Mesa (master): r600d: fix some bugs added reworking literal handling
Christian König
deathsimple at kemper.freedesktop.org
Sun Jan 16 02:38:50 UTC 2011
Module: Mesa
Branch: master
Commit: ac6334145ec8eef42505cdd727aed7fae0831e12
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=ac6334145ec8eef42505cdd727aed7fae0831e12
Author: Christian König <deathsimple at vodafone.de>
Date: Sun Jan 16 01:45:06 2011 +0100
r600d: fix some bugs added reworking literal handling
If a literal slot isn't used it should be set
to 0 instead of an uninitialized value. Also the
channels for pre R700 trig functions were incorrect.
And most important literals were not counted against ndw,
resulting in an invalid force_add_cf detection.
---
src/gallium/drivers/r600/r600_asm.c | 51 +++++++++++++++++++------------
src/gallium/drivers/r600/r600_shader.c | 4 +-
2 files changed, 33 insertions(+), 22 deletions(-)
diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index beefd17..a473fb2 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -846,9 +846,9 @@ static int merge_inst_groups(struct r600_bc *bc, struct r600_bc_alu *slots[5],
{
struct r600_bc_alu *prev[5];
struct r600_bc_alu *result[5] = { NULL };
-
- uint32_t literal[4];
- unsigned nliteral = 0;
+
+ uint32_t literal[4], prev_literal[4];
+ unsigned nliteral = 0, prev_nliteral = 0;
int i, j, r, src, num_src;
int num_once_inst = 0;
@@ -861,8 +861,12 @@ static int merge_inst_groups(struct r600_bc *bc, struct r600_bc_alu *slots[5],
struct r600_bc_alu *alu;
/* check number of literals */
- if (prev[i] && r600_bc_alu_nliterals(bc, prev[i], literal, &nliteral))
- return 0;
+ if (prev[i]) {
+ if (r600_bc_alu_nliterals(bc, prev[i], literal, &nliteral))
+ return 0;
+ if (r600_bc_alu_nliterals(bc, prev[i], prev_literal, &prev_nliteral))
+ return 0;
+ }
if (slots[i] && r600_bc_alu_nliterals(bc, slots[i], literal, &nliteral))
return 0;
@@ -923,6 +927,9 @@ static int merge_inst_groups(struct r600_bc *bc, struct r600_bc_alu *slots[5],
/* looks like everything worked out right, apply the changes */
+ /* undo adding previus literals */
+ bc->cf_last->ndw -= align(prev_nliteral, 2);
+
/* sort instructions */
for (i = 0; i < 5; ++i) {
slots[i] = result[i];
@@ -1108,11 +1115,6 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
if (!bc->cf_last->curr_bs_head) {
bc->cf_last->curr_bs_head = nalu;
}
- /* at most 128 slots, one add alu can add 5 slots + 4 constants(2 slots)
- * worst case */
- if (nalu->last && (bc->cf_last->ndw >> 1) >= 120) {
- bc->force_add_cf = 1;
- }
/* number of gpr == the last gpr used in any alu */
for (i = 0; i < 3; i++) {
if (nalu->src[i].sel >= bc->ngpr && nalu->src[i].sel < 128) {
@@ -1133,6 +1135,8 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
/* process cur ALU instructions for bank swizzle */
if (nalu->last) {
+ uint32_t literal[4];
+ unsigned nliteral;
struct r600_bc_alu *slots[5];
r = assign_alu_units(bc, bc->cf_last->curr_bs_head, slots);
if (r)
@@ -1154,6 +1158,21 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
if (r)
return r;
+ for (i = 0, nliteral = 0; i < 5; i++) {
+ if (slots[i]) {
+ r = r600_bc_alu_nliterals(bc, slots[i], literal, &nliteral);
+ if (r)
+ return r;
+ }
+ }
+ bc->cf_last->ndw += align(nliteral, 2);
+
+ /* at most 128 slots, one add alu can add 5 slots + 4 constants(2 slots)
+ * worst case */
+ if ((bc->cf_last->ndw >> 1) >= 120) {
+ bc->force_add_cf = 1;
+ }
+
bc->cf_last->prev2_bs_head = bc->cf_last->prev_bs_head;
bc->cf_last->prev_bs_head = bc->cf_last->curr_bs_head;
bc->cf_last->curr_bs_head = NULL;
@@ -1455,16 +1474,6 @@ int r600_bc_build(struct r600_bc *bc)
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3):
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3):
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
- nliteral = 0;
- LIST_FOR_EACH_ENTRY(alu, &cf->alu, list) {
- r = r600_bc_alu_nliterals(bc, alu, literal, &nliteral);
- if (r)
- return r;
- if (alu->last) {
- cf->ndw += align(nliteral, 2);
- nliteral = 0;
- }
- }
break;
case V_SQ_CF_WORD1_SQ_CF_INST_TEX:
case V_SQ_CF_WORD1_SQ_CF_INST_VTX:
@@ -1514,6 +1523,7 @@ int r600_bc_build(struct r600_bc *bc)
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3):
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
nliteral = 0;
+ memset(literal, 0, sizeof(literal));
LIST_FOR_EACH_ENTRY(alu, &cf->alu, list) {
r = r600_bc_alu_nliterals(bc, alu, literal, &nliteral);
if (r)
@@ -1539,6 +1549,7 @@ int r600_bc_build(struct r600_bc *bc)
bc->bytecode[addr++] = literal[i];
}
nliteral = 0;
+ memset(literal, 0, sizeof(literal));
}
}
break;
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index e85e829..5d952a0 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -1000,7 +1000,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
alu.src[1].chan = 0;
alu.src[1].value = (uint32_t *)&half_inv_pi;
alu.src[2].sel = V_SQ_ALU_SRC_0_5;
- alu.src[2].chan = 1;
+ alu.src[2].chan = 0;
alu.last = 1;
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
@@ -1034,7 +1034,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
alu.src[1].chan = 0;
alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
- alu.src[2].chan = 1;
+ alu.src[2].chan = 0;
if (ctx->bc->chiprev == CHIPREV_R600) {
alu.src[1].value = (uint32_t *)&double_pi;
More information about the mesa-commit
mailing list