * U ldff1b { z31.s }, p7/z, [z31.s, #31]
# CHECK-NEXT: 1 6 0.33 * U ldff1d { z0.d }, p0/z, [x0, x0, lsl #3]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z0.d }, p0/z, [z0.d]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1d { z31.d }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1d { z31.d }, p7/z, [z31.d, #248]
+# CHECK-NEXT: 4 6 1.25 * U ldff1d { z31.d }, p7/z, [z31.d, #248]
# CHECK-NEXT: 1 6 0.33 * U ldff1h { z0.d }, p0/z, [x0, x0, lsl #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1h { z0.h }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: 1 6 0.33 * U ldff1h { z0.s }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: 4 7 1.33 * U ldff1h { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1h { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1h { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1h { z31.d }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1h { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: 4 6 1.25 * U ldff1h { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: 1 6 0.33 * U ldff1h { z31.h }, p7/z, [sp]
# CHECK-NEXT: 4 7 1.33 * U ldff1h { z31.s }, p7/z, [sp, z31.s, sxtw #1]
# CHECK-NEXT: 4 7 1.33 * U ldff1h { z31.s }, p7/z, [sp, z31.s, uxtw #1]
# CHECK-NEXT: 1 6 0.33 * U ldff1h { z31.s }, p7/z, [sp]
# CHECK-NEXT: 4 7 1.33 * U ldff1h { z31.s }, p7/z, [z31.s, #62]
# CHECK-NEXT: 1 6 0.33 * U ldff1sb { z0.d }, p0/z, [x0, x0]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sb { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sb { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1sb { z0.h }, p0/z, [x0, x0]
# CHECK-NEXT: 1 6 0.33 * U ldff1sb { z0.s }, p0/z, [x0, x0]
# CHECK-NEXT: 4 7 1.33 * U ldff1sb { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1sb { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1sb { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sb { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sb { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1sb { z31.d }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sb { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sb { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: 1 6 0.33 * U ldff1sb { z31.h }, p7/z, [sp]
# CHECK-NEXT: 1 6 0.33 * U ldff1sb { z31.s }, p7/z, [sp]
# CHECK-NEXT: 4 7 1.33 * U ldff1sb { z31.s }, p7/z, [z31.s, #31]
# CHECK-NEXT: 1 6 0.33 * U ldff1sh { z0.d }, p0/z, [x0, x0, lsl #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1sh { z0.s }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: 4 7 1.33 * U ldff1sh { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1sh { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1sh { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1sh { z31.d }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sh { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sh { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: 4 7 1.33 * U ldff1sh { z31.s }, p7/z, [sp, z31.s, sxtw #1]
# CHECK-NEXT: 4 7 1.33 * U ldff1sh { z31.s }, p7/z, [sp, z31.s, uxtw #1]
# CHECK-NEXT: 1 6 0.33 * U ldff1sh { z31.s }, p7/z, [sp]
# CHECK-NEXT: 4 7 1.33 * U ldff1sh { z31.s }, p7/z, [z31.s, #62]
# CHECK-NEXT: 1 6 0.33 * U ldff1sw { z0.d }, p0/z, [x0, x0, lsl #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z0.d }, p0/z, [z0.d]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1sw { z31.d }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1sw { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: 4 6 1.25 * U ldff1sw { z31.d }, p7/z, [z31.d, #124]
# CHECK-NEXT: 1 6 0.33 * U ldff1w { z0.d }, p0/z, [x0, x0, lsl #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1w { z0.s }, p0/z, [x0, x0, lsl #2]
# CHECK-NEXT: 4 7 1.33 * U ldff1w { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1w { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: 4 7 1.33 * U ldff1w { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: 1 6 0.33 * U ldff1w { z31.d }, p7/z, [sp]
-# CHECK-NEXT: 2 6 0.67 * U ldff1w { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: 4 6 1.25 * U ldff1w { z31.d }, p7/z, [z31.d, #124]
# CHECK-NEXT: 4 7 1.33 * U ldff1w { z31.s }, p7/z, [sp, z31.s, sxtw #2]
# CHECK-NEXT: 4 7 1.33 * U ldff1w { z31.s }, p7/z, [sp, z31.s, uxtw #2]
# CHECK-NEXT: 1 6 0.33 * U ldff1w { z31.s }, p7/z, [sp]
@@ -4821,54 +4821,54 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: 1 6 0.33 * U ldnf1w { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: 1 6 0.33 * ldnt1b { z0.b }, p0/z, [x0, x0]
# CHECK-NEXT: 1 6 0.33 * ldnt1b { z0.b }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1b { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1b { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1b { z0.s }, p0/z, [z1.s]
# CHECK-NEXT: 1 6 0.33 * ldnt1b { z21.b }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: 1 6 0.33 * ldnt1b { z23.b }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ldnt1b { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1b { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1b { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: 4 6 1.25 * ldnt1b { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1b { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: 4 7 1.33 * ldnt1b { z31.s }, p7/z, [z31.s]
# CHECK-NEXT: 1 6 0.33 * ldnt1d { z0.d }, p0/z, [x0, x0, lsl #3]
# CHECK-NEXT: 1 6 0.33 * ldnt1d { z0.d }, p0/z, [x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1d { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1d { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: 1 6 0.33 * ldnt1d { z21.d }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: 1 6 0.33 * ldnt1d { z23.d }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ldnt1d { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1d { z31.d }, p7/z, [z31.d]
-# CHECK-NEXT: 2 6 0.67 * ldnt1h { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1d { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: 4 6 1.25 * ldnt1d { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1h { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: 1 6 0.33 * ldnt1h { z0.h }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: 1 6 0.33 * ldnt1h { z0.h }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ldnt1h { z0.s }, p0/z, [z1.s]
# CHECK-NEXT: 1 6 0.33 * ldnt1h { z21.h }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: 1 6 0.33 * ldnt1h { z23.h }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ldnt1h { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1h { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1h { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: 4 6 1.25 * ldnt1h { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1h { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: 4 7 1.33 * ldnt1h { z31.s }, p7/z, [z31.s]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sb { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sb { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1sb { z0.s }, p0/z, [z1.s]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sb { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sb { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sb { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sb { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1sb { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: 4 7 1.33 * ldnt1sb { z31.s }, p7/z, [z31.s]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sh { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sh { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1sh { z0.s }, p0/z, [z1.s]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sh { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sh { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sh { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sh { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1sh { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: 4 7 1.33 * ldnt1sh { z31.s }, p7/z, [z31.s]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sw { z0.d }, p0/z, [z1.d]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sw { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1sw { z31.d }, p7/z, [z31.d]
-# CHECK-NEXT: 2 6 0.67 * ldnt1w { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sw { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sw { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: 4 6 1.25 * ldnt1sw { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1w { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: 1 6 0.33 * ldnt1w { z0.s }, p0/z, [x0, x0, lsl #2]
# CHECK-NEXT: 1 6 0.33 * ldnt1w { z0.s }, p0/z, [x0]
# CHECK-NEXT: 4 7 1.33 * ldnt1w { z0.s }, p0/z, [z1.s]
# CHECK-NEXT: 1 6 0.33 * ldnt1w { z21.s }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: 1 6 0.33 * ldnt1w { z23.s }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: 2 6 0.67 * ldnt1w { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: 2 6 0.67 * ldnt1w { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: 4 6 1.25 * ldnt1w { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: 4 6 1.25 * ldnt1w { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: 4 7 1.33 * ldnt1w { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: 4 7 1.33 * ldnt1w { z31.s }, p7/z, [z31.s]
# CHECK-NEXT: 2 7 0.50 * ldr p0, [x0]
@@ -6847,7 +6847,7 @@ zip2 z31.s, z31.s, z31.s
# CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0.0] [0.1] [1.0] [1.1] [2] [3.0] [3.1] [4] [5] [6.0] [6.1] [7] [8]
-# CHECK-NEXT: - - - - 332.67 481.67 481.67 292.50 224.50 88.50 88.50 1557.00 1400.00
+# CHECK-NEXT: - - - - 252.00 552.25 552.25 292.50 224.50 88.50 88.50 1708.25 1551.25
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0.0] [0.1] [1.0] [1.1] [2] [3.0] [3.1] [4] [5] [6.0] [6.1] [7] [8] Instructions:
@@ -7841,7 +7841,7 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z0.b }, p0/z, [x0, x0]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z0.b }, p0/z, [x0]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1b { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1b { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z0.h }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1b { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1b { z0.s }, p0/z, [x0, z0.s, uxtw]
@@ -7849,53 +7849,53 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1b { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z21.b }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1b { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1b { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1b { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1b { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z21.h }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z21.s }, p5/z, [x10, x21]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z23.d }, p3/z, [x13, x8]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z31.b }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1b { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1b { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1b { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1b { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z31.h }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1b { z31.s }, p7/z, [z31.s, #31]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1b { z5.h }, p3/z, [x17, x16]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1d { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1d { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1d { z23.d }, p3/z, [sp, x8, lsl #3]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1d { z23.d }, p3/z, [x13, x8, lsl #3]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1d { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1d { z31.d }, p7/z, [z31.d, #248]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1d { z31.d }, p7/z, [z31.d, #248]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z0.h }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1h { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1h { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z0.s }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1h { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z21.h }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z21.s }, p5/z, [x10, x21, lsl #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z23.d }, p3/z, [x13, x8, lsl #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1h { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1h { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z31.h }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1h { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1h { z31.s }, p7/z, [sp, z31.s, sxtw #1]
@@ -7956,7 +7956,7 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1rw { z31.d }, p7/z, [sp, #252]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1rw { z31.s }, p7/z, [sp, #252]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sb { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sb { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z0.h }, p0/z, [sp, x0]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z0.h }, p0/z, [x0, x0]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z0.h }, p0/z, [x0]
@@ -7964,73 +7964,73 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z0.s }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sb { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z21.h }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z21.s }, p5/z, [x10, x21]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z23.d }, p3/z, [x13, x8]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sb { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sb { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sb { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sb { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z31.h }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sb { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sb { z31.s }, p7/z, [z31.s, #31]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sh { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sh { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z0.s }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sh { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z21.s }, p5/z, [sp, x21, lsl #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z21.s }, p5/z, [x10, x21, lsl #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z23.d }, p3/z, [x13, x8, lsl #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sh { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sh { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sh { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sh { z31.s }, p7/z, [sp, z31.s, sxtw #1]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sh { z31.s }, p7/z, [sp, z31.s, uxtw #1]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1sh { z31.s }, p7/z, [z31.s, #62]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sw { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sw { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sw { z23.d }, p3/z, [sp, x8, lsl #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sw { z23.d }, p3/z, [x13, x8, lsl #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1sw { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1sw { z31.d }, p7/z, [z31.d, #124]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1sw { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1w { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1w { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z0.s }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1w { z0.s }, p0/z, [z0.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z21.d }, p5/z, [x10, #5, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z21.s }, p5/z, [sp, x21, lsl #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z21.s }, p5/z, [x10, #5, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z21.s }, p5/z, [x10, x21, lsl #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z23.d }, p3/z, [x13, x8, lsl #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z31.d }, p7/z, [sp, #-1, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z31.d }, p7/z, [sp, z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ld1w { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ld1w { z31.d }, p7/z, [z31.d, #124]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ld1w { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1w { z31.s }, p7/z, [sp, z31.s, sxtw #2]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ld1w { z31.s }, p7/z, [sp, z31.s, uxtw #2]
@@ -8096,108 +8096,108 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - 2.50 2.50 ld4w { z23.s - z26.s }, p3/z, [x13, #-32, mul vl]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 1.25 1.25 1.25 1.25 2.50 2.50 ld4w { z5.s - z8.s }, p3/z, [x17, x16, lsl #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1b { z0.d }, p0/z, [x0, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1b { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1b { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1b { z0.h }, p0/z, [x0, x0]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1b { z0.s }, p0/z, [x0, x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1b { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1b { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1b { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1b { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1b { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1b { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1b { z21.d }, p5/z, [x10, z21.d, uxtw]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1b { z31.b }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1b { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1b { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1b { z31.d }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1b { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1b { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1b { z31.h }, p7/z, [sp]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1b { z31.s }, p7/z, [sp]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1b { z31.s }, p7/z, [z31.s, #31]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1d { z0.d }, p0/z, [x0, x0, lsl #3]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z0.d }, p0/z, [z0.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z0.d }, p0/z, [x0, z0.d, sxtw #3]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z0.d }, p0/z, [x0, z0.d, uxtw #3]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z23.d }, p3/z, [x13, z8.d, lsl #3]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1d { z31.d }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1d { z31.d }, p7/z, [z31.d, #248]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1d { z31.d }, p7/z, [z31.d, #248]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1h { z0.d }, p0/z, [x0, x0, lsl #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1h { z0.h }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1h { z0.s }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1h { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1h { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1h { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1h { z31.d }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1h { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1h { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1h { z31.h }, p7/z, [sp]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1h { z31.s }, p7/z, [sp, z31.s, sxtw #1]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1h { z31.s }, p7/z, [sp, z31.s, uxtw #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1h { z31.s }, p7/z, [sp]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1h { z31.s }, p7/z, [z31.s, #62]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sb { z0.d }, p0/z, [x0, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sb { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sb { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sb { z0.h }, p0/z, [x0, x0]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sb { z0.s }, p0/z, [x0, x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sb { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sb { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sb { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sb { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sb { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sb { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sb { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sb { z31.d }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sb { z31.d }, p7/z, [z31.d, #31]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sb { z31.d }, p7/z, [z31.d, #31]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sb { z31.h }, p7/z, [sp]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sb { z31.s }, p7/z, [sp]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sb { z31.s }, p7/z, [z31.s, #31]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sh { z0.d }, p0/z, [x0, x0, lsl #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z0.d }, p0/z, [x0, z0.d, sxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z0.d }, p0/z, [x0, z0.d, uxtw #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sh { z0.s }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sh { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sh { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sh { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z23.d }, p3/z, [x13, z8.d, lsl #1]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sh { z31.d }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sh { z31.d }, p7/z, [z31.d, #62]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sh { z31.d }, p7/z, [z31.d, #62]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sh { z31.s }, p7/z, [sp, z31.s, sxtw #1]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sh { z31.s }, p7/z, [sp, z31.s, uxtw #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sh { z31.s }, p7/z, [sp]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1sh { z31.s }, p7/z, [z31.s, #62]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sw { z0.d }, p0/z, [x0, x0, lsl #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z0.d }, p0/z, [z0.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1sw { z31.d }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1sw { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1sw { z31.d }, p7/z, [z31.d, #124]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1w { z0.d }, p0/z, [x0, x0, lsl #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z0.d }, p0/z, [z0.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z0.d }, p0/z, [x0, z0.d, sxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z0.d }, p0/z, [x0, z0.d, uxtw #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z0.d }, p0/z, [z0.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1w { z0.s }, p0/z, [x0, x0, lsl #2]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1w { z0.s }, p0/z, [x0, z0.s, sxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1w { z0.s }, p0/z, [x0, z0.s, uxtw]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1w { z0.s }, p0/z, [z0.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z21.d }, p5/z, [x10, z21.d, sxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z21.d }, p5/z, [x10, z21.d, uxtw]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z31.d }, p7/z, [sp, z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z21.d }, p5/z, [x10, z21.d, sxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z21.d }, p5/z, [x10, z21.d, uxtw]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z23.d }, p3/z, [x13, z8.d, lsl #2]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z31.d }, p7/z, [sp, z31.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1w { z31.d }, p7/z, [sp]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldff1w { z31.d }, p7/z, [z31.d, #124]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldff1w { z31.d }, p7/z, [z31.d, #124]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1w { z31.s }, p7/z, [sp, z31.s, sxtw #2]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldff1w { z31.s }, p7/z, [sp, z31.s, uxtw #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldff1w { z31.s }, p7/z, [sp]
@@ -8252,54 +8252,54 @@ zip2 z31.s, z31.s, z31.s
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnf1w { z31.s }, p7/z, [sp, #-1, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1b { z0.b }, p0/z, [x0, x0]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1b { z0.b }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1b { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1b { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1b { z0.s }, p0/z, [z1.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1b { z21.b }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1b { z23.b }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1b { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1b { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1b { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1b { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1b { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1b { z31.s }, p7/z, [z31.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1d { z0.d }, p0/z, [x0, x0, lsl #3]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1d { z0.d }, p0/z, [x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1d { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1d { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1d { z21.d }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1d { z23.d }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1d { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1d { z31.d }, p7/z, [z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1h { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1d { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1d { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1h { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1h { z0.h }, p0/z, [x0, x0, lsl #1]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1h { z0.h }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1h { z0.s }, p0/z, [z1.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1h { z21.h }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1h { z23.h }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1h { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1h { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1h { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1h { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1h { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1h { z31.s }, p7/z, [z31.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sb { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sb { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1sb { z0.s }, p0/z, [z1.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sb { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sb { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sb { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sb { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1sb { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1sb { z31.s }, p7/z, [z31.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sh { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sh { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1sh { z0.s }, p0/z, [z1.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sh { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sh { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sh { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sh { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1sh { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1sh { z31.s }, p7/z, [z31.s]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sw { z0.d }, p0/z, [z1.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sw { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1sw { z31.d }, p7/z, [z31.d]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1w { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sw { z0.d }, p0/z, [z1.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sw { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1sw { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1w { z0.d }, p0/z, [z1.d]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1w { z0.s }, p0/z, [x0, x0, lsl #2]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1w { z0.s }, p0/z, [x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1w { z0.s }, p0/z, [z1.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1w { z21.s }, p5/z, [x10, #7, mul vl]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 - - - - - - ldnt1w { z23.s }, p3/z, [x13, #-8, mul vl]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1w { z31.d }, p7/z, [z31.d, x0]
-# CHECK-NEXT: - - - - 0.67 0.67 0.67 - - - - - - ldnt1w { z31.d }, p7/z, [z31.d]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1w { z31.d }, p7/z, [z31.d, x0]
+# CHECK-NEXT: - - - - - 1.25 1.25 - - - - 1.25 1.25 ldnt1w { z31.d }, p7/z, [z31.d]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1w { z31.s }, p7/z, [z31.s, x0]
# CHECK-NEXT: - - - - 1.33 1.33 1.33 - - - - - - ldnt1w { z31.s }, p7/z, [z31.s]
# CHECK-NEXT: - - - - 0.33 0.33 0.33 0.50 0.50 - - - - ldr p0, [x0]
More information about the llvm-commits
mailing list