This patch adds support for "truncating" to a partial SVE vector from
either a full SVE vector or a wider partial vector. This truncation is
actually a no-op and so should have zero cost in the vector cost model.
Tested on aarch64-linux-gnu and applied as r278344.
Richard
2019-11-16 Richard Sandiford <richard.sandif...@arm.com>
gcc/
* config/aarch64/aarch64-sve.md
(trunc<SVE_HSDI:mode><SVE_PARTIAL_I:mode>2): New pattern.
* config/aarch64/aarch64.c (aarch64_integer_truncation_p): New
function.
(aarch64_sve_adjust_stmt_cost): Call it.
gcc/testsuite/
* gcc.target/aarch64/sve/mask_struct_load_1.c: Add
--param aarch64-sve-compare-costs=0.
* gcc.target/aarch64/sve/mask_struct_load_2.c: Likewise.
* gcc.target/aarch64/sve/mask_struct_load_3.c: Likewise.
* gcc.target/aarch64/sve/mask_struct_load_4.c: Likewise.
* gcc.target/aarch64/sve/mask_struct_load_5.c: Likewise.
* gcc.target/aarch64/sve/pack_1.c: Likewise.
* gcc.target/aarch64/sve/truncate_1.c: New test.
Index: gcc/config/aarch64/aarch64-sve.md
===================================================================
--- gcc/config/aarch64/aarch64-sve.md 2019-11-16 11:11:42.929267513 +0000
+++ gcc/config/aarch64/aarch64-sve.md 2019-11-16 11:13:24.236550470 +0000
@@ -72,6 +72,7 @@
;; ---- [INT] General unary arithmetic corresponding to rtx codes
;; ---- [INT] General unary arithmetic corresponding to unspecs
;; ---- [INT] Sign and zero extension
+;; ---- [INT] Truncation
;; ---- [INT] Logical inverse
;; ---- [FP<-INT] General unary arithmetic that maps to unspecs
;; ---- [FP] General unary arithmetic corresponding to unspecs
@@ -2889,6 +2890,29 @@ (define_insn "*cond_uxt<mode>_any"
)
;; -------------------------------------------------------------------------
+;; ---- [INT] Truncation
+;; -------------------------------------------------------------------------
+;; The patterns in this section are synthetic.
+;; -------------------------------------------------------------------------
+
+;; Truncate to a partial SVE vector from either a full vector or a
+;; wider partial vector. This is a no-op, because we can just ignore
+;; the unused upper bits of the source.
+(define_insn_and_split "trunc<SVE_HSDI:mode><SVE_PARTIAL_I:mode>2"
+ [(set (match_operand:SVE_PARTIAL_I 0 "register_operand" "=w")
+ (truncate:SVE_PARTIAL_I
+ (match_operand:SVE_HSDI 1 "register_operand" "w")))]
+ "TARGET_SVE && (~<SVE_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
+ "#"
+ "&& reload_completed"
+ [(set (match_dup 0) (match_dup 1))]
+ {
+ operands[1] = aarch64_replace_reg_mode (operands[1],
+ <SVE_PARTIAL_I:MODE>mode);
+ }
+)
+
+;; -------------------------------------------------------------------------
;; ---- [INT] Logical inverse
;; -------------------------------------------------------------------------
;; Includes:
Index: gcc/config/aarch64/aarch64.c
===================================================================
--- gcc/config/aarch64/aarch64.c 2019-11-16 11:11:42.933267485 +0000
+++ gcc/config/aarch64/aarch64.c 2019-11-16 11:13:24.236550470 +0000
@@ -12901,6 +12901,21 @@ aarch64_extending_load_p (stmt_vec_info
&& DR_IS_READ (STMT_VINFO_DATA_REF (def_stmt_info)));
}
+/* Return true if STMT_INFO is an integer truncation. */
+static bool
+aarch64_integer_truncation_p (stmt_vec_info stmt_info)
+{
+ gassign *assign = dyn_cast <gassign *> (stmt_info->stmt);
+ if (!assign || !CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (assign)))
+ return false;
+
+ tree lhs_type = TREE_TYPE (gimple_assign_lhs (assign));
+ tree rhs_type = TREE_TYPE (gimple_assign_rhs1 (assign));
+ return (INTEGRAL_TYPE_P (lhs_type)
+ && INTEGRAL_TYPE_P (rhs_type)
+ && TYPE_PRECISION (lhs_type) < TYPE_PRECISION (rhs_type));
+}
+
/* STMT_COST is the cost calculated by aarch64_builtin_vectorization_cost
for STMT_INFO, which has cost kind KIND. Adjust the cost as necessary
for SVE targets. */
@@ -12919,6 +12934,11 @@ aarch64_sve_adjust_stmt_cost (vect_cost_
if (kind == vector_stmt && aarch64_extending_load_p (stmt_info))
stmt_cost = 0;
+ /* For similar reasons, vector_stmt integer truncations are a no-op,
+ because we can just ignore the unused upper bits of the source. */
+ if (kind == vector_stmt && aarch64_integer_truncation_p (stmt_info))
+ stmt_cost = 0;
+
return stmt_cost;
}
Index: gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_1.c
===================================================================
--- gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_1.c 2019-03-08
18:14:29.768994780 +0000
+++ gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_1.c 2019-11-16
11:13:24.236550470 +0000
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
+/* { dg-options "-O2 -ftree-vectorize -ffast-math --param
aarch64-sve-compare-costs=0" } */
#include <stdint.h>
Index: gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_2.c
===================================================================
--- gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_2.c 2019-03-08
18:14:29.772994767 +0000
+++ gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_2.c 2019-11-16
11:13:24.236550470 +0000
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
+/* { dg-options "-O2 -ftree-vectorize -ffast-math --param
aarch64-sve-compare-costs=0" } */
#include <stdint.h>
Index: gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_3.c
===================================================================
--- gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_3.c 2019-03-08
18:14:29.772994767 +0000
+++ gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_3.c 2019-11-16
11:13:24.236550470 +0000
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
+/* { dg-options "-O2 -ftree-vectorize -ffast-math --param
aarch64-sve-compare-costs=0" } */
#include <stdint.h>
Index: gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_4.c
===================================================================
--- gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_4.c 2019-03-08
18:14:29.776994751 +0000
+++ gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_4.c 2019-11-16
11:13:24.240550442 +0000
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
+/* { dg-options "-O2 -ftree-vectorize -ffast-math --param
aarch64-sve-compare-costs=0" } */
#include <stdint.h>
Index: gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_5.c
===================================================================
--- gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_5.c 2019-03-08
18:14:29.784994721 +0000
+++ gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_5.c 2019-11-16
11:13:24.240550442 +0000
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
+/* { dg-options "-O2 -ftree-vectorize -ffast-math --param
aarch64-sve-compare-costs=0" } */
#include <stdint.h>
Index: gcc/testsuite/gcc.target/aarch64/sve/pack_1.c
===================================================================
--- gcc/testsuite/gcc.target/aarch64/sve/pack_1.c 2019-03-08
18:14:29.768994780 +0000
+++ gcc/testsuite/gcc.target/aarch64/sve/pack_1.c 2019-11-16
11:13:24.240550442 +0000
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O2 -ftree-vectorize" } */
+/* { dg-options "-O2 -ftree-vectorize --param aarch64-sve-compare-costs=0" } */
#include <stdint.h>
Index: gcc/testsuite/gcc.target/aarch64/sve/truncate_1.c
===================================================================
--- /dev/null 2019-09-17 11:41:18.176664108 +0100
+++ gcc/testsuite/gcc.target/aarch64/sve/truncate_1.c 2019-11-16
11:13:24.240550442 +0000
@@ -0,0 +1,44 @@
+/* { dg-options "-O2 -ftree-vectorize" } */
+
+#include <stdint.h>
+
+#define TEST_LOOP(TYPE1, TYPE2, SHIFT) \
+ void \
+ f_##TYPE1##_##TYPE2 (TYPE2 *restrict dst, TYPE1 *restrict src1, \
+ TYPE1 *restrict src2, int n) \
+ { \
+ for (int i = 0; i < n; ++i)
\
+ dst[i] = (TYPE1) (src1[i] + src2[i]) >> SHIFT; \
+ }
+
+#define TEST_ALL(T) \
+ T (uint16_t, uint8_t, 2) \
+ T (uint32_t, uint8_t, 18) \
+ T (uint64_t, uint8_t, 34) \
+ T (uint32_t, uint16_t, 3) \
+ T (uint64_t, uint16_t, 19) \
+ T (uint64_t, uint32_t, 4)
+
+TEST_ALL (TEST_LOOP)
+
+/* { dg-final { scan-assembler-times {\tld1h\tz[0-9]+\.h,} 2 } } */
+/* { dg-final { scan-assembler-times {\tld1w\tz[0-9]+\.s,} 4 } } */
+/* { dg-final { scan-assembler-times {\tld1d\tz[0-9]+\.d,} 6 } } */
+
+/* { dg-final { scan-assembler-times {\tadd\tz[0-9]+\.h, z[0-9]+\.h,
z[0-9]+\.h\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tadd\tz[0-9]+\.s, z[0-9]+\.s,
z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tadd\tz[0-9]+\.d, z[0-9]+\.d,
z[0-9]+\.d\n} 3 } } */
+
+/* { dg-final { scan-assembler-times {\tlsr\tz[0-9]+\.h, z[0-9]+\.h, #2\n} 1 }
} */
+/* { dg-final { scan-assembler-times {\tlsr\tz[0-9]+\.s, z[0-9]+\.s, #18\n} 1
} } */
+/* { dg-final { scan-assembler-times {\tlsr\tz[0-9]+\.d, z[0-9]+\.d, #34\n} 1
} } */
+/* { dg-final { scan-assembler-times {\tlsr\tz[0-9]+\.s, z[0-9]+\.s, #3\n} 1 }
} */
+/* { dg-final { scan-assembler-times {\tlsr\tz[0-9]+\.d, z[0-9]+\.d, #19\n} 1
} } */
+/* { dg-final { scan-assembler-times {\tlsr\tz[0-9]+\.d, z[0-9]+\.d, #4\n} 1 }
} */
+
+/* { dg-final { scan-assembler-times {\tst1b\tz[0-9]+\.h,} 1 } } */
+/* { dg-final { scan-assembler-times {\tst1b\tz[0-9]+\.s,} 1 } } */
+/* { dg-final { scan-assembler-times {\tst1b\tz[0-9]+\.d,} 1 } } */
+/* { dg-final { scan-assembler-times {\tst1h\tz[0-9]+\.s,} 1 } } */
+/* { dg-final { scan-assembler-times {\tst1h\tz[0-9]+\.d,} 1 } } */
+/* { dg-final { scan-assembler-times {\tst1w\tz[0-9]+\.d,} 1 } } */
