AMDGPU/GlobalISel: Fix bit count ops for non-power-of-2 types

llvm-svn: 354587
This commit is contained in:
Matt Arsenault 2019-02-21 15:22:20 +00:00
parent 411a55f79b
commit b10fa8df3f
6 changed files with 273 additions and 1 deletions

View File

@ -307,7 +307,9 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
.legalFor({{S32, S32}, {S32, S64}})
.clampScalar(0, S32, S32)
.clampScalar(1, S32, S64)
.scalarize(0);
.scalarize(0)
.widenScalarToNextPow2(0, 32)
.widenScalarToNextPow2(1, 32);
// TODO: Expand for > s32
getActionDefinitionsBuilder(G_BSWAP)

View File

@ -155,3 +155,61 @@ body: |
%1:_(<2 x s16>) = G_CTLZ_ZERO_UNDEF %0
$vgpr0 = COPY %1
...
---
name: ctlz_zero_undef_s7_s7
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: ctlz_zero_undef_s7_s7
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[AND]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 25
; CHECK: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[CTLZ_ZERO_UNDEF]], [[C1]]
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SUB]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C2]]
; CHECK: $vgpr0 = COPY [[AND1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s7) = G_TRUNC %0
%2:_(s7) = G_CTLZ_ZERO_UNDEF %1
%3:_(s32) = G_ZEXT %2
$vgpr0 = COPY %3
...
---
name: ctlz_zero_undef_s33_s33
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: ctlz_zero_undef_s33_s33
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 8589934591
; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY [[COPY]](s64)
; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[AND]](s64)
; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[CTLZ_ZERO_UNDEF]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 31
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ZEXT]](s64)
; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[C1]](s64)
; CHECK: [[USUBO:%[0-9]+]]:_(s32), [[USUBO1:%[0-9]+]]:_(s1) = G_USUBO [[UV]], [[UV2]]
; CHECK: [[USUBE:%[0-9]+]]:_(s32), [[USUBE1:%[0-9]+]]:_(s1) = G_USUBE [[UV1]], [[UV3]], [[USUBO1]]
; CHECK: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[USUBO]](s32), [[USUBE]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967295
; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY [[MV]](s64)
; CHECK: [[AND1:%[0-9]+]]:_(s64) = G_AND [[COPY2]], [[C2]]
; CHECK: $vgpr0_vgpr1 = COPY [[AND1]](s64)
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s33) = G_TRUNC %0
%2:_(s33) = G_CTLZ_ZERO_UNDEF %1
%3:_(s64) = G_ANYEXT %2
$vgpr0_vgpr1 = COPY %3
...

View File

@ -155,3 +155,61 @@ body: |
%1:_(<2 x s16>) = G_CTLZ %0
$vgpr0 = COPY %1
...
---
name: ctlz_s7_s7
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: ctlz_s7_s7
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTLZ:%[0-9]+]]:_(s32) = G_CTLZ [[AND]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 25
; CHECK: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[CTLZ]], [[C1]]
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SUB]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C2]]
; CHECK: $vgpr0 = COPY [[AND1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s7) = G_TRUNC %0
%2:_(s7) = G_CTLZ %1
%3:_(s32) = G_ZEXT %2
$vgpr0 = COPY %3
...
---
name: ctlz_s33_s33
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: ctlz_s33_s33
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 8589934591
; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY [[COPY]](s64)
; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTLZ:%[0-9]+]]:_(s32) = G_CTLZ [[AND]](s64)
; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[CTLZ]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 31
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ZEXT]](s64)
; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[C1]](s64)
; CHECK: [[USUBO:%[0-9]+]]:_(s32), [[USUBO1:%[0-9]+]]:_(s1) = G_USUBO [[UV]], [[UV2]]
; CHECK: [[USUBE:%[0-9]+]]:_(s32), [[USUBE1:%[0-9]+]]:_(s1) = G_USUBE [[UV1]], [[UV3]], [[USUBO1]]
; CHECK: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[USUBO]](s32), [[USUBE]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967295
; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY [[MV]](s64)
; CHECK: [[AND1:%[0-9]+]]:_(s64) = G_AND [[COPY2]], [[C2]]
; CHECK: $vgpr0_vgpr1 = COPY [[AND1]](s64)
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s33) = G_TRUNC %0
%2:_(s33) = G_CTLZ %1
%3:_(s64) = G_ANYEXT %2
$vgpr0_vgpr1 = COPY %3
...

View File

@ -149,3 +149,53 @@ body: |
%1:_(<2 x s16>) = G_CTPOP %0
$vgpr0 = COPY %1
...
---
name: ctpop_s7_s7
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: ctpop_s7_s7
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTPOP:%[0-9]+]]:_(s32) = G_CTPOP [[AND]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[CTPOP]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
; CHECK: $vgpr0 = COPY [[AND1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s7) = G_TRUNC %0
%2:_(s7) = G_CTPOP %1
%3:_(s32) = G_ZEXT %2
$vgpr0 = COPY %3
...
---
name: ctpop_s33_s33
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: ctpop_s33_s33
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 8589934591
; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY [[COPY]](s64)
; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTPOP:%[0-9]+]]:_(s32) = G_CTPOP [[AND]](s64)
; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[CTPOP]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967295
; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY [[ZEXT]](s64)
; CHECK: [[AND1:%[0-9]+]]:_(s64) = G_AND [[COPY2]], [[C1]]
; CHECK: $vgpr0_vgpr1 = COPY [[AND1]](s64)
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s33) = G_TRUNC %0
%2:_(s33) = G_CTPOP %1
%3:_(s64) = G_ANYEXT %2
$vgpr0_vgpr1 = COPY %3
...

View File

@ -149,3 +149,53 @@ body: |
%1:_(<2 x s16>) = G_CTTZ_ZERO_UNDEF %0
$vgpr0 = COPY %1
...
---
name: cttz_zero_undef_s7_s7
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: cttz_zero_undef_s7_s7
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTTZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[AND]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[CTTZ_ZERO_UNDEF]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
; CHECK: $vgpr0 = COPY [[AND1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s7) = G_TRUNC %0
%2:_(s7) = G_CTTZ_ZERO_UNDEF %1
%3:_(s32) = G_ZEXT %2
$vgpr0 = COPY %3
...
---
name: cttz_zero_undef_s33_s33
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: cttz_zero_undef_s33_s33
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 8589934591
; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY [[COPY]](s64)
; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY1]], [[C]]
; CHECK: [[CTTZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[AND]](s64)
; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[CTTZ_ZERO_UNDEF]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967295
; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY [[ZEXT]](s64)
; CHECK: [[AND1:%[0-9]+]]:_(s64) = G_AND [[COPY2]], [[C1]]
; CHECK: $vgpr0_vgpr1 = COPY [[AND1]](s64)
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s33) = G_TRUNC %0
%2:_(s33) = G_CTTZ_ZERO_UNDEF %1
%3:_(s64) = G_ANYEXT %2
$vgpr0_vgpr1 = COPY %3
...

View File

@ -155,3 +155,57 @@ body: |
%1:_(<2 x s16>) = G_CTTZ %0
$vgpr0 = COPY %1
...
---
name: cttz_s7_s7
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: cttz_s7_s7
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 128
; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[C1]]
; CHECK: [[CTTZ:%[0-9]+]]:_(s32) = G_CTTZ [[OR]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[CTTZ]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 127
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C2]]
; CHECK: $vgpr0 = COPY [[AND1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s7) = G_TRUNC %0
%2:_(s7) = G_CTTZ %1
%3:_(s32) = G_ZEXT %2
$vgpr0 = COPY %3
...
---
name: cttz_s33_s33
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: cttz_s33_s33
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 8589934591
; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY [[COPY]](s64)
; CHECK: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY1]], [[C]]
; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8589934592
; CHECK: [[OR:%[0-9]+]]:_(s64) = G_OR [[AND]], [[C1]]
; CHECK: [[CTTZ:%[0-9]+]]:_(s32) = G_CTTZ [[OR]](s64)
; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[CTTZ]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967295
; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY [[ZEXT]](s64)
; CHECK: [[AND1:%[0-9]+]]:_(s64) = G_AND [[COPY2]], [[C2]]
; CHECK: $vgpr0_vgpr1 = COPY [[AND1]](s64)
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s33) = G_TRUNC %0
%2:_(s33) = G_CTTZ %1
%3:_(s64) = G_ANYEXT %2
$vgpr0_vgpr1 = COPY %3
...