aboutsummaryrefslogtreecommitdiff
path: root/gcc
diff options
context:
space:
mode:
authorAndrew Stubbs <ams@codesourcery.com>2023-04-26 15:23:48 +0100
committerAndrew Stubbs <ams@codesourcery.com>2023-04-27 10:31:40 +0100
commita410f603fcaf2b963af5607eb359907fe273426e (patch)
treecc4bf9f06e2e99af95d53e4ca56c5937a0102e26 /gcc
parentfe884508f6b92ae84ba888b1fc431327ed0b2388 (diff)
downloadgcc-a410f603fcaf2b963af5607eb359907fe273426e.zip
gcc-a410f603fcaf2b963af5607eb359907fe273426e.tar.gz
gcc-a410f603fcaf2b963af5607eb359907fe273426e.tar.bz2
amdgcn: Fix addsub bug
The vec_fmsubadd instuction actually had add twice, by mistake. Also improve code-gen for all the complex patterns by using properly undefined values. Mostly this just prevents the compiler reserving space in the stack frame. gcc/ChangeLog: * config/gcn/gcn-valu.md (cmul<conj_op><mode>3): Use gcn_gen_undef. (cml<addsub_as><mode>4): Likewise. (vec_addsub<mode>3): Likewise. (cadd<rot><mode>3): Likewise. (vec_fmaddsub<mode>4): Likewise. (vec_fmsubadd<mode>4): Likewise, and use sub for the odd lanes.
Diffstat (limited to 'gcc')
-rw-r--r--gcc/ChangeLog.omp9
-rw-r--r--gcc/config/gcn/gcn-valu.md23
2 files changed, 24 insertions, 8 deletions
diff --git a/gcc/ChangeLog.omp b/gcc/ChangeLog.omp
index 72948e4..c8a12d1 100644
--- a/gcc/ChangeLog.omp
+++ b/gcc/ChangeLog.omp
@@ -1,3 +1,12 @@
+2023-04-27 Andrew Stubbs <ams@codesourcery.com>
+
+ * config/gcn/gcn-valu.md (cmul<conj_op><mode>3): Use gcn_gen_undef.
+ (cml<addsub_as><mode>4): Likewise.
+ (vec_addsub<mode>3): Likewise.
+ (cadd<rot><mode>3): Likewise.
+ (vec_fmaddsub<mode>4): Likewise.
+ (vec_fmsubadd<mode>4): Likewise, and use sub for the odd lanes.
+
2023-04-21 Andrew Stubbs <ams@codesourcery.com>
Backport from mainline:
diff --git a/gcc/config/gcn/gcn-valu.md b/gcc/config/gcn/gcn-valu.md
index ae532d9..a6380f9 100644
--- a/gcc/config/gcn/gcn-valu.md
+++ b/gcc/config/gcn/gcn-valu.md
@@ -2382,8 +2382,9 @@
rtx even = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (even, get_exec (0x5555555555555555UL));
rtx dest = operands[0];
- emit_insn (gen_<cmul_subadd><mode>3_exec (dest, t1, t1_perm, dest, even));
- // a*c-b*d 0
+ emit_insn (gen_<cmul_subadd><mode>3_exec (dest, t1, t1_perm,
+ gcn_gen_undef (<MODE>mode),
+ even)); // a*c-b*d 0
rtx t2_perm = gen_reg_rtx (<MODE>mode);
emit_insn (gen_dpp_swap_pairs<mode> (t2_perm, t2)); // b*c a*d
@@ -2427,7 +2428,8 @@
rtx even = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (even, get_exec (0x5555555555555555UL));
rtx dest = operands[0];
- emit_insn (gen_sub<mode>3_exec (dest, t1, t2_perm, dest, even));
+ emit_insn (gen_sub<mode>3_exec (dest, t1, t2_perm,
+ gcn_gen_undef (<MODE>mode), even));
rtx odd = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (odd, get_exec (0xaaaaaaaaaaaaaaaaUL));
@@ -2451,7 +2453,8 @@
rtx dest = operands[0];
rtx x = operands[1];
rtx y = operands[2];
- emit_insn (gen_sub<mode>3_exec (dest, x, y, dest, even));
+ emit_insn (gen_sub<mode>3_exec (dest, x, y, gcn_gen_undef (<MODE>mode),
+ even));
rtx odd = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (odd, get_exec (0xaaaaaaaaaaaaaaaaUL));
emit_insn (gen_add<mode>3_exec (dest, x, y, dest, odd));
@@ -2478,7 +2481,9 @@
rtx even = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (even, get_exec (0x5555555555555555UL));
- emit_insn (gen_<cadd_subadd><mode>3_exec (dest, x, y, dest, even));
+ emit_insn (gen_<cadd_subadd><mode>3_exec (dest, x, y,
+ gcn_gen_undef (<MODE>mode),
+ even));
rtx odd = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (odd, get_exec (0xaaaaaaaaaaaaaaaaUL));
emit_insn (gen_<cadd_addsub><mode>3_exec (dest, x, y, dest, odd));
@@ -2498,7 +2503,8 @@
rtx even = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (even, get_exec (0x5555555555555555UL));
rtx dest = operands[0];
- emit_insn (gen_sub<mode>3_exec (dest, t1, operands[3], dest, even));
+ emit_insn (gen_sub<mode>3_exec (dest, t1, operands[3],
+ gcn_gen_undef (<MODE>mode), even));
rtx odd = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (odd, get_exec (0xaaaaaaaaaaaaaaaaUL));
emit_insn (gen_add<mode>3_exec (dest, t1, operands[3], dest, odd));
@@ -2518,10 +2524,11 @@
rtx even = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (even, get_exec (0x5555555555555555UL));
rtx dest = operands[0];
- emit_insn (gen_add<mode>3_exec (dest, t1, operands[3], dest, even));
+ emit_insn (gen_add<mode>3_exec (dest, t1, operands[3],
+ gcn_gen_undef (<MODE>mode), even));
rtx odd = gen_rtx_REG (DImode, EXEC_REG);
emit_move_insn (odd, get_exec (0xaaaaaaaaaaaaaaaaUL));
- emit_insn (gen_add<mode>3_exec (dest, t1, operands[3], dest, odd));
+ emit_insn (gen_sub<mode>3_exec (dest, t1, operands[3], dest, odd));
DONE;
})