Patch merge tgsi_r600_arl and tgsi_eg_arl, use tgsi_r600_arl as base
and rename function to tgsi_arl. Fixes the following piglit tests on
Evergreen in the process:
glean
vp1-ARL test
shaders
glsl-fs-varying-array
glsl-fs-vec4-indexing-temp-src
glsl-vs-arrays-2
glsl-vs-mov-after-deref
glsl-vs-texturematrix-2
glsl-vs-vec4-indexing-temp-src
Fixes GPU lockups on Evergreen with:
shaders
glsl-fs-vec4-indexing-temp-dst
glsl-vs-vec4-indexing-temp-dst
glsl-vs-varying-array
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 609d629..ee0aff2 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -2315,51 +2315,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
return tgsi_helper_copy(ctx, inst);
}
-static int tgsi_eg_arl(struct r600_shader_ctx *ctx)
-{
- struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
- struct r600_bc_alu alu;
- int r;
-
- memset(&alu, 0, sizeof(struct r600_bc_alu));
-
- switch (inst->Instruction.Opcode) {
- case TGSI_OPCODE_ARL:
- alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT_FLOOR;
- break;
- case TGSI_OPCODE_ARR:
- alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT;
- break;
- default:
- assert(0);
- return -1;
- }
-
- r600_bc_src(&alu.src[0], &ctx->src[0], 0);
- alu.last = 1;
- alu.dst.sel = ctx->ar_reg;
- alu.dst.write = 1;
- r = r600_bc_add_alu(ctx->bc, &alu);
- if (r)
- return r;
-
- /* TODO: Note that the MOVA can be avoided if we never use AR for
- * indexing non-CB registers in the current ALU clause. Similarly, we
- * need to load AR from ar_reg again if we started a new clause
- * between ARL and AR usage. The easy way to do that is to remove
- * the MOVA here, and load it for the first AR access after ar_reg
- * has been modified in each clause. */
- memset(&alu, 0, sizeof(struct r600_bc_alu));
- alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT;
- alu.src[0].sel = ctx->ar_reg;
- alu.src[0].chan = 0;
- alu.last = 1;
- r = r600_bc_add_alu(ctx->bc, &alu);
- if (r)
- return r;
- return 0;
-}
-static int tgsi_r600_arl(struct r600_shader_ctx *ctx)
+static int tgsi_arl(struct r600_shader_ctx *ctx)
{
/* TODO from r600c, ar values don't persist between clauses */
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
@@ -2369,7 +2325,7 @@ static int tgsi_r600_arl(struct r600_shader_ctx *ctx)
switch (inst->Instruction.Opcode) {
case TGSI_OPCODE_ARL:
memset(&alu, 0, sizeof(alu));
- alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR;
+ alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR);
r600_bc_src(&alu.src[0], &ctx->src[0], 0);
alu.dst.sel = ctx->ar_reg;
alu.dst.write = 1;
@@ -2379,7 +2335,7 @@ static int tgsi_r600_arl(struct r600_shader_ctx *ctx)
return r;
memset(&alu, 0, sizeof(alu));
- alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT;
+ alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT);
alu.src[0].sel = ctx->ar_reg;
alu.dst.sel = ctx->ar_reg;
alu.dst.write = 1;
@@ -2390,7 +2346,7 @@ static int tgsi_r600_arl(struct r600_shader_ctx *ctx)
break;
case TGSI_OPCODE_ARR:
memset(&alu, 0, sizeof(alu));
- alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT;
+ alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT);
r600_bc_src(&alu.src[0], &ctx->src[0], 0);
alu.dst.sel = ctx->ar_reg;
alu.dst.write = 1;
@@ -2403,9 +2359,15 @@ static int tgsi_r600_arl(struct r600_shader_ctx *ctx)
assert(0);
return -1;
}
-
+
+ /* TODO: Note that the MOVA can be avoided if we never use AR for
+ * indexing non-CB registers in the current ALU clause. Similarly, we
+ * need to load AR from ar_reg again if we started a new clause
+ * between ARL and AR usage. The easy way to do that is to remove
+ * the MOVA here, and load it for the first AR access after ar_reg
+ * has been modified in each clause. */
memset(&alu, 0, sizeof(alu));
- alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT;
+ alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT);
alu.src[0].sel = ctx->ar_reg;
alu.last = 1;
@@ -2742,7 +2704,7 @@ static int tgsi_loop_brk_cont(struct r600_shader_ctx *ctx)
}
static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = {
- {TGSI_OPCODE_ARL, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_r600_arl},
+ {TGSI_OPCODE_ARL, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_arl},
{TGSI_OPCODE_MOV, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV, tgsi_op2},
{TGSI_OPCODE_LIT, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_lit},
@@ -2812,7 +2774,7 @@ static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = {
{TGSI_OPCODE_UP4UB, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_X2D, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_ARA, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
- {TGSI_OPCODE_ARR, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_r600_arl},
+ {TGSI_OPCODE_ARR, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_arl},
{TGSI_OPCODE_BRA, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_CAL, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_RET, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
@@ -2906,7 +2868,7 @@ static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = {
};
static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] = {
- {TGSI_OPCODE_ARL, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_eg_arl},
+ {TGSI_OPCODE_ARL, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_arl},
{TGSI_OPCODE_MOV, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV, tgsi_op2},
{TGSI_OPCODE_LIT, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_lit},
{TGSI_OPCODE_RCP, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE, tgsi_trans_srcx_replicate},
@@ -2970,7 +2932,7 @@ static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] = {
{TGSI_OPCODE_UP4UB, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_X2D, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_ARA, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
- {TGSI_OPCODE_ARR, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_eg_arl},
+ {TGSI_OPCODE_ARR, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_arl},
{TGSI_OPCODE_BRA, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_CAL, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
{TGSI_OPCODE_RET, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
_______________________________________________
mesa-dev mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
