diff options
author | Andrew Stubbs <ams@codesourcery.com> | 2023-04-26 15:23:48 +0100 |
---|---|---|
committer | Andrew Stubbs <ams@codesourcery.com> | 2023-04-28 10:15:09 +0100 |
commit | ffc6b225c955a3d8478de1beba5ad08a7396648c (patch) | |
tree | 485b8076a50b823f73adaacfa2e4db86b2bdbff6 /gcc | |
parent | 45b94986b1640f914eeefd876f30987144b65f18 (diff) | |
download | gcc-ffc6b225c955a3d8478de1beba5ad08a7396648c.zip gcc-ffc6b225c955a3d8478de1beba5ad08a7396648c.tar.gz gcc-ffc6b225c955a3d8478de1beba5ad08a7396648c.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.
(cherry picked from commit b17c57b06d90f2ca12ea0395046c4ea7d439065f)
Diffstat (limited to 'gcc')
-rw-r--r-- | gcc/config/gcn/gcn-valu.md | 23 |
1 files changed, 15 insertions, 8 deletions
diff --git a/gcc/config/gcn/gcn-valu.md b/gcc/config/gcn/gcn-valu.md index 44c4846..7290cdc 100644 --- a/gcc/config/gcn/gcn-valu.md +++ b/gcc/config/gcn/gcn-valu.md @@ -2323,8 +2323,9 @@ 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 @@ -2368,7 +2369,8 @@ 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)); @@ -2392,7 +2394,8 @@ 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)); @@ -2419,7 +2422,9 @@ 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)); @@ -2439,7 +2444,8 @@ 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)); @@ -2459,10 +2465,11 @@ 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; }) |