Skip to content

Commit 4e29c6a

Browse files
authored
[AArch64] Correct Neoverse V1 SVE 16-bit sdot/udot schedule pipelines. (#86142)
Fixes #86102
1 parent aeeb7d5 commit 4e29c6a

File tree

2 files changed

+10
-10
lines changed

2 files changed

+10
-10
lines changed

llvm/lib/Target/AArch64/AArch64SchedNeoverseV1.td

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1446,7 +1446,7 @@ def : InstRW<[V1Write_3c_1V01], (instregex "^[SU]DOT_ZZZI?_S$")>;
14461446
def : InstRW<[V1Write_3c_1V], (instrs SUDOT_ZZZI, USDOT_ZZZ, USDOT_ZZZI)>;
14471447

14481448
// Dot product, 16 bit
1449-
def : InstRW<[V1Write_4c_1V01], (instregex "^[SU]DOT_ZZZI?_D$")>;
1449+
def : InstRW<[V1Write_4c_1V0], (instregex "^[SU]DOT_ZZZI?_D$")>;
14501450

14511451
// Duplicate, immediate and indexed form
14521452
def : InstRW<[V1Write_2c_1V01], (instregex "^DUP_ZI_[BHSD]$",

llvm/test/tools/llvm-mca/AArch64/Neoverse/V1-sve-instructions.s

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -4093,8 +4093,8 @@ zip2 z31.s, z31.s, z31.s
40934093
# CHECK-NEXT: 1 12 7.00 sdiv z0.s, p7/m, z0.s, z31.s
40944094
# CHECK-NEXT: 1 20 7.00 sdivr z0.d, p7/m, z0.d, z31.d
40954095
# CHECK-NEXT: 1 12 7.00 sdivr z0.s, p7/m, z0.s, z31.s
4096-
# CHECK-NEXT: 1 4 0.50 sdot z0.d, z1.h, z15.h[1]
4097-
# CHECK-NEXT: 1 4 0.50 sdot z0.d, z1.h, z31.h
4096+
# CHECK-NEXT: 1 4 1.00 sdot z0.d, z1.h, z15.h[1]
4097+
# CHECK-NEXT: 1 4 1.00 sdot z0.d, z1.h, z31.h
40984098
# CHECK-NEXT: 1 3 0.50 sdot z0.s, z1.b, z31.b
40994099
# CHECK-NEXT: 1 3 0.50 sdot z0.s, z1.b, z7.b[3]
41004100
# CHECK-NEXT: 1 2 0.50 sel z23.b, p11, z13.b, z8.b
@@ -4569,8 +4569,8 @@ zip2 z31.s, z31.s, z31.s
45694569
# CHECK-NEXT: 1 12 7.00 udiv z0.s, p7/m, z0.s, z31.s
45704570
# CHECK-NEXT: 1 20 7.00 udivr z0.d, p7/m, z0.d, z31.d
45714571
# CHECK-NEXT: 1 12 7.00 udivr z0.s, p7/m, z0.s, z31.s
4572-
# CHECK-NEXT: 1 4 0.50 udot z0.d, z1.h, z15.h[1]
4573-
# CHECK-NEXT: 1 4 0.50 udot z0.d, z1.h, z31.h
4572+
# CHECK-NEXT: 1 4 1.00 udot z0.d, z1.h, z15.h[1]
4573+
# CHECK-NEXT: 1 4 1.00 udot z0.d, z1.h, z31.h
45744574
# CHECK-NEXT: 1 3 0.50 udot z0.s, z1.b, z31.b
45754575
# CHECK-NEXT: 1 3 0.50 udot z0.s, z1.b, z7.b[3]
45764576
# CHECK-NEXT: 1 2 0.50 umax z0.b, z0.b, #0
@@ -4839,7 +4839,7 @@ zip2 z31.s, z31.s, z31.s
48394839

48404840
# CHECK: Resource pressure per iteration:
48414841
# CHECK-NEXT: [0.0] [0.1] [1.0] [1.1] [2] [3.0] [3.1] [4] [5] [6.0] [6.1] [7] [8] [9] [10]
4842-
# CHECK-NEXT: - - - - 88.67 500.67 500.67 797.50 2.50 92.50 92.50 1250.00 923.00 178.50 181.50
4842+
# CHECK-NEXT: - - - - 88.67 500.67 500.67 797.50 2.50 92.50 92.50 1252.00 921.00 178.50 181.50
48434843

48444844
# CHECK: Resource pressure by instruction:
48454845
# CHECK-NEXT: [0.0] [0.1] [1.0] [1.1] [2] [3.0] [3.1] [4] [5] [6.0] [6.1] [7] [8] [9] [10] Instructions:
@@ -6521,8 +6521,8 @@ zip2 z31.s, z31.s, z31.s
65216521
# CHECK-NEXT: - - - - - - - - - - - 7.00 - - - sdiv z0.s, p7/m, z0.s, z31.s
65226522
# CHECK-NEXT: - - - - - - - - - - - 7.00 - - - sdivr z0.d, p7/m, z0.d, z31.d
65236523
# CHECK-NEXT: - - - - - - - - - - - 7.00 - - - sdivr z0.s, p7/m, z0.s, z31.s
6524-
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - sdot z0.d, z1.h, z15.h[1]
6525-
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - sdot z0.d, z1.h, z31.h
6524+
# CHECK-NEXT: - - - - - - - - - - - 1.00 - - - sdot z0.d, z1.h, z15.h[1]
6525+
# CHECK-NEXT: - - - - - - - - - - - 1.00 - - - sdot z0.d, z1.h, z31.h
65266526
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - sdot z0.s, z1.b, z31.b
65276527
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - sdot z0.s, z1.b, z7.b[3]
65286528
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - sel z23.b, p11, z13.b, z8.b
@@ -6997,8 +6997,8 @@ zip2 z31.s, z31.s, z31.s
69976997
# CHECK-NEXT: - - - - - - - - - - - 7.00 - - - udiv z0.s, p7/m, z0.s, z31.s
69986998
# CHECK-NEXT: - - - - - - - - - - - 7.00 - - - udivr z0.d, p7/m, z0.d, z31.d
69996999
# CHECK-NEXT: - - - - - - - - - - - 7.00 - - - udivr z0.s, p7/m, z0.s, z31.s
7000-
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - udot z0.d, z1.h, z15.h[1]
7001-
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - udot z0.d, z1.h, z31.h
7000+
# CHECK-NEXT: - - - - - - - - - - - 1.00 - - - udot z0.d, z1.h, z15.h[1]
7001+
# CHECK-NEXT: - - - - - - - - - - - 1.00 - - - udot z0.d, z1.h, z31.h
70027002
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - udot z0.s, z1.b, z31.b
70037003
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - udot z0.s, z1.b, z7.b[3]
70047004
# CHECK-NEXT: - - - - - - - - - - - 0.50 0.50 - - umax z0.b, z0.b, #0

0 commit comments

Comments
 (0)