This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
PATCH: Add TARGET_USE_VECTOR_FP_CONVERTS to x86 backend
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: gcc-patches at gcc dot gnu dot org, Uros Bizjak <ubizjak at gmail dot com>
- Date: Wed, 8 Oct 2008 03:04:34 -0700
- Subject: PATCH: Add TARGET_USE_VECTOR_FP_CONVERTS to x86 backend
Hi,
FP to FP splitters in i386.md check both feature and target with
(TARGET_USE_VECTOR_CONVERTS || TARGET_GENERIC)
instead of just feature. This patch adds TARGET_USE_VECTOR_FP_CONVERTS
and turns the splitter condition to just TARGET_USE_VECTOR_FP_CONVERTS.
OK for trunk?
Thanks.
H.J.
---
2008-10-08 H.J. Lu <hongjiu.lu@intel.com>
* config/i386/i386.c (initial_ix86_tune_features): Add
X86_TUNE_USE_VECTOR_FP_CONVERTS.
* config/i386/i386.h (ix86_tune_indices): Likewise.
(TARGET_USE_VECTOR_FP_CONVERTS): New.
* config/i386/i386.md: Check TARGET_USE_VECTOR_FP_CONVERTS
instead of (TARGET_USE_VECTOR_CONVERTS || TARGET_GENERIC)
for FP to FP splitters.
--- gcc/config/i386/i386.c.cvt 2008-10-08 02:46:34.000000000 -0700
+++ gcc/config/i386/i386.c 2008-10-08 02:54:11.000000000 -0700
@@ -1434,6 +1434,10 @@ static unsigned int initial_ix86_tune_fe
replacement is long decoded, so this split helps here as well. */
m_K6,
+ /* X86_TUNE_USE_VECTOR_FP_CONVERTS: Prefer vector packed SSE conversion
+ from FP to FP. */
+ m_AMDFAM10 | m_GENERIC,
+
/* X86_TUNE_USE_VECTOR_CONVERTS: Prefer vector packed SSE conversion
from integer to FP. */
m_AMDFAM10,
--- gcc/config/i386/i386.h.cvt 2008-10-08 02:46:34.000000000 -0700
+++ gcc/config/i386/i386.h 2008-10-08 02:53:17.000000000 -0700
@@ -292,6 +292,7 @@ enum ix86_tune_indices {
X86_TUNE_MOVE_M1_VIA_OR,
X86_TUNE_NOT_UNPAIRABLE,
X86_TUNE_NOT_VECTORMODE,
+ X86_TUNE_USE_VECTOR_FP_CONVERTS,
X86_TUNE_USE_VECTOR_CONVERTS,
X86_TUNE_FUSE_CMP_AND_BRANCH,
@@ -375,6 +376,8 @@ extern unsigned char ix86_tune_features[
#define TARGET_MOVE_M1_VIA_OR ix86_tune_features[X86_TUNE_MOVE_M1_VIA_OR]
#define TARGET_NOT_UNPAIRABLE ix86_tune_features[X86_TUNE_NOT_UNPAIRABLE]
#define TARGET_NOT_VECTORMODE ix86_tune_features[X86_TUNE_NOT_VECTORMODE]
+#define TARGET_USE_VECTOR_FP_CONVERTS \
+ ix86_tune_features[X86_TUNE_USE_VECTOR_FP_CONVERTS]
#define TARGET_USE_VECTOR_CONVERTS \
ix86_tune_features[X86_TUNE_USE_VECTOR_CONVERTS]
#define TARGET_FUSE_CMP_AND_BRANCH \
--- gcc/config/i386/i386.md.cvt 2008-10-08 02:46:34.000000000 -0700
+++ gcc/config/i386/i386.md 2008-10-08 02:52:53.000000000 -0700
@@ -4216,7 +4216,7 @@
[(set (match_operand:DF 0 "register_operand" "")
(float_extend:DF
(match_operand:SF 1 "nonimmediate_operand" "")))]
- "(TARGET_USE_VECTOR_CONVERTS || TARGET_GENERIC)
+ "TARGET_USE_VECTOR_FP_CONVERTS
&& optimize_insn_for_speed_p ()
&& reload_completed && SSE_REG_P (operands[0])"
[(set (match_dup 2)
@@ -4356,7 +4356,7 @@
[(set (match_operand:SF 0 "register_operand" "")
(float_truncate:SF
(match_operand:DF 1 "nonimmediate_operand" "")))]
- "(TARGET_USE_VECTOR_CONVERTS || TARGET_GENERIC)
+ "TARGET_USE_VECTOR_FP_CONVERTS
&& optimize_insn_for_speed_p ()
&& reload_completed && SSE_REG_P (operands[0])"
[(set (match_dup 2)