Mesa (master): freedreno/a3xx/compiler: better const handling
Rob Clark
robclark at kemper.freedesktop.org
Thu Aug 29 22:01:00 UTC 2013
Module: Mesa
Branch: master
Commit: 4a2b5b23840237e54bc810aaaa23b3a81832e1e3
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=4a2b5b23840237e54bc810aaaa23b3a81832e1e3
Author: Rob Clark <robclark at freedesktop.org>
Date: Sat Aug 24 17:30:50 2013 -0400
freedreno/a3xx/compiler: better const handling
Seems like most/all instructions have some restrictions about const src
registers. In seems like the 2 src (cat2) instructions can take at most
one const, and the 3 src (cat3) instructions can take at most one const
in the first 2 arguments. And so on. Handle this properly now.
Signed-off-by: Rob Clark <robclark at freedesktop.org>
---
src/gallium/drivers/freedreno/a3xx/fd3_compiler.c | 211 ++++++++++++---------
1 files changed, 121 insertions(+), 90 deletions(-)
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c b/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c
index 477053b..dcdd2d9 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c
@@ -355,20 +355,47 @@ is_const(struct tgsi_src_register *src)
(src->File == TGSI_FILE_IMMEDIATE);
}
+static type_t
+get_ftype(struct fd3_compile_context *ctx)
+{
+ return ctx->so->half_precision ? TYPE_F16 : TYPE_F32;
+}
+
+static type_t
+get_utype(struct fd3_compile_context *ctx)
+{
+ return ctx->so->half_precision ? TYPE_U16 : TYPE_U32;
+}
+
+static unsigned
+src_swiz(struct tgsi_src_register *src, int chan)
+{
+ switch (chan) {
+ case 0: return src->SwizzleX;
+ case 1: return src->SwizzleY;
+ case 2: return src->SwizzleZ;
+ case 3: return src->SwizzleW;
+ }
+ assert(0);
+ return 0;
+}
+
/* for instructions that cannot take a const register as src, if needed
* generate a move to temporary gpr:
*/
static struct tgsi_src_register *
get_unconst(struct fd3_compile_context *ctx, struct tgsi_src_register *src)
{
- if (is_const(src)) {
- static struct tgsi_dst_register tmp_dst;
- struct tgsi_src_register *tmp_src =
- get_internal_temp(ctx, &tmp_dst);
- create_mov(ctx, &tmp_dst, src);
- src = tmp_src;
- }
- return src;
+ struct tgsi_dst_register tmp_dst;
+ struct tgsi_src_register *tmp_src;
+
+ compile_assert(ctx, is_const(src));
+
+ tmp_src = get_internal_temp(ctx, &tmp_dst);
+
+ create_mov(ctx, &tmp_dst, src);
+
+ return tmp_src;
}
static void
@@ -418,31 +445,6 @@ get_immediate(struct fd3_compile_context *ctx,
reg->SwizzleW = swiz2tgsi[swiz];
}
-static type_t
-get_ftype(struct fd3_compile_context *ctx)
-{
- return ctx->so->half_precision ? TYPE_F16 : TYPE_F32;
-}
-
-static type_t
-get_utype(struct fd3_compile_context *ctx)
-{
- return ctx->so->half_precision ? TYPE_U16 : TYPE_U32;
-}
-
-static unsigned
-src_swiz(struct tgsi_src_register *src, int chan)
-{
- switch (chan) {
- case 0: return src->SwizzleX;
- case 1: return src->SwizzleY;
- case 2: return src->SwizzleZ;
- case 3: return src->SwizzleW;
- }
- assert(0);
- return 0;
-}
-
static void
create_mov(struct fd3_compile_context *ctx, struct tgsi_dst_register *dst,
struct tgsi_src_register *src)
@@ -463,7 +465,6 @@ create_mov(struct fd3_compile_context *ctx, struct tgsi_dst_register *dst,
ir3_instr_create(ctx->ir, 0, OPC_NOP);
}
}
-
}
static void
@@ -584,6 +585,15 @@ vectorize(struct fd3_compile_context *ctx, struct ir3_instruction *instr,
* native instructions:
*/
+static inline void
+get_swiz(unsigned *swiz, struct tgsi_src_register *src)
+{
+ swiz[0] = src->SwizzleX;
+ swiz[1] = src->SwizzleY;
+ swiz[2] = src->SwizzleZ;
+ swiz[3] = src->SwizzleW;
+}
+
static void
trans_dotp(const struct instr_translater *t,
struct fd3_compile_context *ctx,
@@ -595,34 +605,31 @@ trans_dotp(const struct instr_translater *t,
struct tgsi_dst_register *dst = &inst->Dst[0].Register;
struct tgsi_src_register *src0 = &inst->Src[0].Register;
struct tgsi_src_register *src1 = &inst->Src[1].Register;
- unsigned swiz0[] = { src0->SwizzleX, src0->SwizzleY, src0->SwizzleZ, src0->SwizzleW };
- unsigned swiz1[] = { src1->SwizzleX, src1->SwizzleY, src1->SwizzleZ, src1->SwizzleW };
+ unsigned swiz0[4];
+ unsigned swiz1[4];
opc_t opc_mad = ctx->so->half_precision ? OPC_MAD_F16 : OPC_MAD_F32;
unsigned n = t->arg; /* number of components */
- unsigned i;
+ unsigned i, swapped = 0;
tmp_src = get_internal_temp_repl(ctx, &tmp_dst);
- /* Blob compiler never seems to use a const in src1 position for
- * mad.*, although there does seem (according to disassembler
- * hidden in libllvm-a3xx.so) to be a bit to indicate that src1
- * is a const. Not sure if this is a hw bug, or simply that the
- * disassembler lies.
+ /* in particular, can't handle const for src1 for cat3/mad:
*/
if (is_const(src1)) {
-
- /* the mov to tmp unswizzles src1, so now we have tmp.xyzw:
- */
- for (i = 0; i < 4; i++)
- swiz1[i] = i;
-
- /* the first mul.f will clobber tmp.x, but that is ok
- * because after that point we no longer need tmp.x:
- */
- create_mov(ctx, &tmp_dst, src1);
- src1 = tmp_src;
+ if (!is_const(src0)) {
+ struct tgsi_src_register *tmp;
+ tmp = src0;
+ src0 = src1;
+ src1 = tmp;
+ swapped = 1;
+ } else {
+ src0 = get_unconst(ctx, src0);
+ }
}
+ get_swiz(swiz0, src0);
+ get_swiz(swiz1, src1);
+
instr = ir3_instr_create(ctx->ir, 2, OPC_MUL_F);
add_dst_reg(ctx, instr, &tmp_dst, 0);
add_src_reg(ctx, instr, src0, swiz0[0]);
@@ -640,22 +647,20 @@ trans_dotp(const struct instr_translater *t,
/* DPH(a,b) = (a.x * b.x) + (a.y * b.y) + (a.z * b.z) + b.w */
if (t->tgsi_opc == TGSI_OPCODE_DPH) {
- ir3_instr_create(ctx->ir, 0, OPC_NOP);
+ ir3_instr_create(ctx->ir, 0, OPC_NOP)->repeat = 1;
instr = ir3_instr_create(ctx->ir, 2, OPC_ADD_F);
add_dst_reg(ctx, instr, &tmp_dst, 0);
- add_src_reg(ctx, instr, src1, swiz1[i]);
+ if (swapped)
+ add_src_reg(ctx, instr, src0, swiz0[i]);
+ else
+ add_src_reg(ctx, instr, src1, swiz1[i]);
add_src_reg(ctx, instr, tmp_src, 0);
n++;
}
- ir3_instr_create(ctx->ir, 0, OPC_NOP);
-
- /* pad out to multiple of 4 scalar instructions: */
- for (i = 2 * n; i % 4; i++) {
- ir3_instr_create(ctx->ir, 0, OPC_NOP);
- }
+ ir3_instr_create(ctx->ir, 0, OPC_NOP)->repeat = 2;
create_mov(ctx, dst, tmp_src);
}
@@ -670,6 +675,11 @@ trans_lrp(const struct instr_translater *t,
struct tgsi_dst_register tmp_dst1, tmp_dst2;
struct tgsi_src_register *tmp_src1, *tmp_src2;
struct tgsi_src_register tmp_const;
+ struct tgsi_src_register *src0 = &inst->Src[0].Register;
+ struct tgsi_src_register *src1 = &inst->Src[1].Register;
+
+ if (is_const(src0) && is_const(src1))
+ src0 = get_unconst(ctx, src0);
tmp_src1 = get_internal_temp(ctx, &tmp_dst1);
tmp_src2 = get_internal_temp(ctx, &tmp_dst2);
@@ -678,15 +688,12 @@ trans_lrp(const struct instr_translater *t,
/* tmp1 = (a * b) */
instr = ir3_instr_create(ctx->ir, 2, OPC_MUL_F);
- vectorize(ctx, instr, &tmp_dst1, 2,
- &inst->Src[0].Register, 0,
- &inst->Src[1].Register, 0);
+ vectorize(ctx, instr, &tmp_dst1, 2, src0, 0, src1, 0);
/* tmp2 = (1 - a) */
instr = ir3_instr_create(ctx->ir, 2, OPC_ADD_F);
- vectorize(ctx, instr, &tmp_dst2, 2,
- &tmp_const, 0,
- &inst->Src[0].Register, IR3_REG_NEGATE);
+ vectorize(ctx, instr, &tmp_dst2, 2, &tmp_const, 0,
+ src0, IR3_REG_NEGATE);
/* tmp2 = tmp2 * c */
instr = ir3_instr_create(ctx->ir, 2, OPC_MUL_F);
@@ -930,10 +937,6 @@ trans_cmp(const struct instr_translater *t,
return;
}
- /* NOTE: seems blob compiler will move a const to a gpr if both
- * src args to cmps.f are const. Need to check if this applies
- * to other instructions..
- */
if (is_const(a0) && is_const(a1))
a0 = get_unconst(ctx, a0);
@@ -1041,6 +1044,9 @@ trans_if(const struct instr_translater *t,
get_immediate(ctx, &constval, fui(0.0));
+ if (is_const(src))
+ src = get_unconst(ctx, src);
+
instr = ir3_instr_create(ctx->ir, 2, OPC_CMPS_F);
ir3_reg_create(instr, regid(REG_P0, 0), 0);
add_src_reg(ctx, instr, src, src->SwizzleX);
@@ -1122,11 +1128,11 @@ instr_cat2(const struct instr_translater *t,
struct tgsi_full_instruction *inst)
{
struct tgsi_dst_register *dst = get_dst(ctx, inst);
+ struct tgsi_src_register *src0 = &inst->Src[0].Register;
+ struct tgsi_src_register *src1 = &inst->Src[1].Register;
struct ir3_instruction *instr;
unsigned src0_flags = 0;
- instr = ir3_instr_create(ctx->ir, 2, t->opc);
-
switch (t->tgsi_opc) {
case TGSI_OPCODE_ABS:
src0_flags = IR3_REG_ABS;
@@ -1149,41 +1155,65 @@ instr_cat2(const struct instr_translater *t,
case OPC_SETRM:
case OPC_CBITS_B:
/* these only have one src reg */
- vectorize(ctx, instr, dst, 1,
- &inst->Src[0].Register, src0_flags);
+ instr = ir3_instr_create(ctx->ir, 2, t->opc);
+ vectorize(ctx, instr, dst, 1, src0, src0_flags);
break;
default:
- vectorize(ctx, instr, dst, 2,
- &inst->Src[0].Register, src0_flags,
- &inst->Src[1].Register, 0);
+ if (is_const(src0) && is_const(src1))
+ src0 = get_unconst(ctx, src0);
+
+ instr = ir3_instr_create(ctx->ir, 2, t->opc);
+ vectorize(ctx, instr, dst, 2, src0, src0_flags, src1, 0);
break;
}
put_dst(ctx, inst, dst);
}
+static bool is_mad(opc_t opc)
+{
+ switch (opc) {
+ case OPC_MAD_U16:
+ case OPC_MADSH_U16:
+ case OPC_MAD_S16:
+ case OPC_MADSH_M16:
+ case OPC_MAD_U24:
+ case OPC_MAD_S24:
+ case OPC_MAD_F16:
+ case OPC_MAD_F32:
+ return true;
+ default:
+ return false;
+ }
+}
+
static void
instr_cat3(const struct instr_translater *t,
struct fd3_compile_context *ctx,
struct tgsi_full_instruction *inst)
{
struct tgsi_dst_register *dst = get_dst(ctx, inst);
- struct tgsi_src_register *src1;
+ struct tgsi_src_register *src0 = &inst->Src[0].Register;
+ struct tgsi_src_register *src1 = &inst->Src[1].Register;
struct ir3_instruction *instr;
- /* Blob compiler never seems to use a const in src1 position..
- * although there does seem (according to disassembler hidden
- * in libllvm-a3xx.so) to be a bit to indicate that src1 is a
- * const. Not sure if this is a hw bug, or simply that the
- * disassembler lies.
+ /* in particular, can't handle const for src1 for cat3..
+ * for mad, we can swap first two src's if needed:
*/
- src1 = get_unconst(ctx, &inst->Src[1].Register);
+ if (is_const(src1)) {
+ if (is_mad(t->opc) && !is_const(src0)) {
+ struct tgsi_src_register *tmp;
+ tmp = src0;
+ src0 = src1;
+ src1 = tmp;
+ } else {
+ src0 = get_unconst(ctx, src0);
+ }
+ }
instr = ir3_instr_create(ctx->ir, 3,
ctx->so->half_precision ? t->hopc : t->opc);
- vectorize(ctx, instr, dst, 3,
- &inst->Src[0].Register, 0,
- src1, 0,
+ vectorize(ctx, instr, dst, 3, src0, 0, src1, 0,
&inst->Src[2].Register, 0);
put_dst(ctx, inst, dst);
}
@@ -1194,11 +1224,12 @@ instr_cat4(const struct instr_translater *t,
struct tgsi_full_instruction *inst)
{
struct tgsi_dst_register *dst = get_dst(ctx, inst);
- struct tgsi_src_register *src;
+ struct tgsi_src_register *src = &inst->Src[0].Register;
struct ir3_instruction *instr;
/* seems like blob compiler avoids const as src.. */
- src = get_unconst(ctx, &inst->Src[0].Register);
+ if (is_const(src))
+ src = get_unconst(ctx, src);
ir3_instr_create(ctx->ir, 0, OPC_NOP)->repeat = 5;
instr = ir3_instr_create(ctx->ir, 4, t->opc);
More information about the mesa-commit
mailing list