[llvm] 90b83a6 - [SVE] Add isel for 32-bit add/sub(cntp()) -> incp/decp.
Paul Walker via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 6 07:26:22 PDT 2023
Author: Paul Walker
Date: 2023-07-06T14:25:18Z
New Revision: 90b83a6d6caa651ea32987c94955ed18fffcb40c
URL: https://github.com/llvm/llvm-project/commit/90b83a6d6caa651ea32987c94955ed18fffcb40c
DIFF: https://github.com/llvm/llvm-project/commit/90b83a6d6caa651ea32987c94955ed18fffcb40c.diff
LOG: [SVE] Add isel for 32-bit add/sub(cntp()) -> incp/decp.
Patterns already exist for 64-bit that I've simply copied and
converted to include the necessary truncation.
Differential Revision: https://reviews.llvm.org/D154350
Added:
Modified:
llvm/lib/Target/AArch64/SVEInstrFormats.td
llvm/test/CodeGen/AArch64/sve-cntp-combine-i32.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/SVEInstrFormats.td b/llvm/lib/Target/AArch64/SVEInstrFormats.td
index e7f8eb4202e45c..118862b8c317cb 100644
--- a/llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -953,6 +953,42 @@ multiclass sve_int_count_r_x64<bits<5> opc, string asm,
(!cast<Instruction>(NAME # _S) PPRAny:$pred, $Rn)>;
def : Pat<(i64 (combine_op GPR64:$Rn, (int_aarch64_sve_cntp_oneuse (nxv2i1 PPRAny:$pred), (nxv2i1 PPRAny:$pred)))),
(!cast<Instruction>(NAME # _D) PPRAny:$pred, $Rn)>;
+
+ // combine_op(x, trunc(cntp(all_active, p))) ==> inst p, x
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv16i1 (SVEAllActive)), (nxv16i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _B) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv8i1 (SVEAllActive)), (nxv8i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _H) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv4i1 (SVEAllActive)), (nxv4i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _S) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv2i1 (SVEAllActive)), (nxv2i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _D) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
+
+ // combine_op(x, trunc(cntp(p, p))) ==> inst p, x
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv16i1 PPRAny:$pred), (nxv16i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _B) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv8i1 PPRAny:$pred), (nxv8i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _H) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv4i1 PPRAny:$pred), (nxv4i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _S) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
+ def : Pat<(i32 (combine_op GPR32:$Rn, (trunc (int_aarch64_sve_cntp_oneuse (nxv2i1 PPRAny:$pred), (nxv2i1 PPRAny:$pred))))),
+ (i32 (EXTRACT_SUBREG (!cast<Instruction>(NAME # _D) PPRAny:$pred,
+ (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32)),
+ sub_32))>;
}
class sve_int_count_v<bits<2> sz8_64, bits<5> opc, string asm,
diff --git a/llvm/test/CodeGen/AArch64/sve-cntp-combine-i32.ll b/llvm/test/CodeGen/AArch64/sve-cntp-combine-i32.ll
index 859c0ffad4c72a..d04dc500287653 100644
--- a/llvm/test/CodeGen/AArch64/sve-cntp-combine-i32.ll
+++ b/llvm/test/CodeGen/AArch64/sve-cntp-combine-i32.ll
@@ -8,9 +8,9 @@ target triple = "aarch64-unknown-linux-gnu"
define i32 @cntp_add_all_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_all_active_nxv16i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.b
-; CHECK-NEXT: cntp x8, p1, p0.b
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.b
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv16i1(<vscale x 16 x i1> %1, <vscale x 16 x i1> %pg)
@@ -22,9 +22,9 @@ define i32 @cntp_add_all_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
define i32 @cntp_add_all_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_all_active_nxv8i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.h
-; CHECK-NEXT: cntp x8, p1, p0.h
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.h
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv8i1(<vscale x 8 x i1> %1, <vscale x 8 x i1> %pg)
@@ -36,9 +36,9 @@ define i32 @cntp_add_all_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
define i32 @cntp_add_all_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_all_active_nxv4i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.s
-; CHECK-NEXT: cntp x8, p1, p0.s
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.s
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %pg)
@@ -50,9 +50,9 @@ define i32 @cntp_add_all_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
define i32 @cntp_add_all_active_nxv2i1(i32 %x, <vscale x 2 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_all_active_nxv2i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.d
-; CHECK-NEXT: cntp x8, p1, p0.d
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.d
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv2i1(<vscale x 2 x i1> %1, <vscale x 2 x i1> %pg)
@@ -64,9 +64,9 @@ define i32 @cntp_add_all_active_nxv2i1(i32 %x, <vscale x 2 x i1> %pg) #0 {
define i32 @cntp_add_all_active_nxv8i1_via_cast(i32 %x, <vscale x 8 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_all_active_nxv8i1_via_cast:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.b
-; CHECK-NEXT: cntp x8, p1, p0.h
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.h
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
%2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %1)
@@ -97,8 +97,9 @@ define i64 @cntp_add_all_active_nxv2i1_multiuse(i32 %x, i64 %y, <vscale x 2 x i1
define i32 @cntp_add_same_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_same_active_nxv16i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.b
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.b
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %pg)
%2 = trunc i64 %1 to i32
@@ -109,8 +110,9 @@ define i32 @cntp_add_same_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
define i32 @cntp_add_same_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_same_active_nxv8i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.h
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.h
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv8i1(<vscale x 8 x i1> %pg, <vscale x 8 x i1> %pg)
%2 = trunc i64 %1 to i32
@@ -121,8 +123,9 @@ define i32 @cntp_add_same_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
define i32 @cntp_add_same_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_same_active_nxv4i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.s
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.s
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv4i1(<vscale x 4 x i1> %pg, <vscale x 4 x i1> %pg)
%2 = trunc i64 %1 to i32
@@ -133,8 +136,9 @@ define i32 @cntp_add_same_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
define i32 @cntp_add_same_active_nxv2i1(i32 %x, <vscale x 2 x i1> %pg) #0 {
; CHECK-LABEL: cntp_add_same_active_nxv2i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.d
-; CHECK-NEXT: add w0, w8, w0
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: incp x0, p0.d
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv2i1(<vscale x 2 x i1> %pg, <vscale x 2 x i1> %pg)
%2 = trunc i64 %1 to i32
@@ -163,9 +167,9 @@ define i64 @cntp_add_same_active_nxv2i1_multiuse(i32 %x, i64 %y, <vscale x 2 x i
define i32 @cntp_sub_all_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_all_active_nxv16i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.b
-; CHECK-NEXT: cntp x8, p1, p0.b
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.b
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv16i1(<vscale x 16 x i1> %1, <vscale x 16 x i1> %pg)
@@ -177,9 +181,9 @@ define i32 @cntp_sub_all_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
define i32 @cntp_sub_all_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_all_active_nxv8i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.h
-; CHECK-NEXT: cntp x8, p1, p0.h
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.h
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv8i1(<vscale x 8 x i1> %1, <vscale x 8 x i1> %pg)
@@ -191,9 +195,9 @@ define i32 @cntp_sub_all_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
define i32 @cntp_sub_all_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_all_active_nxv4i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.s
-; CHECK-NEXT: cntp x8, p1, p0.s
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.s
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %pg)
@@ -205,9 +209,9 @@ define i32 @cntp_sub_all_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
define i32 @cntp_sub_all_active_nxv2i1(i32 %x, <vscale x 2 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_all_active_nxv2i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.d
-; CHECK-NEXT: cntp x8, p1, p0.d
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.d
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)
%2 = tail call i64 @llvm.aarch64.sve.cntp.nxv2i1(<vscale x 2 x i1> %1, <vscale x 2 x i1> %pg)
@@ -219,9 +223,9 @@ define i32 @cntp_sub_all_active_nxv2i1(i32 %x, <vscale x 2 x i1> %pg) #0 {
define i32 @cntp_sub_all_active_nxv8i1_via_cast(i32 %x, <vscale x 8 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_all_active_nxv8i1_via_cast:
; CHECK: // %bb.0:
-; CHECK-NEXT: ptrue p1.b
-; CHECK-NEXT: cntp x8, p1, p0.h
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.h
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
%2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %1)
@@ -252,8 +256,9 @@ define i64 @cntp_sub_all_active_nxv2i1_multiuse(i32 %x, i64 %y, <vscale x 2 x i1
define i32 @cntp_sub_same_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_same_active_nxv16i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.b
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.b
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %pg)
%2 = trunc i64 %1 to i32
@@ -264,8 +269,9 @@ define i32 @cntp_sub_same_active_nxv16i1(i32 %x, <vscale x 16 x i1> %pg) #0 {
define i32 @cntp_sub_same_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_same_active_nxv8i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.h
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.h
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv8i1(<vscale x 8 x i1> %pg, <vscale x 8 x i1> %pg)
%2 = trunc i64 %1 to i32
@@ -276,8 +282,9 @@ define i32 @cntp_sub_same_active_nxv8i1(i32 %x, <vscale x 8 x i1> %pg) #0 {
define i32 @cntp_sub_same_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_same_active_nxv4i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.s
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.s
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv4i1(<vscale x 4 x i1> %pg, <vscale x 4 x i1> %pg)
%2 = trunc i64 %1 to i32
@@ -288,8 +295,9 @@ define i32 @cntp_sub_same_active_nxv4i1(i32 %x, <vscale x 4 x i1> %pg) #0 {
define i32 @cntp_sub_same_active_nxv2i1(i32 %x, <vscale x 2 x i1> %pg) #0 {
; CHECK-LABEL: cntp_sub_same_active_nxv2i1:
; CHECK: // %bb.0:
-; CHECK-NEXT: cntp x8, p0, p0.d
-; CHECK-NEXT: sub w0, w0, w8
+; CHECK-NEXT: // kill: def $w0 killed $w0 def $x0
+; CHECK-NEXT: decp x0, p0.d
+; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%1 = tail call i64 @llvm.aarch64.sve.cntp.nxv2i1(<vscale x 2 x i1> %pg, <vscale x 2 x i1> %pg)
%2 = trunc i64 %1 to i32
More information about the llvm-commits
mailing list