public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-21 21:05 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-21 21:05 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:63e61c4651301839e07e8e2818a174774476d8af
commit 63e61c4651301839e07e8e2818a174774476d8af
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Fri Apr 21 17:04:38 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
In addition, I changed the vec_extract of V4SFmode where the element number is
constant without conversion to do the split before register allocation. I also
simplified the alternatives.
2023-04-21 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_load): Allow split before
register allocation.
(vsx_extract_v4sf_to_df_load): New insn.
gcc/testsuite/
* gcc.target/powerpc/vec-extract-mem-float-1.c: New test.
Diff:
---
gcc/config/rs6000/vsx.md | 38 +++++++++++++++++-----
| 31 ++++++++++++++++++
2 files changed, 61 insertions(+), 8 deletions(-)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 417aff5e24b..17e56ab1ce4 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -3549,23 +3549,45 @@
[(set_attr "length" "8")
(set_attr "type" "fp")])
+;; V4SF extract from memory convert to DFmode with constant element number. If
+;; the element number is 0, we don't need a temporary register.
(define_insn_and_split "*vsx_extract_v4sf_load"
- [(set (match_operand:SF 0 "register_operand" "=f,v,v,?r")
+ [(set (match_operand:SF 0 "register_operand" "=wa,wa,?r,?r")
(vec_select:SF
- (match_operand:V4SF 1 "memory_operand" "m,Z,m,m")
- (parallel [(match_operand:QI 2 "const_0_to_3_operand" "n,n,n,n")])))
- (clobber (match_scratch:P 3 "=&b,&b,&b,&b"))]
+ (match_operand:V4SF 1 "memory_operand" "m,Q,m,Q")
+ (parallel
+ [(match_operand:QI 2 "const_0_to_3_operand" "O,n,O,n")])))
+ (clobber (match_scratch:P 3 "=X,&b,X,&b"))]
"VECTOR_MEM_VSX_P (V4SFmode)"
"#"
- "&& reload_completed"
+ "&& 1"
[(set (match_dup 0) (match_dup 4))]
{
operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
operands[3], SFmode);
}
- [(set_attr "type" "fpload,fpload,fpload,load")
- (set_attr "length" "8")
- (set_attr "isa" "*,p7v,p9v,*")])
+ [(set_attr "type" "fpload,fpload,load,load")
+ (set_attr "length" "4,8,4,8")])
+
+;; V4SF extract from memory and convert to DFmode with constant element number.
+(define_insn_and_split "*vsx_extract_v4sf_load_to_df"
+ [(set (match_operand:DF 0 "register_operand" "=wa,wa")
+ (float_extend:DF
+ (vec_select:SF
+ (match_operand:V4SF 1 "memory_operand" "m,Q")
+ (parallel [(match_operand:QI 2 "const_0_to_3_operand" "O,n")]))))
+ (clobber (match_scratch:P 3 "=X,&b"))]
+ "VECTOR_MEM_VSX_P (V4SFmode)"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (float_extend:DF (match_dup 4)))]
+{
+ operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
+ operands[3], SFmode);
+}
+ [(set_attr "type" "fpload")
+ (set_attr "length" "4,8")])
;; Variable V4SF extract from a register
(define_insn_and_split "vsx_extract_v4sf_var"
--git a/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
new file mode 100644
index 00000000000..34ebc574339
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
@@ -0,0 +1,31 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target powerpc_vsx_ok } */
+/* { dg-options "-O2 -mvsx" } */
+
+/* Test to verify that the vec_extract with constant element numbers can load
+ float (SF) variables into a GPR without doing a LFS or STFS. */
+
+#include <altivec.h>
+
+void
+extract_float_0_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 0);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+void
+extract_float_1_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 1);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+/* { dg-final { scan-assembler-times {\mlwz\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mstwz\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mlfs\M|\mlxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mstfs\M|\mstxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mm[tf]vsd} } } */
+/* { dg-final { scan-assembler-not {\mxscvdpspn?\M} } } */
^ permalink raw reply [flat|nested] 8+ messages in thread
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-21 19:55 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-21 19:55 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:bb3324da2b643fa18dacd9b6e7554e991dca0d35
commit bb3324da2b643fa18dacd9b6e7554e991dca0d35
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Fri Apr 21 15:55:18 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
In addition, I changed the vec_extract of V4SFmode where the element number is
constant without conversion to do the split before register allocation.
2023-04-21 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_load): Allow split before
register allocation.
(vsx_extract_v4sf_to_df_load): New insn.
gcc/testsuite/
* gcc.target/powerpc/vec-extract-mem-float-1.c: New test.
Diff:
---
gcc/config/rs6000/vsx.md | 24 ++++++++++++++-
| 36 ++++++++++++++++++++++
2 files changed, 59 insertions(+), 1 deletion(-)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 417aff5e24b..c3848de5d4f 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -3549,6 +3549,7 @@
[(set_attr "length" "8")
(set_attr "type" "fp")])
+;; V4SF extract from memory convert to DFmode with constant element number.
(define_insn_and_split "*vsx_extract_v4sf_load"
[(set (match_operand:SF 0 "register_operand" "=f,v,v,?r")
(vec_select:SF
@@ -3557,7 +3558,7 @@
(clobber (match_scratch:P 3 "=&b,&b,&b,&b"))]
"VECTOR_MEM_VSX_P (V4SFmode)"
"#"
- "&& reload_completed"
+ "&& 1"
[(set (match_dup 0) (match_dup 4))]
{
operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
@@ -3567,6 +3568,27 @@
(set_attr "length" "8")
(set_attr "isa" "*,p7v,p9v,*")])
+;; V4SF extract from memory and convert to DFmode with constant element number.
+(define_insn_and_split "*vsx_extract_v4sf_load_to_df"
+ [(set (match_operand:DF 0 "register_operand" "=f,v,v")
+ (float_extend:DF
+ (vec_select:SF
+ (match_operand:V4SF 1 "memory_operand" "m,Z,m")
+ (parallel [(match_operand:QI 2 "const_0_to_3_operand" "n,n,n")]))))
+ (clobber (match_scratch:P 3 "=&b,&b,&b"))]
+ "VECTOR_MEM_VSX_P (V4SFmode)"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (float_extend:DF (match_dup 4)))]
+{
+ operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
+ operands[3], SFmode);
+}
+ [(set_attr "type" "fpload")
+ (set_attr "length" "8")
+ (set_attr "isa" "*,p7v,p9v")])
+
;; Variable V4SF extract from a register
(define_insn_and_split "vsx_extract_v4sf_var"
[(set (match_operand:SF 0 "gpc_reg_operand" "=wa")
--git a/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
new file mode 100644
index 00000000000..e110d308307
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
@@ -0,0 +1,36 @@
+/* { dg-do compile { target lp64 } } */
+/* { dg-require-effective-target powerpc_p8vector_ok } */
+/* { dg-options "-O2 -mdejagnu-cpu=power8" } */
+
+/* Test to verify that the vec_extract with constant element numbers can load
+ float (SF) variables into a GPR without doing a LFS or STFS.
+
+ Target LP64 and power8 vector are needed because the compiler only does the
+ vec_extract optimizations on 64-bit machines that have direct move support.
+ On earlier machines, vec_extract is done by storing the V4SF value into
+ memory, and just doing the load from memory. */
+
+#include <altivec.h>
+
+float
+extract_float_0_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 0);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+float
+extract_float_3_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 3);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+/* { dg-final { scan-assembler-times {\mlwz\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mstwz\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mlfs\M|\mlxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mstfs\M|\mstxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mm[tf]vsd} } } */
+/* { dg-final { scan-assembler-not {\mxscvdpspn?\M} } } */
^ permalink raw reply [flat|nested] 8+ messages in thread
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-21 19:44 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-21 19:44 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:87ef9d86fc6fd04eb87eebc4a74b22349b7235b3
commit 87ef9d86fc6fd04eb87eebc4a74b22349b7235b3
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Fri Apr 21 15:44:03 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
In addition, I changed the vec_extract of V4SFmode where the element number is
constant without conversion to do the split before register allocation.
2023-04-21 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_load): Allow split before
register allocation.
(vsx_extract_v4sf_to_df_load): New insn.
gcc/testsuite/
* gcc.target/powerpc/vec-extract-mem-float-1.c: New test.
Diff:
---
gcc/config/rs6000/vsx.md | 24 ++++++++++++++++-
| 31 ++++++++++++++++++++++
2 files changed, 54 insertions(+), 1 deletion(-)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 417aff5e24b..c3848de5d4f 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -3549,6 +3549,7 @@
[(set_attr "length" "8")
(set_attr "type" "fp")])
+;; V4SF extract from memory convert to DFmode with constant element number.
(define_insn_and_split "*vsx_extract_v4sf_load"
[(set (match_operand:SF 0 "register_operand" "=f,v,v,?r")
(vec_select:SF
@@ -3557,7 +3558,7 @@
(clobber (match_scratch:P 3 "=&b,&b,&b,&b"))]
"VECTOR_MEM_VSX_P (V4SFmode)"
"#"
- "&& reload_completed"
+ "&& 1"
[(set (match_dup 0) (match_dup 4))]
{
operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
@@ -3567,6 +3568,27 @@
(set_attr "length" "8")
(set_attr "isa" "*,p7v,p9v,*")])
+;; V4SF extract from memory and convert to DFmode with constant element number.
+(define_insn_and_split "*vsx_extract_v4sf_load_to_df"
+ [(set (match_operand:DF 0 "register_operand" "=f,v,v")
+ (float_extend:DF
+ (vec_select:SF
+ (match_operand:V4SF 1 "memory_operand" "m,Z,m")
+ (parallel [(match_operand:QI 2 "const_0_to_3_operand" "n,n,n")]))))
+ (clobber (match_scratch:P 3 "=&b,&b,&b"))]
+ "VECTOR_MEM_VSX_P (V4SFmode)"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (float_extend:DF (match_dup 4)))]
+{
+ operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
+ operands[3], SFmode);
+}
+ [(set_attr "type" "fpload")
+ (set_attr "length" "8")
+ (set_attr "isa" "*,p7v,p9v")])
+
;; Variable V4SF extract from a register
(define_insn_and_split "vsx_extract_v4sf_var"
[(set (match_operand:SF 0 "gpc_reg_operand" "=wa")
--git a/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
new file mode 100644
index 00000000000..cc10fb41894
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
@@ -0,0 +1,31 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target powerpc_p8vector_ok } */
+/* { dg-options "-O2 -mdejagnu-cpu=power8" } */
+
+/* Test to verify that the vec_extract with constant element numbers can load
+ float (SF) variables into a GPR without doing a LFS or STFS. */
+
+#include <altivec.h>
+
+float
+extract_float_0_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 0);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+float
+extract_float_3_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 3);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+/* { dg-final { scan-assembler-times {\mlwz\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mstwz\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mlfs\M|\mlxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mstfs\M|\mstxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mm[tf]vsd} } } */
+/* { dg-final { scan-assembler-not {\mxscvdpspn?\M} } } */
^ permalink raw reply [flat|nested] 8+ messages in thread
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-21 19:04 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-21 19:04 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:5a5c396c1052046c96e5e823eee80c05ef53b0a5
commit 5a5c396c1052046c96e5e823eee80c05ef53b0a5
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Fri Apr 21 15:04:07 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
In addition, I changed the vec_extract of V4SFmode where the element number is
constant without conversion to do the split before register allocation.
2023-04-21 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_load): Allow split before
register allocation.
(vsx_extract_v4sf_to_df_load): New insn.
gcc/testsuite/
* gcc.target/powerpc/vec-extract-mem-float-1.c: New test.
Diff:
---
gcc/config/rs6000/vsx.md | 24 +++++++++++++++++-
| 29 ++++++++++++++++++++++
2 files changed, 52 insertions(+), 1 deletion(-)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 417aff5e24b..c3848de5d4f 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -3549,6 +3549,7 @@
[(set_attr "length" "8")
(set_attr "type" "fp")])
+;; V4SF extract from memory convert to DFmode with constant element number.
(define_insn_and_split "*vsx_extract_v4sf_load"
[(set (match_operand:SF 0 "register_operand" "=f,v,v,?r")
(vec_select:SF
@@ -3557,7 +3558,7 @@
(clobber (match_scratch:P 3 "=&b,&b,&b,&b"))]
"VECTOR_MEM_VSX_P (V4SFmode)"
"#"
- "&& reload_completed"
+ "&& 1"
[(set (match_dup 0) (match_dup 4))]
{
operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
@@ -3567,6 +3568,27 @@
(set_attr "length" "8")
(set_attr "isa" "*,p7v,p9v,*")])
+;; V4SF extract from memory and convert to DFmode with constant element number.
+(define_insn_and_split "*vsx_extract_v4sf_load_to_df"
+ [(set (match_operand:DF 0 "register_operand" "=f,v,v")
+ (float_extend:DF
+ (vec_select:SF
+ (match_operand:V4SF 1 "memory_operand" "m,Z,m")
+ (parallel [(match_operand:QI 2 "const_0_to_3_operand" "n,n,n")]))))
+ (clobber (match_scratch:P 3 "=&b,&b,&b"))]
+ "VECTOR_MEM_VSX_P (V4SFmode)"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (float_extend:DF (match_dup 4)))]
+{
+ operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
+ operands[3], SFmode);
+}
+ [(set_attr "type" "fpload")
+ (set_attr "length" "8")
+ (set_attr "isa" "*,p7v,p9v")])
+
;; Variable V4SF extract from a register
(define_insn_and_split "vsx_extract_v4sf_var"
[(set (match_operand:SF 0 "gpc_reg_operand" "=wa")
--git a/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
new file mode 100644
index 00000000000..eab7892ed80
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
@@ -0,0 +1,29 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target powerpc_vsx_ok } */
+/* { dg-options "-O2 -mvsx" } */
+
+/* Test to verify that the vec_extract with constant element numbers can load
+ float (SF) variables into a GPR without doing a LFS or STFS. */
+
+#include <altivec.h>
+
+float
+extract_float_0_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 0);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+float
+extract_float_3_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 3);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+/* { dg-final { scan-assembler-times {\mlwz\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mstwz\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mlfs\M|\mlxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mstfs\M|\mstxsspx?\M} } } */
^ permalink raw reply [flat|nested] 8+ messages in thread
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-21 18:57 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-21 18:57 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:db1da1abcc562f3f2cf109f37f4d932bff0571b9
commit db1da1abcc562f3f2cf109f37f4d932bff0571b9
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Fri Apr 21 14:57:30 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
In addition, I changed the vec_extract of V4SFmode where the element number is
constant without conversion to do the split before register allocation.
2023-04-21 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_load): Allow split before
register allocation.
(vsx_extract_v4sf_to_df_load): New insn.
gcc/testsuite/
* gcc.target/powerpc/vec-extract-mem-float-1.c: New test.
Diff:
---
gcc/config/rs6000/vsx.md | 24 +++++++++++++++++-
| 29 ++++++++++++++++++++++
2 files changed, 52 insertions(+), 1 deletion(-)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 417aff5e24b..84a41158d6b 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -3549,6 +3549,7 @@
[(set_attr "length" "8")
(set_attr "type" "fp")])
+;; V4SF extract from memory and convert to DFmode with constant element number
(define_insn_and_split "*vsx_extract_v4sf_load"
[(set (match_operand:SF 0 "register_operand" "=f,v,v,?r")
(vec_select:SF
@@ -3557,7 +3558,7 @@
(clobber (match_scratch:P 3 "=&b,&b,&b,&b"))]
"VECTOR_MEM_VSX_P (V4SFmode)"
"#"
- "&& reload_completed"
+ "&& 1"
[(set (match_dup 0) (match_dup 4))]
{
operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
@@ -3567,6 +3568,27 @@
(set_attr "length" "8")
(set_attr "isa" "*,p7v,p9v,*")])
+;; V4SF extract from memory and convert to DFmode with constant element number
+(define_insn_and_split "*vsx_extract_v4sf_load_to_df"
+ [(set (match_operand:DF 0 "register_operand" "=f,v,v")
+ (float_extend:DF
+ (vec_select:SF
+ (match_operand:V4SF 1 "memory_operand" "m,Z,m")
+ (parallel [(match_operand:QI 2 "const_0_to_3_operand" "n,n,n")]))))
+ (clobber (match_scratch:P 3 "=&b,&b,&b"))]
+ "VECTOR_MEM_VSX_P (V4SFmode)"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (float_extend:DF (match_dup 4)))]
+{
+ operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
+ operands[3], SFmode);
+}
+ [(set_attr "type" "fpload")
+ (set_attr "length" "8")
+ (set_attr "isa" "*,p7v,p9v")])
+
;; Variable V4SF extract from a register
(define_insn_and_split "vsx_extract_v4sf_var"
[(set (match_operand:SF 0 "gpc_reg_operand" "=wa")
--git a/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
new file mode 100644
index 00000000000..eab7892ed80
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/vec-extract-mem-float-1.c
@@ -0,0 +1,29 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target powerpc_vsx_ok } */
+/* { dg-options "-O2 -mvsx" } */
+
+/* Test to verify that the vec_extract with constant element numbers can load
+ float (SF) variables into a GPR without doing a LFS or STFS. */
+
+#include <altivec.h>
+
+float
+extract_float_0_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 0);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+float
+extract_float_3_gpr (vector float *p, float *q)
+{
+ float x = vec_extract (*p, 3);
+ __asm__ ("# %0" : "+r" (x)); /* lwz. */
+ *q = x;
+}
+
+/* { dg-final { scan-assembler-times {\mlwz\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mstwz\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mlfs\M|\mlxsspx?\M} } } */
+/* { dg-final { scan-assembler-not {\mstfs\M|\mstxsspx?\M} } } */
^ permalink raw reply [flat|nested] 8+ messages in thread
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-19 19:32 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-19 19:32 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:189afb0d76cb331f3754c586133a2ea3982e1342
commit 189afb0d76cb331f3754c586133a2ea3982e1342
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Wed Apr 19 15:32:35 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
2023-04-18 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_to_df_load): New insn.
Diff:
---
gcc/config/rs6000/vsx.md | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 417aff5e24b..ebc986fc6ac 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -3567,6 +3567,27 @@
(set_attr "length" "8")
(set_attr "isa" "*,p7v,p9v,*")])
+;; V4SF extract from memory and convert to DFmode with constant element number
+(define_insn_and_split "*vsx_extract_v4sf_to_df_load"
+ [(set (match_operand:DF 0 "register_operand" "=f,v")
+ (float_extend:DF
+ (vec_select:SF
+ (match_operand:V4SF 1 "memory_operand" "m,m")
+ (parallel [(match_operand:QI 2 "const_0_to_3_operand" "n,n")]))))
+ (clobber (match_scratch:P 3 "=&b,&b"))]
+ "VECTOR_MEM_VSX_P (V4SFmode)"
+ "#"
+ "&& reload_completed"
+ [(set (match_dup 0)
+ (float_extend:DF (match_dup 4)))]
+{
+ operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
+ operands[3], SFmode);
+}
+ [(set_attr "type" "fpload")
+ (set_attr "length" "8")
+ (set_attr "isa" "*,p8v")])
+
;; Variable V4SF extract from a register
(define_insn_and_split "vsx_extract_v4sf_var"
[(set (match_operand:SF 0 "gpc_reg_operand" "=wa")
^ permalink raw reply [flat|nested] 8+ messages in thread
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-19 19:25 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-19 19:25 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:354e5bb1234064a4a5a0cb88061470738107668f
commit 354e5bb1234064a4a5a0cb88061470738107668f
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Wed Apr 19 15:24:36 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
2023-04-18 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_to_df_load): New insn.
Diff:
---
gcc/config/rs6000/vsx.md | 29 ++++++++++++++++++++---------
1 file changed, 20 insertions(+), 9 deletions(-)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 0e681844243..c3b870640ed 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -223,6 +223,12 @@
(V8HI "v")
(V4SI "wa")])
+;; Mode attribute to give the isa constraint for accessing Altivec registers
+;; with vector extract and insert operations.
+(define_mode_attr VSX_EX_ISA [(V16QI "p9v")
+ (V8HI "p9v")
+ (V4SI "p8v")])
+
;; Mode iterator for binary floating types other than double to
;; optimize convert to that floating point type from an extract
;; of an integer type
@@ -3951,23 +3957,28 @@
}
[(set_attr "type" "mfvsr")])
-;; Optimize extracting a single scalar element from memory.
+;; Extract a V16QI/V8HI/V4SI element from memory with a constant element number.
+;; If the element number is 0, we don't need to do a load immediate operation.
+;; Likewise for GPRs with offsettable loads, we can fold the offset into the
+;; address. For vector registers, we are limited to X-FORM memory addresses.
+;; PowerPC64 is needed because we need a DI temporary base register.
(define_insn_and_split "*vsx_extract_<mode>_load"
- [(set (match_operand:<VEC_base> 0 "register_operand" "=r")
+ [(set (match_operand:<VEC_base> 0 "register_operand" "=r,r,r,<VSX_EX>,<VSX_EX>")
(vec_select:<VEC_base>
- (match_operand:VSX_EXTRACT_I 1 "memory_operand" "m")
- (parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "n")])))
- (clobber (match_scratch:DI 3 "=&b"))]
- "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_DIRECT_MOVE_64BIT"
+ (match_operand:VSX_EXTRACT_I 1 "memory_operand" "m,o,m,Z,Q")
+ (parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "0,n,n,0,n")])))
+ (clobber (match_scratch:DI 3 "=X,X,&b,X,&b"))]
+ "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_POWERPC64"
"#"
- "&& reload_completed"
+ "&& 1"
[(set (match_dup 0) (match_dup 4))]
{
operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
operands[3], <VEC_base>mode);
}
- [(set_attr "type" "load")
- (set_attr "length" "8")])
+ [(set_attr "type" "load,load,load,fpload,fpload")
+ (set_attr "length" "4,4,8,4,8")
+ (set_attr "isa" "*,*,*,<VSX_EX_ISA>,<VSX_EX_ISA>")])
;; Variable V16QI/V8HI/V4SI extract from a register
(define_insn_and_split "vsx_extract_<mode>_var"
^ permalink raw reply [flat|nested] 8+ messages in thread
* [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert.
@ 2023-04-19 19:18 Michael Meissner
0 siblings, 0 replies; 8+ messages in thread
From: Michael Meissner @ 2023-04-19 19:18 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:2ecd33ae0e474e043aa8b509ba2bc0af1cbc8597
commit 2ecd33ae0e474e043aa8b509ba2bc0af1cbc8597
Author: Michael Meissner <meissner@linux.ibm.com>
Date: Wed Apr 19 15:16:48 2023 -0400
Combine vec_extract of V4SF with DF convert.
This patch adds a combine insn that merges loading up a vec_extract of V4SFmode
where the element number is constant combined with a conversion to DFmode.
2023-04-18 Michael Meissner <meissner@linux.ibm.com>
gcc/
* config/rs6000/vsx.md (vsx_extract_v4sf_to_df_load): New insn.
Diff:
---
gcc/config/rs6000/vsx.md | 29 ++++++++++++++++++++---------
1 file changed, 20 insertions(+), 9 deletions(-)
diff --git a/gcc/config/rs6000/vsx.md b/gcc/config/rs6000/vsx.md
index 0e681844243..c3b870640ed 100644
--- a/gcc/config/rs6000/vsx.md
+++ b/gcc/config/rs6000/vsx.md
@@ -223,6 +223,12 @@
(V8HI "v")
(V4SI "wa")])
+;; Mode attribute to give the isa constraint for accessing Altivec registers
+;; with vector extract and insert operations.
+(define_mode_attr VSX_EX_ISA [(V16QI "p9v")
+ (V8HI "p9v")
+ (V4SI "p8v")])
+
;; Mode iterator for binary floating types other than double to
;; optimize convert to that floating point type from an extract
;; of an integer type
@@ -3951,23 +3957,28 @@
}
[(set_attr "type" "mfvsr")])
-;; Optimize extracting a single scalar element from memory.
+;; Extract a V16QI/V8HI/V4SI element from memory with a constant element number.
+;; If the element number is 0, we don't need to do a load immediate operation.
+;; Likewise for GPRs with offsettable loads, we can fold the offset into the
+;; address. For vector registers, we are limited to X-FORM memory addresses.
+;; PowerPC64 is needed because we need a DI temporary base register.
(define_insn_and_split "*vsx_extract_<mode>_load"
- [(set (match_operand:<VEC_base> 0 "register_operand" "=r")
+ [(set (match_operand:<VEC_base> 0 "register_operand" "=r,r,r,<VSX_EX>,<VSX_EX>")
(vec_select:<VEC_base>
- (match_operand:VSX_EXTRACT_I 1 "memory_operand" "m")
- (parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "n")])))
- (clobber (match_scratch:DI 3 "=&b"))]
- "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_DIRECT_MOVE_64BIT"
+ (match_operand:VSX_EXTRACT_I 1 "memory_operand" "m,o,m,Z,Q")
+ (parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "0,n,n,0,n")])))
+ (clobber (match_scratch:DI 3 "=X,X,&b,X,&b"))]
+ "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_POWERPC64"
"#"
- "&& reload_completed"
+ "&& 1"
[(set (match_dup 0) (match_dup 4))]
{
operands[4] = rs6000_adjust_vec_address (operands[0], operands[1], operands[2],
operands[3], <VEC_base>mode);
}
- [(set_attr "type" "load")
- (set_attr "length" "8")])
+ [(set_attr "type" "load,load,load,fpload,fpload")
+ (set_attr "length" "4,4,8,4,8")
+ (set_attr "isa" "*,*,*,<VSX_EX_ISA>,<VSX_EX_ISA>")])
;; Variable V16QI/V8HI/V4SI extract from a register
(define_insn_and_split "vsx_extract_<mode>_var"
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2023-04-21 21:05 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-04-21 21:05 [gcc(refs/users/meissner/heads/work119)] Combine vec_extract of V4SF with DF convert Michael Meissner
-- strict thread matches above, loose matches on Subject: below --
2023-04-21 19:55 Michael Meissner
2023-04-21 19:44 Michael Meissner
2023-04-21 19:04 Michael Meissner
2023-04-21 18:57 Michael Meissner
2023-04-19 19:32 Michael Meissner
2023-04-19 19:25 Michael Meissner
2023-04-19 19:18 Michael Meissner
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).