Mesa (master): radv: add support for anisotropic filtering on SI-CI

[Dave Airlie]

Module: Mesa
Branch: master
Commit: 5cbcbc75f437b805e5c3c0c3b5c1192c0d68c1f5
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=5cbcbc75f437b805e5c3c0c3b5c1192c0d68c1f5

Author: Fredrik Höglund <fredrik at kde.org>
Date:   Wed Nov 23 23:04:58 2016 +0100

radv: add support for anisotropic filtering on SI-CI

Ported from radeonsi.

Note that si_make_texture_descriptor() already sets img7 to the mask
value referred to in the comment.

Reviewed-by: Dave Airlie <airlied at redhat.com>

---

 src/amd/common/ac_nir_to_llvm.c | 31 +++++++++++++++++++++++++++++++
 src/amd/vulkan/radv_device.c    | 11 ++---------
 2 files changed, 33 insertions(+), 9 deletions(-)

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index c68cb9c..06645be 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -3151,6 +3151,35 @@ static void set_tex_fetch_args(struct nir_to_llvm_context *ctx,
 	tinfo->arg_count = num_args;
 }
 
+/* Disable anisotropic filtering if BASE_LEVEL == LAST_LEVEL.
+ *
+ * SI-CI:
+ *   If BASE_LEVEL == LAST_LEVEL, the shader must disable anisotropic
+ *   filtering manually. The driver sets img7 to a mask clearing
+ *   MAX_ANISO_RATIO if BASE_LEVEL == LAST_LEVEL. The shader must do:
+ *     s_and_b32 samp0, samp0, img7
+ *
+ * VI:
+ *   The ANISO_OVERRIDE sampler field enables this fix in TA.
+ */
+static LLVMValueRef sici_fix_sampler_aniso(struct nir_to_llvm_context *ctx,
+                                           LLVMValueRef res, LLVMValueRef samp)
+{
+	LLVMBuilderRef builder = ctx->builder;
+	LLVMValueRef img7, samp0;
+
+	if (ctx->options->chip_class >= VI)
+		return samp;
+
+	img7 = LLVMBuildExtractElement(builder, res,
+	                               LLVMConstInt(ctx->i32, 7, 0), "");
+	samp0 = LLVMBuildExtractElement(builder, samp,
+	                                LLVMConstInt(ctx->i32, 0, 0), "");
+	samp0 = LLVMBuildAnd(builder, samp0, img7, "");
+	return LLVMBuildInsertElement(builder, samp, samp0,
+	                              LLVMConstInt(ctx->i32, 0, 0), "");
+}
+
 static void tex_fetch_ptrs(struct nir_to_llvm_context *ctx,
 			   nir_tex_instr *instr,
 			   LLVMValueRef *res_ptr, LLVMValueRef *samp_ptr,
@@ -3165,6 +3194,8 @@ static void tex_fetch_ptrs(struct nir_to_llvm_context *ctx,
 			*samp_ptr = get_sampler_desc(ctx, instr->sampler, DESC_SAMPLER);
 		else
 			*samp_ptr = get_sampler_desc(ctx, instr->texture, DESC_SAMPLER);
+		if (instr->sampler_dim < GLSL_SAMPLER_DIM_RECT)
+			*samp_ptr = sici_fix_sampler_aniso(ctx, *res_ptr, *samp_ptr);
 	}
 	if (fmask_ptr && !instr->sampler && (instr->op == nir_texop_txf_ms ||
 					     instr->op == nir_texop_samples_identical))
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 00f3a07..70c94e9 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -371,7 +371,7 @@ void radv_GetPhysicalDeviceFeatures(
 		.largePoints                              = true,
 		.alphaToOne                               = true,
 		.multiViewport                            = false,
-		.samplerAnisotropy                        = false, /* FINISHME */
+		.samplerAnisotropy                        = true,
 		.textureCompressionETC2                   = false,
 		.textureCompressionASTC_LDR               = false,
 		.textureCompressionBC                     = true,
@@ -1805,14 +1805,7 @@ radv_init_sampler(struct radv_device *device,
 	uint32_t max_aniso = pCreateInfo->anisotropyEnable && pCreateInfo->maxAnisotropy > 1.0 ?
 					(uint32_t) pCreateInfo->maxAnisotropy : 0;
 	uint32_t max_aniso_ratio = radv_tex_aniso_filter(max_aniso);
-	bool is_vi;
-	is_vi = (device->instance->physicalDevice.rad_info.chip_class >= VI);
-
-	if (!is_vi && max_aniso > 0) {
-		radv_finishme("Anisotropic filtering must be disabled manually "
-		              "by the shader on SI-CI when BASE_LEVEL == LAST_LEVEL\n");
-		max_aniso = max_aniso_ratio = 0;
-	}
+	bool is_vi = (device->instance->physicalDevice.rad_info.chip_class >= VI);
 
 	sampler->state[0] = (S_008F30_CLAMP_X(radv_tex_wrap(pCreateInfo->addressModeU)) |
 			     S_008F30_CLAMP_Y(radv_tex_wrap(pCreateInfo->addressModeV)) |