Bootstrapped and regtested on x86-64-pc-linux-gnu{-m32,}.
Ready to push to trunk after first patch is approved.
gcc/ChangeLog:
PR target/101989
* config/i386/predicates.md (reg_or_notreg_operand): Rename to ..
(regmem_or_bitnot_regmem_operand): .. and extend to handle
memory_operand.
* config/i386/sse.md (*<avx512>_vpternlog<mode>_1): Force_reg
the operands which are required to be register_operand.
(*<avx512>_vpternlog<mode>_2): Ditto.
(*<avx512>_vpternlog<mode>_3): Ditto.
(*<avx512>_vternlog<mode>_all): Disallow embeded broadcast for
vector HFmodes since it's not a real AVX512FP16 instruction.
gcc/testsuite/ChangeLog:
* gcc.target/i386/pr101989-3.c: New test.
---
gcc/config/i386/predicates.md | 6 ++--
gcc/config/i386/sse.md | 41 +++++++++++++++-------
gcc/testsuite/gcc.target/i386/pr101989-3.c | 40 +++++++++++++++++++++
3 files changed, 72 insertions(+), 15 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr101989-3.c
diff --git a/gcc/config/i386/predicates.md b/gcc/config/i386/predicates.md
index df5acb425d4..114d8d448f1 100644
--- a/gcc/config/i386/predicates.md
+++ b/gcc/config/i386/predicates.md
@@ -1046,10 +1046,10 @@ (define_predicate "reg_or_pm1_operand"
;; True for registers, or (not: registers). Used to optimize 3-operand
;; bitwise operation.
-(define_predicate "reg_or_notreg_operand"
- (ior (match_operand 0 "register_operand")
+(define_predicate "regmem_or_bitnot_regmem_operand"
+ (ior (match_operand 0 "nonimmediate_operand")
(and (match_code "not")
- (match_test "register_operand (XEXP (op, 0), mode)"))))
+ (match_test "nonimmediate_operand (XEXP (op, 0), mode)"))))
;; True if OP is acceptable as operand of DImode shift expander.
(define_predicate "shiftdi_operand"
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 2764a250229..5aeb6065f13 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -11655,7 +11655,11 @@ (define_insn "*<avx512>_vternlog<mode>_all"
(match_operand:V 3 "bcst_vector_operand" "vmBr")
(match_operand:SI 4 "const_0_to_255_operand")]
UNSPEC_VTERNLOG))]
- "TARGET_AVX512F"
+ "TARGET_AVX512F
+/* Disallow embeded broadcast for vector HFmode since
+ it's not real AVX512FP16 instruction. */
+ && (GET_MODE_SIZE (GET_MODE_INNER (<MODE>mode)) >= 4
+ || GET_CODE (operands[3]) != VEC_DUPLICATE)"
"vpternlog<ternlogsuffix>\t{%4, %3, %2, %0|%0, %2, %3, %4}"
[(set_attr "type" "sselog")
(set_attr "prefix" "evex")
@@ -11683,11 +11687,11 @@ (define_insn_and_split "*<avx512>_vpternlog<mode>_1"
[(set (match_operand:V 0 "register_operand")
(any_logic:V
(any_logic1:V
- (match_operand:V 1 "reg_or_notreg_operand")
- (match_operand:V 2 "reg_or_notreg_operand"))
+ (match_operand:V 1 "regmem_or_bitnot_regmem_operand")
+ (match_operand:V 2 "regmem_or_bitnot_regmem_operand"))
(any_logic2:V
- (match_operand:V 3 "reg_or_notreg_operand")
- (match_operand:V 4 "reg_or_notreg_operand"))))]
+ (match_operand:V 3 "regmem_or_bitnot_regmem_operand")
+ (match_operand:V 4 "regmem_or_bitnot_regmem_operand"))))]
"(<MODE_SIZE> == 64 || TARGET_AVX512VL)
&& ix86_pre_reload_split ()
&& (rtx_equal_p (STRIP_UNARY (operands[1]),
@@ -11756,6 +11760,10 @@ (define_insn_and_split "*<avx512>_vpternlog<mode>_1"
operands[1] = STRIP_UNARY (operands[1]);
operands[2] = STRIP_UNARY (operands[2]);
operands[6] = STRIP_UNARY (operands[6]);
+ if (!register_operand (operands[2], <MODE>mode))
+ operands[2] = force_reg (<MODE>mode, operands[2]);
+ if (!register_operand (operands[6], <MODE>mode))
+ operands[6] = force_reg (<MODE>mode, operands[6]);
operands[5] = GEN_INT (reg_mask);
})
@@ -11764,10 +11772,10 @@ (define_insn_and_split "*<avx512>_vpternlog<mode>_2"
(any_logic:V
(any_logic1:V
(any_logic2:V
- (match_operand:V 1 "reg_or_notreg_operand")
- (match_operand:V 2 "reg_or_notreg_operand"))
- (match_operand:V 3 "reg_or_notreg_operand"))
- (match_operand:V 4 "reg_or_notreg_operand")))]
+ (match_operand:V 1 "regmem_or_bitnot_regmem_operand")
+ (match_operand:V 2 "regmem_or_bitnot_regmem_operand"))
+ (match_operand:V 3 "regmem_or_bitnot_regmem_operand"))
+ (match_operand:V 4 "regmem_or_bitnot_regmem_operand")))]
"(<MODE_SIZE> == 64 || TARGET_AVX512VL)
&& ix86_pre_reload_split ()
&& (rtx_equal_p (STRIP_UNARY (operands[1]),
@@ -11837,15 +11845,20 @@ (define_insn_and_split "*<avx512>_vpternlog<mode>_2"
operands[2] = STRIP_UNARY (operands[2]);
operands[6] = STRIP_UNARY (operands[6]);
operands[5] = GEN_INT (reg_mask);
+ if (!register_operand (operands[2], <MODE>mode))
+ operands[2] = force_reg (<MODE>mode, operands[2]);
+ if (!register_operand (operands[6], <MODE>mode))
+ operands[6] = force_reg (<MODE>mode, operands[6]);
+
})
(define_insn_and_split "*<avx512>_vpternlog<mode>_3"
[(set (match_operand:V 0 "register_operand")
(any_logic:V
(any_logic1:V
- (match_operand:V 1 "reg_or_notreg_operand")
- (match_operand:V 2 "reg_or_notreg_operand"))
- (match_operand:V 3 "reg_or_notreg_operand")))]
+ (match_operand:V 1 "regmem_or_bitnot_regmem_operand")
+ (match_operand:V 2 "regmem_or_bitnot_regmem_operand"))
+ (match_operand:V 3 "regmem_or_bitnot_regmem_operand")))]
"(<MODE_SIZE> == 64 || TARGET_AVX512VL)
&& ix86_pre_reload_split ()"
"#"
@@ -11876,6 +11889,10 @@ (define_insn_and_split "*<avx512>_vpternlog<mode>_3"
operands[2] = STRIP_UNARY (operands[2]);
operands[3] = STRIP_UNARY (operands[3]);
operands[4] = GEN_INT (reg_mask);
+ if (!register_operand (operands[2], <MODE>mode))
+ operands[2] = force_reg (<MODE>mode, operands[2]);
+ if (!register_operand (operands[3], <MODE>mode))
+ operands[3] = force_reg (<MODE>mode, operands[3]);
})
diff --git a/gcc/testsuite/gcc.target/i386/pr101989-3.c
b/gcc/testsuite/gcc.target/i386/pr101989-3.c
new file mode 100644
index 00000000000..dfd89918c17
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr101989-3.c
@@ -0,0 +1,40 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpternlog" 5 } } */
+/* { dg-final { scan-assembler-not "vpxor" } } */
+/* { dg-final { scan-assembler-not "vpor" } } */
+/* { dg-final { scan-assembler-not "vpand" } } */
+
+#include<immintrin.h>
+
+extern __m256i src1, src2, src3;
+
+__m256i
+foo (void)
+{
+ return (src2 & ~src1) | (src3 & src1);
+}
+
+__m256i
+foo1 (void)
+{
+ return (src2 & src1) | (src3 & ~src1);
+}
+
+__m256i
+foo2 (void)
+{
+ return (src2 & src1) | (~src3 & src1);
+}
+
+__m256i
+foo3 (void)
+{
+ return (~src2 & src1) | (src3 & src1);
+}
+
+__m256i
+foo4 (void)
+{
+ return src3 & src2 ^ src1;
+}
--
2.18.1
