Here is the same version again with an extra test added:
The popcount expansion uses SIMD instructions acting on 64-bit values.
As a result a popcount of a 32-bit integer requires zero-extension before
moving the zero-extended value into an FP register. This patch adds
support for zero-extended int->FP moves to avoid the redundant uxtw.
Similarly, add support for 32-bit zero-extending load->FP register
and 32-bit zero-extending FP->FP and FP->int moves.
Add a missing 'fp' arch attribute to the related 8/16-bit pattern and
fix an incorrect type attribute.
To complete zero-extended load support, add a new alternative to
load_pair_zero_extendsidi2_aarch64 to support LDP into FP registers too.
int f (int a)
{
return __builtin_popcount (a);
}
Before:
uxtw x0, w0
fmov d0, x0
cnt v0.8b, v0.8b
addv b0, v0.8b
fmov w0, s0
ret
After:
fmov s0, w0
cnt v0.8b, v0.8b
addv b0, v0.8b
fmov w0, s0
ret
Passes regress on AArch64, OK for commit?
ChangeLog:
2018-10-11 Wilco Dijkstra <wdijk...@arm.com>
gcc/
* config/aarch64/aarch64.md (zero_extendsidi2_aarch64): Add alternatives
to zero-extend between int and floating-point registers.
(load_pair_zero_extendsidi2_aarch64): Add alternative to emit
zero-extended
ldp into floating-point registers. Add type and arch attributes.
(zero_extend<SHORT:mode><GPI:mode>2_aarch64): Add arch attribute.
Use f_loads for type attribute.
testsuite/
* gcc.target/aarch64/popcnt.c: Test zero-extended popcount.
* gcc.target/aarch64/vec_zeroextend.c: Test zero-extended vectors.
--
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index
ef2368706e88a551b9d0d2db2385860112bdbdde..5e42485a4727079cc7647f4ded45dd175030b3fb
100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -1588,26 +1588,34 @@ (define_insn "*load_pair_extendsidi2_aarch64"
)
(define_insn "*zero_extendsidi2_aarch64"
- [(set (match_operand:DI 0 "register_operand" "=r,r")
- (zero_extend:DI (match_operand:SI 1 "nonimmediate_operand" "r,m")))]
+ [(set (match_operand:DI 0 "register_operand" "=r,r,w,w,r,w")
+ (zero_extend:DI (match_operand:SI 1 "nonimmediate_operand"
"r,m,r,m,w,w")))]
""
"@
uxtw\t%0, %w1
- ldr\t%w0, %1"
- [(set_attr "type" "extend,load_4")]
+ ldr\t%w0, %1
+ fmov\t%s0, %w1
+ ldr\t%s0, %1
+ fmov\t%w0, %s1
+ fmov\t%s0, %s1"
+ [(set_attr "type" "extend,load_4,f_mcr,f_loads,f_mrc,fmov")
+ (set_attr "arch" "*,*,fp,fp,fp,fp")]
)
(define_insn "*load_pair_zero_extendsidi2_aarch64"
- [(set (match_operand:DI 0 "register_operand" "=r")
- (zero_extend:DI (match_operand:SI 1 "aarch64_mem_pair_operand" "Ump")))
- (set (match_operand:DI 2 "register_operand" "=r")
- (zero_extend:DI (match_operand:SI 3 "memory_operand" "m")))]
+ [(set (match_operand:DI 0 "register_operand" "=r,w")
+ (zero_extend:DI (match_operand:SI 1 "aarch64_mem_pair_operand"
"Ump,Ump")))
+ (set (match_operand:DI 2 "register_operand" "=r,w")
+ (zero_extend:DI (match_operand:SI 3 "memory_operand" "m,m")))]
"rtx_equal_p (XEXP (operands[3], 0),
plus_constant (Pmode,
XEXP (operands[1], 0),
GET_MODE_SIZE (SImode)))"
- "ldp\\t%w0, %w2, %1"
- [(set_attr "type" "load_8")]
+ "@
+ ldp\t%w0, %w2, %1
+ ldp\t%s0, %s2, %1"
+ [(set_attr "type" "load_8,neon_load1_2reg")
+ (set_attr "arch" "*,fp")]
)
(define_expand "<ANY_EXTEND:optab><SHORT:mode><GPI:mode>2"
@@ -1634,7 +1642,8 @@ (define_insn "*zero_extend<SHORT:mode><GPI:mode>2_aarch64"
and\t%<GPI:w>0, %<GPI:w>1, <SHORT:short_mask>
ldr<SHORT:size>\t%w0, %1
ldr\t%<SHORT:size>0, %1"
- [(set_attr "type" "logic_imm,load_4,load_4")]
+ [(set_attr "type" "logic_imm,load_4,f_loads")
+ (set_attr "arch" "*,*,fp")]
)
(define_expand "<optab>qihi2"
diff --git a/gcc/testsuite/gcc.target/aarch64/popcnt.c
b/gcc/testsuite/gcc.target/aarch64/popcnt.c
index
7e957966d8e81b8633a444bb42944d0da82ae5db..2b5e9f3e2c0245438ed7bcc5d0d4e01efe01b1ee
100644
--- a/gcc/testsuite/gcc.target/aarch64/popcnt.c
+++ b/gcc/testsuite/gcc.target/aarch64/popcnt.c
@@ -19,5 +19,16 @@ foo2 (long long x)
return __builtin_popcountll (x);
}
-/* { dg-final { scan-assembler-not "popcount" } } */
-/* { dg-final { scan-assembler-times "cnt\t" 3 } } */
+int
+foo3 (int *p)
+{
+ return __builtin_popcount (*p);
+}
+
+/* { dg-final { scan-assembler-not {popcount} } } */
+/* { dg-final { scan-assembler-times {cnt\t} 4 } } */
+/* { dg-final { scan-assembler-times {fmov\ts} 1 {target lp64} } } */
+/* { dg-final { scan-assembler-times {fmov\td} 2 {target lp64} } } */
+/* { dg-final { scan-assembler-times {fmov\ts} 2 {target ilp32} } } */
+/* { dg-final { scan-assembler-times {fmov\td} 1 {target ilp32} } } */
+/* { dg-final { scan-assembler-times {ldr\ts} 1 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/vec_zeroextend.c
b/gcc/testsuite/gcc.target/aarch64/vec_zeroextend.c
new file mode 100644
index
0000000000000000000000000000000000000000..9c3971f036add8b1ee0204141d09db24720d05c6
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/vec_zeroextend.c
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+
+#define vector __attribute__((vector_size(16) ))
+
+vector unsigned long long
+f1(vector unsigned long long b, vector unsigned int a)
+{
+ b[0] = a[0];
+ return b;
+}
+
+unsigned long long
+f2(vector unsigned int a)
+{
+ return a[0];
+}
+
+/* { dg-final { scan-assembler-times {fmov} 2 } } */
+/* { dg-final { scan-assembler-not {umov} } } */
+/* { dg-final { scan-assembler-not {uxtw} } } */
