mesa/mesa-8.0-nouveau-vieux-nvfx-lowmem.patch

From a8c4c4dde253db0500c3b6115edf87d2caf1667f Mon Sep 17 00:00:00 2001
From: Ben Skeggs <bskeggs@redhat.com>
Date: Fri, 11 May 2012 01:31:36 +1000
Subject: [PATCH] vieux/nvfx lowmem hacks

---
 src/gallium/drivers/nvfx/nvfx_fragprog.c       |   23 ++++++++++----------
 src/gallium/drivers/nvfx/nvfx_screen.h         |    2 ++
 src/gallium/drivers/nvfx/nvfx_surface.c        |    9 +++++++-
 src/gallium/drivers/nvfx/nvfx_vbo.c            |    5 +++++
 src/gallium/drivers/nvfx/nvfx_vertprog.c       |    4 +++-
 src/mesa/drivers/dri/nouveau/nouveau_context.c |    5 +++++
 src/mesa/drivers/dri/nouveau/nouveau_context.h |    4 ++++
 src/mesa/drivers/dri/nouveau/nouveau_driver.c  |    3 +++
 src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c   |    3 +++
 src/mesa/drivers/dri/nouveau/nv04_surface.c    |   27 +++++++++++++++++++++---
 10 files changed, 68 insertions(+), 17 deletions(-)

diff --git a/src/gallium/drivers/nvfx/nvfx_fragprog.c b/src/gallium/drivers/nvfx/nvfx_fragprog.c
index 0babcbb..f5b7a0e 100644
--- a/src/gallium/drivers/nvfx/nvfx_fragprog.c
+++ b/src/gallium/drivers/nvfx/nvfx_fragprog.c
@@ -224,7 +224,6 @@ nvfx_fp_emit(struct nvfx_fpc *fpc, struct nvfx_insn insn)
 	if(insn.unit >= 0)
 	{
 		hw[0] |= (insn.unit << NVFX_FP_OP_TEX_UNIT_SHIFT);
-		fp->samplers |= (1 << insn.unit);
 	}
 
 	emit_dst(fpc, insn.dst);
@@ -1355,17 +1354,17 @@ update_slots:
 
 			fp->slot_to_fp_input[fp->num_slots] = sprite_real_input;
 
-			if(nvfx->is_nv4x)
-			{
-				fp->or = 0;
-				for(i = 0; i <= fp->num_slots; ++i) {
-					unsigned fp_input = fp->slot_to_fp_input[i];
-					if(fp_input == NVFX_FP_OP_INPUT_SRC_TC(8))
-						fp->or |= (1 << 12);
-					else if(fp_input == NVFX_FP_OP_INPUT_SRC_TC(9))
-						fp->or |= (1 << 13);
-					else if(fp_input >= NVFX_FP_OP_INPUT_SRC_TC(0) && fp_input <= NVFX_FP_OP_INPUT_SRC_TC(7))
-						fp->or |= (1 << (fp_input - NVFX_FP_OP_INPUT_SRC_TC0 + 14));
+			fp->or = 0;
+			fp->samplers = 0;
+			for(i = 0; i <= fp->num_slots; ++i) {
+				unsigned fp_input = fp->slot_to_fp_input[i];
+				if(fp_input == NVFX_FP_OP_INPUT_SRC_TC(8))
+					fp->or |= (1 << 12);
+				else if(fp_input == NVFX_FP_OP_INPUT_SRC_TC(9))
+					fp->or |= (1 << 13);
+				else if(fp_input >= NVFX_FP_OP_INPUT_SRC_TC(0) && fp_input <= NVFX_FP_OP_INPUT_SRC_TC(7)) {
+					fp->or |= (1 << (fp_input - NVFX_FP_OP_INPUT_SRC_TC0 + 14));
+					fp->samplers |= 1 << (fp_input - NVFX_FP_OP_INPUT_SRC_TC(0));
 				}
 			}
 
diff --git a/src/gallium/drivers/nvfx/nvfx_screen.h b/src/gallium/drivers/nvfx/nvfx_screen.h
index 35c9d91..55ed439 100644
--- a/src/gallium/drivers/nvfx/nvfx_screen.h
+++ b/src/gallium/drivers/nvfx/nvfx_screen.h
@@ -14,6 +14,8 @@ struct nvfx_screen {
 
 	struct nvfx_context *cur_ctx;
 
+	boolean hack;
+
 	unsigned is_nv4x; /* either 0 or ~0 */
 	unsigned use_nv4x; /* either 0 or ~0 */
 	boolean force_swtnl;
diff --git a/src/gallium/drivers/nvfx/nvfx_surface.c b/src/gallium/drivers/nvfx/nvfx_surface.c
index cac3fca..3937869 100644
--- a/src/gallium/drivers/nvfx/nvfx_surface.c
+++ b/src/gallium/drivers/nvfx/nvfx_surface.c
@@ -273,8 +273,12 @@ nvfx_resource_copy_region(struct pipe_context *pipe,
 	small = (w * h <= copy_threshold);
 	if((!dst_to_gpu || !src_on_gpu) && small)
 		ret = -1; /* use the CPU */
-	else
+	else {
 		ret = nv04_region_copy_2d(ctx, &dst, &src, w, h, dst_to_gpu, src_on_gpu);
+		if (nvfx_screen(pipe->screen)->hack)
+			FIRE_RING(nvfx_screen(pipe->screen)->base.channel);
+	}
+
 	if(!ret)
 	{}
 	else if(ret > 0
@@ -335,6 +339,9 @@ nvfx_surface_fill(struct pipe_context* pipe, struct pipe_surface *dsts,
 	h = util_format_get_nblocksy(dsts->format, h);
 
 	ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
+	if (nvfx_screen(pipe->screen)->hack)
+		FIRE_RING(nvfx_screen(pipe->screen)->base.channel);
+
 	if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 		return 1;
 	else if(ret)
diff --git a/src/gallium/drivers/nvfx/nvfx_vbo.c b/src/gallium/drivers/nvfx/nvfx_vbo.c
index b72379d..5860b54 100644
--- a/src/gallium/drivers/nvfx/nvfx_vbo.c
+++ b/src/gallium/drivers/nvfx/nvfx_vbo.c
@@ -237,6 +237,11 @@ void nvfx_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 		}
 	}
 
+	if (nvfx->screen->hack) {
+		struct nouveau_channel *chan = nvfx->screen->base.channel;
+		FIRE_RING(chan);
+	}
+
 	if (nvfx->screen->force_swtnl || !nvfx_state_validate(nvfx))
 		nvfx_draw_vbo_swtnl(pipe, info);
 	else
diff --git a/src/gallium/drivers/nvfx/nvfx_vertprog.c b/src/gallium/drivers/nvfx/nvfx_vertprog.c
index a11941f..e19a288 100644
--- a/src/gallium/drivers/nvfx/nvfx_vertprog.c
+++ b/src/gallium/drivers/nvfx/nvfx_vertprog.c
@@ -288,6 +288,8 @@ nvfx_vp_emit(struct nvfx_vpc *vpc, struct nvfx_insn insn)
 
 	hw = vpc->vpi->data;
 
+	if (insn.cc_test != NVFX_COND_TR)
+		hw[0] |= NVFX_VP(INST_COND_TEST_ENABLE);
 	hw[0] |= (insn.cc_test << NVFX_VP(INST_COND_SHIFT));
 	hw[0] |= ((insn.cc_swz[0] << NVFX_VP(INST_COND_SWZ_X_SHIFT)) |
 		  (insn.cc_swz[1] << NVFX_VP(INST_COND_SWZ_Y_SHIFT)) |
@@ -649,7 +651,7 @@ nvfx_vertprog_parse_instruction(struct nvfx_context* nvfx, struct nvfx_vpc *vpc,
 		nvfx_vp_emit(vpc, arith(sat, VEC, SNE, dst, mask, src[0], src[1], none));
 		break;
 	case TGSI_OPCODE_SSG:
-		nvfx_vp_emit(vpc, arith(sat, VEC, SSG, dst, mask, src[0], src[1], none));
+		nvfx_vp_emit(vpc, arith(sat, VEC, SSG, dst, mask, src[0], none, none));
 		break;
 	case TGSI_OPCODE_STR:
 		nvfx_vp_emit(vpc, arith(sat, VEC, STR, dst, mask, src[0], src[1], none));
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_context.c b/src/mesa/drivers/dri/nouveau/nouveau_context.c
index 0cb918e..9051e18 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_context.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_context.c
@@ -106,6 +106,11 @@ nouveau_context_init(struct gl_context *ctx, struct nouveau_screen *screen,
 	nctx->screen = screen;
 	nctx->fallback = HWTNL;
 
+	if (getenv("NOUVEAU_UNHACK"))
+		nctx->hack = GL_FALSE;
+	else
+		nctx->hack = GL_TRUE;
+
 	/* Initialize the function pointers. */
 	_mesa_init_driver_functions(&functions);
 	nouveau_driver_functions_init(&functions);
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_context.h b/src/mesa/drivers/dri/nouveau/nouveau_context.h
index cd4a9fb..7777246 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_context.h
+++ b/src/mesa/drivers/dri/nouveau/nouveau_context.h
@@ -74,6 +74,8 @@ struct nouveau_context {
 		GLboolean clear_blocked;
 		int clear_seq;
 	} hierz;
+
+	GLboolean hack;
 };
 
 #define to_nouveau_context(ctx)	((struct nouveau_context *)(ctx))
@@ -88,6 +90,8 @@ struct nouveau_context {
 	(to_nouveau_context(ctx)->hw.eng3d)
 #define context_drv(ctx) \
 	(to_nouveau_context(ctx)->screen->driver)
+#define context_hack(ctx) \
+	(to_nouveau_context(ctx)->hack)
 #define context_dirty(ctx, s) \
 	BITSET_SET(to_nouveau_context(ctx)->dirty, NOUVEAU_STATE_##s)
 #define context_dirty_i(ctx, s, i) \
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_driver.c b/src/mesa/drivers/dri/nouveau/nouveau_driver.c
index ba818f0..8532032 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_driver.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_driver.c
@@ -128,6 +128,9 @@ nouveau_clear(struct gl_context *ctx, GLbitfield buffers)
 
 	if (buffers)
 		_mesa_meta_Clear(ctx, buffers);
+
+	if (context_hack(ctx))
+		FIRE_RING(context_chan(ctx));
 }
 
 void
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
index de04d18..2337a73 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
@@ -456,6 +456,9 @@ TAG(vbo_render_prims)(struct gl_context *ctx,
 			     max_index);
 
 	vbo_deinit_arrays(ctx, ib, arrays);
+
+	if (context_hack(ctx))
+		FIRE_RING(context_chan(ctx));
 }
 
 /* VBO rendering entry points. */
diff --git a/src/mesa/drivers/dri/nouveau/nv04_surface.c b/src/mesa/drivers/dri/nouveau/nv04_surface.c
index 4a3000c..2383711 100644
--- a/src/mesa/drivers/dri/nouveau/nv04_surface.c
+++ b/src/mesa/drivers/dri/nouveau/nv04_surface.c
@@ -209,6 +209,13 @@ nv04_surface_copy_swizzle(struct gl_context *ctx,
 	unsigned sub_h = h > max_h ? max_h : h;
 	unsigned x, y;
 
+	if (context_hack(ctx)) {
+		void *flush_notify = chan->flush_notify;
+		chan->flush_notify = NULL;
+		FIRE_RING(chan);
+		chan->flush_notify = flush_notify;
+	}
+
         /* Swizzled surfaces must be POT  */
 	assert(_mesa_is_pow_two(dst->width) &&
 	       _mesa_is_pow_two(dst->height));
@@ -263,7 +270,7 @@ nv04_surface_copy_swizzle(struct gl_context *ctx,
 
 	nouveau_bo_context_reset(bctx);
 
-	if (context_chipset(ctx) < 0x10)
+	if (context_hack(ctx) || context_chipset(ctx) < 0x10)
 		FIRE_RING(chan);
 }
 
@@ -282,6 +289,13 @@ nv04_surface_copy_m2mf(struct gl_context *ctx,
 	unsigned dst_offset = dst->offset + dy * dst->pitch + dx * dst->cpp;
 	unsigned src_offset = src->offset + sy * src->pitch + sx * src->cpp;
 
+	if (context_hack(ctx)) {
+		void *flush_notify = chan->flush_notify;
+		chan->flush_notify = NULL;
+		FIRE_RING(chan);
+		chan->flush_notify = flush_notify;
+	}
+
 	nouveau_bo_marko(bctx, m2mf, NV04_M2MF_DMA_BUFFER_IN,
 			 src->bo, bo_flags | NOUVEAU_BO_RD);
 	nouveau_bo_marko(bctx, m2mf, NV04_M2MF_DMA_BUFFER_OUT,
@@ -311,7 +325,7 @@ nv04_surface_copy_m2mf(struct gl_context *ctx,
 
 	nouveau_bo_context_reset(bctx);
 
-	if (context_chipset(ctx) < 0x10)
+	if (context_hack(ctx) || context_chipset(ctx) < 0x10)
 		FIRE_RING(chan);
 }
 
@@ -429,6 +443,13 @@ nv04_surface_fill(struct gl_context *ctx,
 	struct nouveau_grobj *rect = hw->rect;
 	unsigned bo_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART;
 
+	if (context_hack(ctx)) {
+		void *flush_notify = chan->flush_notify;
+		chan->flush_notify = NULL;
+		FIRE_RING(chan);
+		chan->flush_notify = flush_notify;
+	}
+
 	MARK_RING (chan, 19, 4);
 
 	BEGIN_RING(chan, surf2d, NV04_CONTEXT_SURFACES_2D_DMA_IMAGE_SOURCE, 2);
@@ -454,7 +475,7 @@ nv04_surface_fill(struct gl_context *ctx,
 	OUT_RING  (chan, (dx << 16) | dy);
 	OUT_RING  (chan, ( w << 16) |  h);
 
-	if (context_chipset(ctx) < 0x10)
+	if (context_hack(ctx) || context_chipset(ctx) < 0x10)
 		FIRE_RING(chan);
 }
 
-- 
1.7.10