r600g: Why all this fiddling with tgsi_helper_copy?

tgsi_helper_copy is used on several occasions to copy a temporary result
into the real destination register to emulate writemasks for OP3 and
reduction operations. According to R600 ISA that's unnecessary.

This patch fixes this use for MAD, CMP and DP4.

diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index d6455023a3af94fbc86c057bdcd80aba3d22b3db..64c8b82c1120a0820af603bbdead2cb8fcdad00c 100644 (file)
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -1587,6 +1587,13 @@ static int tgsi_op3(struct r600_shader_ctx *ctx)
        struct r600_bc_alu_src r600_src[3];
        struct r600_bc_alu alu;
        int i, j, r;
+       int lasti = 0;
+
+       for (i = 0; i < 4; i++) {
+               if (inst->Dst[0].Register.WriteMask & (1 << i)) {
+                       lasti = i;
+               }
+       }
 
        r = tgsi_split_constant(ctx, r600_src);
        if (r)
@@ -1594,26 +1601,32 @@ static int tgsi_op3(struct r600_shader_ctx *ctx)
        r = tgsi_split_literal_constant(ctx, r600_src);
        if (r)
                return r;
-       /* do it in 2 step as op3 doesn't support writemask */
-       for (i = 0; i < 4; i++) {
+       for (i = 0; i < lasti + 1; i++) {
+               if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+                       continue;
+
                memset(&alu, 0, sizeof(struct r600_bc_alu));
                alu.inst = ctx->inst_info->r600_opcode;
                for (j = 0; j < inst->Instruction.NumSrcRegs; j++) {
                        alu.src[j] = r600_src[j];
                        alu.src[j].chan = tgsi_chan(&inst->Src[j], i);
                }
-               alu.dst.sel = ctx->temp_reg;
+
+               r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+               if (r)
+                       return r;
+
                alu.dst.chan = i;
                alu.dst.write = 1;
                alu.is_op3 = 1;
-               if (i == 3) {
+               if (i == lasti) {
                        alu.last = 1;
                }
                r = r600_bc_add_alu(ctx->bc, &alu);
                if (r)
                        return r;
        }
-       return tgsi_helper_copy(ctx, inst);
+       return 0;
 }
 
 static int tgsi_dp(struct r600_shader_ctx *ctx)
@@ -1636,7 +1649,13 @@ static int tgsi_dp(struct r600_shader_ctx *ctx)
                        alu.src[j] = r600_src[j];
                        alu.src[j].chan = tgsi_chan(&inst->Src[j], i);
                }
-               alu.dst.sel = ctx->temp_reg;
+               if(inst->Dst[0].Register.WriteMask & (1 << i)) {
+                       r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+                       if (r)
+                               return r;
+               } else {
+                       alu.dst.sel = ctx->temp_reg;
+               }
                alu.dst.chan = i;
                alu.dst.write = 1;
                /* handle some special cases */
@@ -1670,7 +1689,7 @@ static int tgsi_dp(struct r600_shader_ctx *ctx)
                if (r)
                        return r;
        }
-       return tgsi_helper_copy(ctx, inst);
+       return 0;
 }
 
 static int tgsi_tex(struct r600_shader_ctx *ctx)
@@ -1851,6 +1870,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
        }
 
        if (src_not_temp) {
+                assert(0);
                for (i = 0; i < 4; i++) {
                        memset(&alu, 0, sizeof(struct r600_bc_alu));
                        alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV);
@@ -2000,8 +2020,14 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
        struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
        struct r600_bc_alu_src r600_src[3];
        struct r600_bc_alu alu;
-       int use_temp = 0;
        int i, r;
+       int lasti = 0;
+
+       for (i = 0; i < 4; i++) {
+               if (inst->Dst[0].Register.WriteMask & (1 << i)) {
+                       lasti = i;
+               }
+       }
 
        r = tgsi_split_constant(ctx, r600_src);
        if (r)
@@ -2010,10 +2036,10 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
        if (r)
                return r;
 
-       if (inst->Dst[0].Register.WriteMask != 0xf)
-               use_temp = 1;
+       for (i = 0; i < lasti + 1; i++) {
+               if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+                       continue;
 
-       for (i = 0; i < 4; i++) {
                memset(&alu, 0, sizeof(struct r600_bc_alu));
                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE);
                alu.src[0] = r600_src[0];
@@ -2025,24 +2051,19 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
                alu.src[2] = r600_src[1];
                alu.src[2].chan = tgsi_chan(&inst->Src[1], i);
 
-               if (use_temp)
-                       alu.dst.sel = ctx->temp_reg;
-               else {
-                       r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
-                       if (r)
-                               return r;
-               }
+               r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+               if (r)
+                       return r;
+
                alu.dst.chan = i;
                alu.dst.write = 1;
                alu.is_op3 = 1;
-               if (i == 3)
+               if (i == lasti)
                        alu.last = 1;
                r = r600_bc_add_alu(ctx->bc, &alu);
                if (r)
                        return r;
        }
-       if (use_temp)
-               return tgsi_helper_copy(ctx, inst);
        return 0;
 }