summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c
diff options
context:
space:
mode:
authorChristoph Bumiller <e0425955@student.tuwien.ac.at>2010-08-22 22:59:01 +0200
committerChristoph Bumiller <e0425955@student.tuwien.ac.at>2010-08-23 14:25:51 +0200
commit33f45c5a8afd353ad9bbd8647fa5c6dfc59cdfd7 (patch)
tree5b0bfdbe6cbec1a14ed6fbac74ed5d1769421af8 /src/gallium/drivers/nv50/nv50_tgsi_to_nc.c
parent3e54d63429fe7ca5db3c75c181abbaf7a7f55724 (diff)
nv50: DP2, fix ARL
Diffstat (limited to 'src/gallium/drivers/nv50/nv50_tgsi_to_nc.c')
-rw-r--r--src/gallium/drivers/nv50/nv50_tgsi_to_nc.c44
1 files changed, 27 insertions, 17 deletions
diff --git a/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c b/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c
index dafff725b8..7b2ccef704 100644
--- a/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c
+++ b/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c
@@ -1199,6 +1199,25 @@ bld_tex(struct bld_context *bld, struct nv_value *dst0[4],
nvi->tex_argc = arg;
}
+static INLINE struct nv_value *
+bld_dot(struct bld_context *bld, const struct tgsi_full_instruction *insn,
+ int n)
+{
+ struct nv_value *dotp, *src0, *src1;
+ int c;
+
+ src0 = emit_fetch(bld, insn, 0, 0);
+ src1 = emit_fetch(bld, insn, 1, 0);
+ dotp = bld_insn_2(bld, NV_OP_MUL, src0, src1);
+
+ for (c = 1; c < n; ++c) {
+ src0 = emit_fetch(bld, insn, 0, c);
+ src1 = emit_fetch(bld, insn, 1, c);
+ dotp = bld_insn_3(bld, NV_OP_MAD, src0, src1, dotp);
+ }
+ return dotp;
+}
+
#define FOR_EACH_DST0_ENABLED_CHANNEL(chan, inst) \
for (chan = 0; chan < 4; ++chan) \
if ((inst)->Dst[0].Register.WriteMask & (1 << chan))
@@ -1232,7 +1251,7 @@ bld_instruction(struct bld_context *bld,
src1 = bld_imm_u32(bld, 4);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) {
src0 = emit_fetch(bld, insn, 0, c);
- (temp = bld_insn_1(bld, NV_OP_FLOOR, temp))->reg.type = NV_TYPE_S32;
+ (temp = bld_insn_1(bld, NV_OP_FLOOR, src0))->reg.type = NV_TYPE_S32;
dst0[c] = bld_insn_2(bld, NV_OP_SHL, temp, src1);
}
break;
@@ -1269,27 +1288,18 @@ bld_instruction(struct bld_context *bld,
temp = bld_insn_1(bld, NV_OP_PRESIN, src0);
dst0[3] = bld_insn_1(bld, opcode, temp);
break;
+ case TGSI_OPCODE_DP2:
+ temp = bld_dot(bld, insn, 2);
+ FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
+ dst0[c] = temp;
+ break;
case TGSI_OPCODE_DP3:
- src0 = emit_fetch(bld, insn, 0, 0);
- src1 = emit_fetch(bld, insn, 1, 0);
- temp = bld_insn_2(bld, NV_OP_MUL, src0, src1);
- for (c = 1; c < 3; ++c) {
- src0 = emit_fetch(bld, insn, 0, c);
- src1 = emit_fetch(bld, insn, 1, c);
- temp = bld_insn_3(bld, NV_OP_MAD, src0, src1, temp);
- }
+ temp = bld_dot(bld, insn, 3);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;
case TGSI_OPCODE_DP4:
- src0 = emit_fetch(bld, insn, 0, 0);
- src1 = emit_fetch(bld, insn, 1, 0);
- temp = bld_insn_2(bld, NV_OP_MUL, src0, src1);
- for (c = 1; c < 4; ++c) {
- src0 = emit_fetch(bld, insn, 0, c);
- src1 = emit_fetch(bld, insn, 1, c);
- temp = bld_insn_3(bld, NV_OP_MAD, src0, src1, temp);
- }
+ temp = bld_dot(bld, insn, 4);
FOR_EACH_DST0_ENABLED_CHANNEL(c, insn)
dst0[c] = temp;
break;