r300: Simplify DCE by assuming all output writes are used.

No change on shader-db.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14847>

3 files changed, 8 insertions, 35 deletions

diff --git a/src/gallium/drivers/r300/compiler/r3xx_fragprog.c b/src/gallium/drivers/r300/compiler/r3xx_fragprog.c
index 8aa4ef9e86b..1d41a552dd2 100644
--- a/src/gallium/drivers/r300/compiler/r3xx_fragprog.c
+++ b/src/gallium/drivers/r300/compiler/r3xx_fragprog.c
@@ -37,17 +37,6 @@
 #include "r500_fragprog.h"
 
 
-static void dataflow_outputs_mark_use(void * userdata, void * data,
-		void (*callback)(void *, unsigned int, unsigned int))
-{
-	struct r300_fragment_program_compiler * c = userdata;
-	callback(data, c->OutputColor[0], RC_MASK_XYZW);
-	callback(data, c->OutputColor[1], RC_MASK_XYZW);
-	callback(data, c->OutputColor[2], RC_MASK_XYZW);
-	callback(data, c->OutputColor[3], RC_MASK_XYZW);
-	callback(data, c->OutputDepth, RC_MASK_W);
-}
-
 static void rc_rewrite_depth_out(struct radeon_compiler *cc, void *user)
 {
 	struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)cc;
@@ -128,7 +117,7 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
 		{"transform IF",		1, is_r500,	rc_local_transform,		rewrite_if},
 		{"native rewrite",		1, is_r500,	rc_local_transform,		native_rewrite_r500},
 		{"native rewrite",		1, !is_r500,	rc_local_transform,		native_rewrite_r300},
-		{"deadcode",			1, opt,		rc_dataflow_deadcode,		dataflow_outputs_mark_use},
+		{"deadcode",			1, opt,		rc_dataflow_deadcode,		NULL},
 		{"emulate loops",		1, !is_r500,	rc_emulate_loops,		NULL},
 		{"register rename",		1, !is_r500 || opt,		rc_rename_regs,			NULL},
 		{"dataflow optimize",		1, opt,		rc_optimize,			NULL},
diff --git a/src/gallium/drivers/r300/compiler/r3xx_vertprog.c b/src/gallium/drivers/r300/compiler/r3xx_vertprog.c
index de0d98240f3..559a8ff54e7 100644
--- a/src/gallium/drivers/r300/compiler/r3xx_vertprog.c
+++ b/src/gallium/drivers/r300/compiler/r3xx_vertprog.c
@@ -772,18 +772,6 @@ static void rc_vs_add_artificial_outputs(struct radeon_compiler *c, void *user)
 	}
 }
 
-static void dataflow_outputs_mark_used(void * userdata, void * data,
-		void (*callback)(void *, unsigned int, unsigned int))
-{
-	struct r300_vertex_program_compiler * c = userdata;
-	int i;
-
-	for(i = 0; i < 32; ++i) {
-		if (c->RequiredOutputs & (1U << i))
-			callback(data, i, RC_MASK_XYZW);
-	}
-}
-
 static int swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
 {
 	(void) opcode;
@@ -909,7 +897,7 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler *c)
 		{"native rewrite",		1, is_r500,	rc_local_transform,		alu_rewrite_r500},
 		{"native rewrite",		1, !is_r500,	rc_local_transform,		alu_rewrite_r300},
 		{"emulate modifiers",		1, !is_r500,	rc_local_transform,		emulate_modifiers},
-		{"deadcode",			1, opt,		rc_dataflow_deadcode,		dataflow_outputs_mark_used},
+		{"deadcode",			1, opt,		rc_dataflow_deadcode,		NULL},
 		{"dataflow optimize",		1, opt,		rc_optimize,			NULL},
 		/* This pass must be done after optimizations. */
 		{"source conflict resolve",	1, 1,		rc_local_transform,		resolve_src_conflicts},
diff --git a/src/gallium/drivers/r300/compiler/radeon_dataflow_deadcode.c b/src/gallium/drivers/r300/compiler/radeon_dataflow_deadcode.c
index 44fc13377fa..22b6d1ba58d 100644
--- a/src/gallium/drivers/r300/compiler/radeon_dataflow_deadcode.c
+++ b/src/gallium/drivers/r300/compiler/radeon_dataflow_deadcode.c
@@ -204,18 +204,10 @@ static void update_instruction(struct deadcode_state * s, struct rc_instruction
 	}
 }
 
-static void mark_output_use(void * data, unsigned int index, unsigned int mask)
-{
-	struct deadcode_state * s = data;
-
-	mark_used(s, RC_FILE_OUTPUT, index, mask);
-}
-
 void rc_dataflow_deadcode(struct radeon_compiler * c, void *user)
 {
 	struct deadcode_state s;
 	unsigned int nr_instructions;
-	rc_dataflow_mark_outputs_fn dce = (rc_dataflow_mark_outputs_fn)user;
 	unsigned int ip;
 
 	memset(&s, 0, sizeof(s));
@@ -225,13 +217,17 @@ void rc_dataflow_deadcode(struct radeon_compiler * c, void *user)
 	s.Instructions = memory_pool_malloc(&c->Pool, sizeof(struct instruction_state)*nr_instructions);
 	memset(s.Instructions, 0, sizeof(struct instruction_state)*nr_instructions);
 
-	dce(c, &s, &mark_output_use);
-
 	for(struct rc_instruction * inst = c->Program.Instructions.Prev;
 	    inst != &c->Program.Instructions;
 	    inst = inst->Prev) {
 		const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->U.I.Opcode);
 
+		/* Assume all output regs are live.  Anything else should have been
+		 * eliminated before it got to us.
+		 */
+		if (opcode->HasDstReg)
+			mark_used(&s, RC_FILE_OUTPUT, inst->U.I.DstReg.Index, inst->U.I.DstReg.WriteMask);
+
 		switch(opcode->Opcode){
 		/* Mark all sources in the loop body as used before doing
 		 * normal deadcode analysis.  This is probably not optimal.